40 Commits

Author SHA1 Message Date
admin a8eaa4c367 255漂移问题稳定时间改30秒 2026-09-18 13:37:03 +08:00
admin 32b6af62e2 完善行程标定与视觉运动判定 2026-09-17 18:25:36 +08:00
admin af2dc9c38f 新增 O30 右手行程标定功能包并完善相机恢复
支持三机位 Tag 两端边界标定、滑块示教、顺序避让恢复、同步四指扫描和 JSON 离线复算;修复相机时钟异常导致采集线程退出的问题,并忽略本地行程标定输出。
2026-09-17 12:41:53 +08:00
admin 7a04780b52 重构O6验证过 2026-09-16 14:05:08 +08:00
admin 467651fc47 末节采用原 CAD 零位 2026-09-11 17:45:44 +08:00
admin 69c2da6808 标定改造方案:按关节锁定零位,输出实测 JSON 与近似 URDF 2026-09-11 14:43:44 +08:00
admin f94cf2c500 重构 2026-09-11 10:07:45 +08:00
admin c4ad2b968a O12重构一版提交 2026-09-09 15:05:09 +08:00
admin a8ddcc6296 O6gui预设动作修改、O12标定初始化提交 2026-09-04 18:35:04 +08:00
admin 2356bd6247 O12 gui 2026-09-04 15:35:45 +08:00
admin 889e0ea8db o12右手原始urdf 2026-09-04 10:48:19 +08:00
admin e1fb458eff 新增o6/l6左手原始urdf 2026-09-03 17:39:37 +08:00
admin 8de69c34a1 o6右手标定 2026-09-03 10:11:13 +08:00
admin d6b7bd6209 通用 URDF patch engine 抽取 2026-09-02 15:17:50 +08:00
admin 8a749a3687 根据json修正urdf 2026-09-02 14:20:29 +08:00
admin 08fe190b3a O6原始urdf 2026-09-02 13:36:16 +08:00
admin f7aeef87a8 L6右手标定 2026-09-02 13:26:32 +08:00
admin 2b7c1f92e7 原始urdf位置修改 2026-09-01 13:57:37 +08:00
admin 1ed36ecdd8 标定代码结构修改 2026-09-01 11:51:28 +08:00
admin 7f84225ba8 refactor: dispatch formal calibration through model profiles 2026-08-31 19:22:51 +08:00
admin ba9f1b25e8 refactor: establish reusable calibration architecture 2026-08-31 18:33:31 +08:00
admin 0d606c2ba2 refactor: rename calibration package 2026-08-31 18:06:15 +08:00
admin 06c050e446 test: restore calibration regression baseline 2026-08-31 17:58:11 +08:00
admin 286581bcba 大拇指单独标定,yaw正确和稳定性修改 2026-08-31 17:17:39 +08:00
admin 4dadfb954b 大拇指零位正确性稳定性修改 2026-08-30 17:35:23 +08:00
admin 83c69b48c2 标定稳定性 2026-08-27 16:24:01 +08:00
admin 4e594ddb09 G20右手四指独立标定(少thumb_mcp) 2026-08-24 10:12:23 +08:00
admin ef65681230 G20四指单独标定(少末端tag) 2026-08-21 12:21:39 +08:00
admin a609d521a0 g20右手标定 2026-08-11 15:51:36 +08:00
admin 41ff4a61a9 新零位相机外参标定方案 2026-08-07 16:22:24 +08:00
admin 4107da4c22 三机位标定 2026-08-05 18:43:10 +08:00
admin 5d206bcb73 三机位标定 2026-08-05 18:32:51 +08:00
admin 05634f5472 拇指侧摆零位标定 2026-08-03 15:36:49 +08:00
admin ce9d0129b9 拇指零位标定 2026-07-30 17:05:22 +08:00
admin 9210373fb2 圆心标定方案 2026-07-29 18:30:08 +08:00
admin 44975620a7 标定 2026-07-29 11:49:16 +08:00
admin 0d92e5f998 大拇指安全限制去除 2026-07-27 17:40:57 +08:00
admin fc7c66d30e 大拇指横摆安全限制去除 2026-07-27 17:28:06 +08:00
admin b7cf448a4d 多手势标定 2026-07-24 17:36:23 +08:00
admin 1bec806c6e chore: save current code 2026-07-23 11:10:00 +08:00
731 changed files with 118327 additions and 2193 deletions
+50 -1
View File
@@ -50,12 +50,61 @@ Thumbs.db
# Runtime and calibration scratch files
/logs/
# The camera SDK writes logs relative to the launch working directory.
MvSdkLog/
*.tmp
*.log
*.bak
*.orig
*.rej
# Operator/device-specific calibration artifacts
# Reproducible seed profiles remain under
# src/linkerhand_retarget/resource/linkerforce_v2/profiles/.
/profiles/
# Calibration outputs may be created from the workspace root or a subdirectory.
calibration_output/
range_calibration_output/
/config/*_three_camera_extrinsics.yaml
# Superseded local O6 camera calibrations. Keep the active extrinsics and
# intrinsics referenced by o6_right_product.yaml available for version control.
/config/o6_three_camera_extrinsics_20260915.yaml
/config/o6_three_camera_extrinsics_intrinsics_20260915.yaml
/config/o6_three_camera_extrinsics_recalibrated.yaml
*.wear_check.json
*.checkpoint.json
*.verification.json
*_mapping_quality.json
# Device-specific robot descriptions derived from local calibration runs
# Includes full/partial zero-calibration outputs and local copies.
/src/linkerhand_calibration/urdf/*/*_calibrated_*.urdf
/src/linkerhand_calibration/urdf/*/*_zero_calibrated*.urdf
/src/linkerhand_calibration/urdf/*/*_transferred_from_*.urdf
# Unreferenced local URDF experiment; never use as a protected CAD input.
/src/linkerhand_calibration/urdf/o6_right/linkerhand_o6_right111.urdf
/src/linkerhand_retarget/linkerhand_retarget/assets/robots/hands/linker_hand/g20_left/linkerhand_g20_left_cmc_pitch_*.urdf
/src/linkerhand_retarget/linkerhand_retarget/assets/robots/hands/linker_hand/g20_left/linkerhand_g20_left_calibrated_*.urdf
/src/linkerhand_retarget/linkerhand_retarget/assets/robots/hands/linker_hand/g20_left/linkerhand_g20_left_zero_calibrated_*.urdf
/src/linkerhand_retarget/linkerhand_retarget/assets/robots/hands/linker_hand/g20_right/linkerhand_g20_right_cmc_pitch_*.urdf
/src/linkerhand_retarget/linkerhand_retarget/assets/robots/hands/linker_hand/g20_right/linkerhand_g20_right_calibrated_*.urdf
/src/linkerhand_retarget/linkerhand_retarget/assets/robots/hands/linker_hand/g20_right/linkerhand_g20_right_zero_calibrated_*.urdf
/src/linkerhand_retarget/linkerhand_retarget/assets/robots/hands/linker_hand/g20_right/linkerhand_G20_RIGHT_tag.urdf
# ROS bag / MCAP recordings and CAN captures
rosbag2_*/
/bags/
/recordings/
/captures/
/sessions/
/reports/
*.db3
*.mcap
candump-*
l10_*_state_*/
*_state_*/
# Local Codex/agent workspace metadata
/.agents/
/.codex/
/.codebuddy/
/.zcode/
+2
View File
@@ -0,0 +1,2 @@
1.不要补丁式修复问题,避免产生屎山代码,要求代码结构清晰可读性好。
2.永远用中文回复。
+4 -1
View File
@@ -1,5 +1,8 @@
# 1. LinkerFFG手套
> 当前FFG多手势标定、G20/O6 profile映射及实机操作请优先参考:
> [FFG多手势标定映射与遥操操作说明](docs/FFG多手势标定映射与遥操操作说明.md)。
## 1.1 产品介绍
本产品的具体介绍参考,内含标定示例说明
附件1、Linker FFG(FFG01)产品说明手册
@@ -416,4 +419,4 @@ if self.calibrationoriginal is not None and self.calibrationfistpose is not None
改写成如上图的示例,即可启用右机械手的校准,启动后就会让机械手按照映射的角度固定在当前角度
当左右两手都达到期望的对指位置后,就可以恢复原状,按正常顺序使用遥操系统
当左右两手都达到期望的对指位置后,就可以恢复原状,按正常顺序使用遥操系统
@@ -0,0 +1,55 @@
image_width: 1624
image_height: 1240
camera_name: hikrobot_top_DB2163739
camera_matrix:
rows: 3
cols: 3
data:
- 3608.7915725635876
- 0.0
- 814.529414176388
- 0.0
- 3602.9474866035775
- 596.7866652774017
- 0.0
- 0.0
- 1.0
distortion_model: plumb_bob
distortion_coefficients:
rows: 1
cols: 5
data:
- -0.05774549441026762
- -0.5653814259405426
- -0.0034831870836035525
- -0.0007572304719307008
- 0.0
rectification_matrix:
rows: 3
cols: 3
data:
- 1.0
- 0.0
- 0.0
- 0.0
- 1.0
- 0.0
- 0.0
- 0.0
- 1.0
projection_matrix:
rows: 3
cols: 4
data:
- 3590.9833984375
- 0.0
- 813.6314052451635
- 0.0
- 0.0
- 3591.4892578125
- 595.0547132430074
- 0.0
- 0.0
- 0.0
- 1.0
- 0.0
@@ -0,0 +1,55 @@
image_width: 1624
image_height: 1240
camera_name: hikrobot_front_DB2163742
camera_matrix:
rows: 3
cols: 3
data:
- 3520.83963745858
- 0.0
- 971.8024440668337
- 0.0
- 3503.4665942352735
- 636.2092891200001
- 0.0
- 0.0
- 1.0
distortion_model: plumb_bob
distortion_coefficients:
rows: 1
cols: 5
data:
- -0.06456607194260039
- -0.3209716179987898
- 0.0016533290649511723
- 0.014394085090833234
- 0.0
rectification_matrix:
rows: 3
cols: 3
data:
- 1.0
- 0.0
- 0.0
- 0.0
- 1.0
- 0.0
- 0.0
- 0.0
- 1.0
projection_matrix:
rows: 3
cols: 4
data:
- 3485.41357421875
- 0.0
- 980.9585478080553
- 0.0
- 0.0
- 3500.062255859375
- 636.4474700149367
- 0.0
- 0.0
- 0.0
- 1.0
- 0.0
@@ -0,0 +1,55 @@
image_width: 1624
image_height: 1240
camera_name: hikrobot_side_DB2163749
camera_matrix:
rows: 3
cols: 3
data:
- 3564.4557723541934
- 0.0
- 633.5422135270659
- 0.0
- 3549.8257661608695
- 614.1194434338138
- 0.0
- 0.0
- 1.0
distortion_model: plumb_bob
distortion_coefficients:
rows: 1
cols: 5
data:
- -0.023159717917942222
- -0.6611401590594669
- 0.0010272016326466154
- -0.014523561130335804
- 0.0
rectification_matrix:
rows: 3
cols: 3
data:
- 1.0
- 0.0
- 0.0
- 0.0
- 1.0
- 0.0
- 0.0
- 0.0
- 1.0
projection_matrix:
rows: 3
cols: 4
data:
- 3531.000244140625
- 0.0
- 623.4124604711214
- 0.0
- 0.0
- 3551.248291015625
- 614.0312247691472
- 0.0
- 0.0
- 0.0
- 1.0
- 0.0
@@ -0,0 +1,60 @@
schema_version: 1
reference_view: front
cameras:
front:
serial_number: DB2163742
width: 1624
height: 1240
intrinsics_sha256: 3cdecb8fcede41561019882a48a342d7265a44855cf0282ec96ec7f69a9b1ee5
side:
serial_number: DB2163749
width: 1624
height: 1240
intrinsics_sha256: dbe0cdc73460dc1e51c0e3967395d14b4415a71cee952597e0717ad0babd3449
top:
serial_number: DB2163739
width: 1624
height: 1240
intrinsics_sha256: 315d5cd6f28964ad57ff67088222fff3d9679cce076acc3f05a22b76038e5352
front_from_view:
front:
translation_xyz_m:
- 0.0
- 0.0
- 0.0
quaternion_xyzw:
- 0.0
- 0.0
- 0.0
- 1.0
side:
translation_xyz_m:
- -0.8459732881345877
- -0.000555404488719817
- 0.9729938257334647
quaternion_xyzw:
- -0.04404064417946292
- 0.6703485002266404
- -0.01659275028535042
- 0.7405524900654377
top:
translation_xyz_m:
- -0.10806018719194868
- -0.5842672050520756
- 1.0419844264616385
quaternion_xyzw:
- 0.7284460604085318
- -0.016210595429770967
- 0.019511507802845565
- -0.6846333724953536
quality:
passed: true
reprojection_rms_px: 0.46496404895142424
maximum_rotation_repeatability_deg: 0.019287272875226844
maximum_translation_repeatability_m: 0.0002246640979668814
front_side_captures: 15
front_top_captures: 15
front_side_candidates: 15
front_top_candidates: 15
front_side_rejected: 0
front_top_rejected: 0
@@ -0,0 +1,985 @@
# FFG多手势标定映射与遥操技术实现
## 1. 文档定位
本文面向 `linkerforce_v2` 的开发、联调和维护人员,说明新版FFG手套映射遥操链路的
软件架构、标定拟合方法、实时映射算法、ROS 2接口、profile约束和安全门控。
实机标定与启动步骤见
[《FFG多手势标定映射与遥操操作说明》](./FFG多手势标定映射与遥操操作说明.md)。
本文不重复完整操作流程,而是回答以下实现问题:
- 21维FFG数据如何变成模型无关的手部语义;
- 张手、桌面、钩拳、握拳四个锚点如何解耦根部和末端屈伸;
- G20和O6如何共用手套语义、同时保持各自的机械执行标尺;
- 捏合与握持为什么不会把整只手锁定到离散模板;
- `cmd_u8``actuation_target``q_nominal`三类目标有什么区别;
- profile如何生成、校验、配对和追踪;
- 节点在什么条件下允许或撤销实机控制。
当前实现基于:
- profile schema`schema_version=1`
- 手套侧:单只左手FFG,21维输入;
- 机械手侧:左手G20和左手O6
- 映射模式:`factorized_paired_v2`
- 机械手profile策略:`paired_continuous_v1`
- 仿真策略:`semantic_urdf_v1`
- 标定等级:`provisional`,没有真实关节角GT。
## 2. 代码组织
核心实现位于
[`linkerforce_v2`](../src/linkerhand_retarget/linkerhand_retarget/motion/linkerforce_v2/)
| 文件 | 职责 |
|---|---|
| `constants.py` | FFG关节名、手部语义名、静态/动态手势集合和默认参数 |
| `calibrate_glove.py` | 订阅FFG原始话题,交互采集完整标定和快速佩戴检查 |
| `calibrate_robot.py` | 从GUI命令、快照和SDK状态生成G20/O6实机profile |
| `calibration.py` | 鲁棒统计、FFG特征拟合、机械手通道权重和分段曲线拟合 |
| `profiles.py` | profile加载、严格校验、规范化哈希和原子保存 |
| `mapping.py` | 人手语义提取、连续配对映射、捏合/握持修正和命令滤波 |
| `node.py` | ROS 2实时节点、话题、服务、定时循环和安全门控 |
| `safety.py` | 与ROS无关的超时撤权判定 |
| `simulation.py` | 按关节名重排仿真目标并执行限位检查 |
| `quality.py` | 静态、捏合、动态轨迹的离线回放质量检查 |
| `verify_robot_profile.py` | 低速回放机械手标定姿势并生成独立人工复核报告 |
| `session_manifest.py` | 生成provisional数采清单并绑定profile、URDF和设备信息 |
ROS 2入口在
[`setup.py`](../src/linkerhand_retarget/setup.py),双手机型启动文件为
[`ffg_dual_g20_o6.launch.py`](../src/linkerhand_retarget/launch/ffg_dual_g20_o6.launch.py)
可提交的机械手种子profile位于
[`resource/linkerforce_v2/profiles`](../src/linkerhand_retarget/resource/linkerforce_v2/profiles/)。
## 3. 总体架构
```text
┌──────────────────────────────┐
FFG串口 / ROS JointState│ 21维左手套原始弧度 raw[21] │
└──────────────┬───────────────┘
│ 可选逐维Kalman
┌──────────────────────────────┐
│ HandIntentExtractor │
│ 21维 → 22维0~1人手语义 │
└──────────────┬───────────────┘
│ 同一hand_intent
┌───────────────────┴───────────────────┐
▼ ▼
┌───────────────────┐ ┌───────────────────┐
│ G20 RobotMapper │ │ O6 RobotMapper │
│ 16个主动执行语义 │ │ 6个主动执行语义 │
└─────────┬─────────┘ └─────────┬─────────┘
│ │
┌─────────┴─────────┐ ┌─────────┴─────────┐
▼ ▼ ▼ ▼
20维cmd_u8 16维q_nominal 6维cmd_u8 6维q_nominal
G20实机电机空间 G20仿真弧度目标 O6实机电机空间 O6仿真弧度目标
```
架构的关键是分成三层:
1. **传感器层**:FFG原始值和佩戴差异由手套profile吸收;
2. **解剖语义层**`hand_intent`只描述人的手部动作,不依赖G20或O6;
3. **执行器层**:每个机械手profile独立定义语义到本机电机命令和URDF目标的映射。
因此,换机械手通常不需要改手套语义提取器;换手套或操作者也不需要改G20/O6的
通道定义,只需重新建立对应profile并在运行时配对。
## 4. 数据契约
### 4.1 FFG 21维原始数据
原始输入使用 `sensor_msgs/msg/JointState`。名称和规范顺序为:
```text
thumb_0 ... thumb_4
index_0 ... index_3
middle_0 ... middle_3
ring_0 ... ring_3
pinky_0 ... pinky_3
```
串口解析器将设备角度转换为弧度。实时节点只接受长度为21、全部有限的数据。
- 串口模式直接使用上述顺序;
- topic模式允许输入名称顺序不同,但要求名称集合完整且唯一,节点按规范顺序重排;
- 标定工具要求输入消息已经使用规范顺序;
- 不支持右手FFG,也不会因右手套缺失而退出。
普通四指每指4个原始量:第0维主要用于侧摆,第1~3维共同参与根部和末端屈伸拟合。
拇指5个原始量由不同组合共同拟合旋转、外展、对掌、根部屈伸和末端屈伸。
### 4.2 人手语义
`hand_intent`共22维,所有值限制在 `[0, 1]`
| 类别 | 名称 |
|---|---|
| 拇指 | `thumb_rotate``thumb_abduction``thumb_opposition``thumb_root``thumb_tip` |
| 四指屈伸 | 每指的 `<finger>_root``<finger>_tip` |
| 四指侧摆 | 每指的 `<finger>_splay` |
| 捏合证据 | `pinch_index``pinch_middle``pinch_ring``pinch_pinky` |
| 整体握持 | `power_grasp` |
这里的0和1是由个人手套标定定义的语义端点,不是机械手角度,也不代表统一的物理角度。
### 4.3 G20命令空间
G20输出完整20维 `cmd_u8`
| 下标 | 通道 |
|---:|---|
| 0 | `thumb_cmc_pitch` |
| 14 | `index/middle/ring/pinky_mcp_pitch` |
| 5 | `thumb_cmc_roll` |
| 69 | `index/middle/ring/pinky_mcp_roll` |
| 10 | `thumb_cmc_yaw` |
| 1114 | `reserved_11``reserved_14`,固定为255 |
| 15 | `thumb_mcp` |
| 1619 | `index/middle/ring/pinky_pip` |
其中16个通道是主动映射通道,4个保留通道不参与映射。新版种子profile中
`thumb_cmc_yaw`使用完整的 `[0, 255]` 命令范围,不再继承旧版的80下限。
### 4.4 O6命令空间
O6输出6维 `cmd_u8`
```text
thumb_cmc_pitch
thumb_cmc_yaw
index_mcp_pitch
middle_mcp_pitch
ring_mcp_pitch
pinky_mcp_pitch
```
O6没有独立的四指PIP和侧摆执行通道,因此每个普通手指的单一屈伸通道由
`root``tip`语义融合得到。
### 4.5 三种输出标尺
| 输出 | 范围/单位 | 含义 |
|---|---|---|
| `actuation_target` | `[0,1]` | 当前型号各主动通道的归一化语义激活量 |
| `cmd_u8_preview` / 实机命令 | `[0,255]` | 设备电机命令空间,包含机械耦合和本机标定 |
| `joint_target_nominal` | rad | 根据语义激活量和URDF名义端点生成的仿真目标 |
`state_u8`是SDK返回的设备状态,仍属于设备空间。它既不是编码器关节角,也不能作为
`q_nominal`或真实物理关节角的GT。
## 5. FFG手套profile的生成
### 5.1 鲁棒采样统计
每次采集保留:
```text
sample_count
median[21]
mad[21]
raw_frames[N][21]
```
对第 `j` 维:
```text
median_j = median(raw[:, j])
MAD_j = median(abs(raw[:, j] - median_j))
```
每个静态姿势和动态轨迹还保存3次独立重复的上述统计。`approved_for_runtime=true`
要求:
- 11个静态姿势全部存在;
- 7个动态轨迹全部存在;
- 每项恰好3次重复;
- 每次至少50帧;
- 汇总帧数等于3次重复的帧数之和。
因此,CLI虽然允许修改静态 `--repeats`,但不是3次时生成的profile只能用于预览。
### 5.2 基础语义特征拟合
每个语义特征定义一组原始下标和带标签的标定姿势。以某个特征为例:
1. 从语义标签为0的姿势求原始端点 `low`
2. 从语义标签为1的姿势求原始端点 `high`
3. 将各标定姿势归一化为:
```text
n_j = clip((raw[index_j] - low_j) / (high_j - low_j), 0, 1)
```
4. 用最小二乘拟合各原始维度对语义标签的贡献;
5. 将负权重截为0,再归一化为权重和1;
6. 运行时计算:
```text
feature = clip(sum(weight_j * n_j), 0, 1)
```
无有效跨度的维度不参与归一化。若拟合后所有权重都接近0,则回退为等权。
普通四指的 `root``tip`故意使用同一组3个屈伸传感器,但使用不同姿势标签:
| 姿势 | root目标 | tip目标 |
|---|---:|---:|
| 张手/并拢 | 0 | 0 |
| 桌面 | 1 | 0 |
| 钩拳 | 0 | 1 |
| 握拳 | 1 | 1 |
这一设计先得到两个可能仍有耦合的初始特征,再由下一步二维标定面解耦。
### 5.3 根部—末端双线性解耦
对每个普通手指,在初始 `(root_feature, tip_feature)` 平面中取得四个锚点:
```text
p00 = 张手
p10 = 桌面
p01 = 钩拳
p11 = 握拳
```
建立双线性标定面:
```text
p(u, v) = p00
+ (p10 - p00) * u
+ (p01 - p00) * v
+ (p11 - p10 - p01 + p00) * u * v
```
其中 `u`是解耦后的根部屈伸,`v`是解耦后的末端屈伸。运行时先用线性最小二乘得到
初值,再执行最多5次Newton迭代反解 `(u, v)`,最后限制到 `[0,1]`
只有标定四边形在四角的Jacobian行列式符号一致,且最小绝对值不小于 `1e-3` 时才启用
该解码器。退化或发生折叠的标定面不会用于反解,此时保留基础特征结果。
### 5.4 动态屈伸对侧摆的串扰补偿
每个普通手指的独立屈伸往返轨迹假设该手指侧摆应基本不变。对每次重复:
```text
x = 0.5 * (root + tip)
y = splay
x, y分别减去各自中位数
coefficient = dot(x, y) / (dot(x, x) + ridge)
```
其中 `ridge = 1e-3 * max(dot(x,x), 1e-6)`
以下情况拒绝学习该次轨迹:
- 少于10帧、长度错误或存在非有限值;
- 屈伸变化范围小于0.05
- 侧摆几乎没有变化,无法估计;
- 补偿后残差方差仍大于原方差的80%;
- 三次重复的有效系数方向互相矛盾。
最终系数取各有效重复的中位数并限制到 `[-1,1]`,运行时执行:
```text
splay_corrected = clip(
splay - coefficient * 0.5 * root
- coefficient * 0.5 * tip,
0,
1
)
```
补偿只发生在人手语义层,不直接学习或修改任何G20/O6电机系数。
### 5.5 捏合证据
每种捏合只使用拇指5维和目标手指4维。标定时保存:
```text
center = 目标捏合姿势中位数
scale = max(abs(center - open), 6 * pinch_pose_MAD, 0.02)
```
并计算张手到捏合中心的归一化距离 `open_distance`。运行时:
```text
distance = RMS((raw_selected - center) / scale)
pinch_strength = clip(1 - distance / open_distance, 0, 1)
```
这4个值是候选证据,不直接等于4个离散状态;最终是否施加捏合修正还要经过竞争门控。
### 5.6 快速佩戴检查
快速检查重新采集张手、握拳和食指捏合。每个姿势计算:
```text
normalized_error =
RMS((observed_median - reference_median) / max(6 * MAD, 0.05))
```
默认要求每个误差不大于4.0。凭据保存当前手套profile的规范化SHA-256、检查时间、
阈值、各姿势误差和通过状态。
实时节点只在加载profile时检查凭据:
- `kind=ffg_wear_check`
- `passed=true`
- 绑定哈希等于当前手套profile哈希;
- 凭据年龄在配置范围内,默认12小时。
节点不会在长时间运行期间周期性重新读取凭据或重新计算年龄。需要跨时段运行时,应按
作业流程主动重启节点并重新执行佩戴检查。
## 6. 机械手profile的生成
### 6.1 捕获数据
`hand_pose_capture`同时监听:
- GUI连续命令 `/<model>/cb_left_hand_control_cmd`
- SDK状态 `/<model>/cb_left_hand_state`
- GUI保存快照 `/<model>/calibration_pose_snapshot`
消息名称允许任意顺序,但必须与seed中的 `command_names`集合完全一致;保存前统一重排为
profile顺序。每个姿势保存:
```text
cmd_u8
command_names
state_u8
state_names
status = exact | approximate | unsupported
confirmed
captured_at
```
每次人工确认后立即原子写入checkpoint。恢复时会核对seed哈希、型号、输出路径、
SN、固件、CAN、操作者、命令名和姿势列表;身份不一致时拒绝续标。若旧checkpoint中
某个命令超出新的安全范围,只删除该姿势并要求重拍。
最终 `approved_for_control=true` 同时要求:
- 用户在最后明确批准;
- SN、CAN和操作者非空;
- 所有必需姿势均已确认;
- 命令与状态名称完整;
- 所有命令位于profile安全范围。
`unsupported`表示该姿势不参与对应运行时约束,但该姿势记录本身仍需人工确认并保存。
### 6.2 多源执行通道权重
一个机械手主动通道可以融合多个人手语义。对seed中列出的 `fit_sources`,使用
张手、桌面、钩拳和握拳的实机命令拟合。
先以张手和握拳命令归一化该通道:
```text
y_pose = (cmd_pose - cmd_open) / (cmd_fist - cmd_open)
```
设计矩阵来自各姿势的规范语义目标,然后执行最小二乘;负权重截为0并归一化。
- G20大部分主动通道只有一个语义源;
- O6普通手指通道同时使用对应的 `root``tip`,权重由实机捕获结果拟合;
- 若张手和握拳命令跨度退化,回退为等权。
### 6.3 分段曲线与单调约束
profile生成时,根据通道语义激活量和各标定姿势的 `cmd_u8`产生曲线点。
- `piecewise`:同一激活量的命令取中位数,然后按激活量排序;
- `monotonic_piecewise`:在上述基础上使用相邻违例合并算法执行等距单调回归;
- 曲线至少需要两个不同的激活量;
- 命令点必须位于 `[0,255]`和该通道 `command_bounds`内。
该profile曲线是一条可独立验证和追踪的型号级基线,也是在运行时手套锚点退化时的
单通道回退曲线。
## 7. 运行时分解式配对映射
### 7.1 配对曲线构造
`RobotMapper`同时收到手套profile和机械手profile时,不直接使用抽象规范姿势坐标,
而是:
1. 用当前手套profile的静态姿势中位数重新计算真实 `hand_intent`
2. 找出手套和机械手共有且未标为 `unsupported` 的姿势;
3. 对每个机械手主动通道,选择真正定义该解剖通道的姿势;
4. 以当前手套语义激活量为横轴、当前实机profile命令为纵轴重建分段曲线;
5. 对单调通道再次执行单调回归。
姿势选择规则为:
| 通道 | 使用的基础姿势 |
|---|---|
| 普通四指屈伸 | 张手、并拢、桌面、钩拳、握拳中双方共有的姿势 |
| G20普通四指侧摆 | 并拢、张手 |
| 拇指基础通道 | 最大外展、张手、横跨掌心 |
捏合姿势不进入普通通道曲线,握拳也不直接进入拇指基础曲线;它们分别由局部残差分支
处理。这样,某个捏合捕获中的非目标手指残留命令不会污染普通手指曲线。
如果某个通道的实际手套锚点退化为少于两个不同激活量,该通道使用机械手profile中
已经校验的曲线;其他通道仍可保持配对曲线。
完成构造后,`mapping_mode``factorized_paired_v2`
### 7.2 基础通道映射
对第 `k` 个主动通道,其来源权重为 `w_ki`,当前人手语义为 `h_i`
```text
a_k = clip(sum(w_ki * h_i) / sum(w_ki), 0, 1)
```
`a_k`组成 `actuation_target`。基础电机命令由该通道配对曲线分段线性插值得到:
```text
cmd_base[index_k] = piecewise_linear(a_k, paired_points_k)
```
完整命令向量先以张手命令初始化,主动通道逐个覆盖;未映射的保留通道之后强制写回
固定值。
### 7.3 竞争式局部捏合修正
#### 7.3.1 标定自适应阈值
映射器先对手套profile中的所有静态姿势计算4种捏合分数。对每个真实捏合姿势记录:
- 目标分数;
- 目标分数相对其他3种分数的领先量。
对所有非捏合姿势记录:
- 最大误触分数;
- 第一名相对第二名的误触领先量。
满激活阈值取4个目标姿势中的最弱值,起始阈值位于最大负样本与满激活阈值之间的20%:
```text
score_onset = negative_score + 0.2 * (score_full - negative_score)
margin_onset = negative_margin + 0.2 * (margin_full - negative_margin)
```
若当前手套profile无法在正负样本间形成有效分数或领先量间隔,所有捏合门均保持0。
#### 7.3.2 单赢家连续门控
运行时仅选择当前分数最高的候选,并计算:
```text
score_gate = smoothstep((top_score - score_onset) / score_span)
margin_gate = smoothstep((top_score - second_score - margin_onset) / margin_span)
pinch_gate = score_gate * margin_gate
```
`smoothstep(x)=x²(3-2x)`,输入先限制到 `[0,1]`。其他3种捏合门为0。
因此:
- 证据不足时保持普通连续映射;
- 两种捏合证据接近时,领先量门控将修正降到0;
- 不存在确认帧数、进入/退出滞回或历史姿势锁存;
- 捏合切换只依赖当前帧,且权重连续变化。
#### 7.3.3 局部残差
对每种捏合,在该手套捏合中位数处先计算基础命令,再与机械手目标捏合命令做差:
```text
residual = robot_pinch_target - base_command_at_glove_pinch
```
运行时只把 `pinch_gate * residual`加到:
- 所有拇指主动通道;
- 当前目标手指的主动通道。
其他3根手指不参与该分支。G20的保留通道也不参与。
### 7.4 握持时的拇指协调
`power_grasp`是8个普通四指 `root/tip`语义的平均值。握持分数进一步要求拇指主动折叠:
```text
grasp_score = min(
power_grasp,
thumb_opposition,
thumb_root,
thumb_tip
)
```
满分取手套握拳姿势,负样本取其他静态姿势的最高分,门控同样使用从负样本到握拳分数
20%处开始的 `smoothstep`
握持残差是机械手握拳目标与握拳处基础命令的差,但只施加到拇指主动通道。四指仍由
各自连续屈伸曲线决定,普通拇指动作也不会仅因四指弯曲而被强制成握拳拇指。
### 7.5 安全范围与保留通道
局部修正完成后依次执行:
1. 写回保留通道固定值;
2. 按每通道 `command_bounds`裁剪;
3. 执行可选命令滤波;
4. 再次裁剪并再次写回保留通道;
5. 最终四舍五入为整数命令。
`raw_command`保留滤波前浮点目标,当前ROS节点不发布该字段;`cmd_u8_preview`发布滤波后
并取整的最终目标。
## 8. 滤波与实时执行
### 8.1 输入Kalman
输入滤波是21个互相独立的一维Kalman滤波器,共享参数:
```text
P_pred = P + process_variance
K = P_pred / (P_pred + measurement_variance)
x = x + K * (z - x)
P = (1 - K) * P_pred
```
首帧、时间倒退或帧间隔超过 `input_filter_reset_gap` 时直接重置到当前测量,避免断流后
从旧状态缓慢追赶。默认关闭。
### 8.2 输出命令滤波
`CommandFilter`支持:
| 模式 | 行为 |
|---|---|
| `passthrough` | 直接使用本帧目标,仅应用deadband |
| `ema` | `step=clip(alpha*(target-last), ±max_step)` |
| `acceleration_limited` | 同时限制速度、帧间加速度,并根据剩余距离提前制动 |
默认参数匹配旧版左手G20的有效执行路径:
```text
input_filter_enabled=false
command_filter_mode=passthrough
command_filter_ema_alpha=1.0
command_filter_max_step_u8=255
command_filter_deadband_u8=0
```
实时节点没有单独暴露 `command_filter_max_acceleration_u8_per_frame2` 参数;
`acceleration_limited`模式下它使用与 `command_filter_max_step_u8`相同的值。
### 8.3 30 Hz最新帧策略
实时节点的处理定时器默认30 Hz
1. 串口模式从线程安全快照取得最新序列号、数据和接收时刻;
2. 只有出现新FFG序列时,才发布/更新raw、filtered、intent和frame metadata
3. 每个定时周期都用最近一次有效intent重新计算两个型号目标;
4. 预览始终发布,只有已使能型号才发布到SDK命令话题;
5. 硬件命令QoS为 `RELIABLE + KEEP_LAST(depth=1)`
固定控制心跳不会排队重放旧手套帧。FFG停止更新时,节点可在超时窗口内短暂复用最后
intent,随后watchdog撤销使能。
使能某型号时,命令滤波器会重置到该型号最新有效SDK状态,而不是张手或上一次内部
目标,从而降低重新使能的第一帧跳变。
## 9. 独立仿真目标
仿真目标不从 `cmd_u8`反解。对主动通道激活量 `a_k`
```text
q_nominal_k = clip(
q_open_k + a_k * (q_closed_k - q_open_k),
q_lower_k,
q_upper_k
)
```
其输入是基础解剖通道激活量,不使用电机命令曲线,也不直接使用捏合或握持的电机残差。
因此实机姿势捕获中的机械耦合、偶然残留值和保留通道不会污染仿真弧度目标。
仿真消费者必须按 `JointState.name`建立映射。`simulation.py`提供:
- `build_name_mapping()`:检查空名、重名、缺名和多余名称;
- `reorder_named_target()`:按仿真模型顺序重排,检查有限值并应用仿真限位。
名称合同不满足时抛出 `JointNameMismatch`,不得按裸下标猜测。
profile中的 `urdf_sha256`用于追踪生成名义端点时对应的URDF版本,但实时映射节点本身
不读取或重新计算URDF文件哈希;数采manifest工具会执行文件哈希核对。
## 10. ROS 2实时节点
### 10.1 输入与输出话题
| 话题 | 类型 | 维度 | 发布条件 |
|---|---|---:|---|
| `/ffg/left/raw_joint_state` | `JointState` | 21 | 串口模式收到新帧;topic模式直接使用上游话题 |
| `/ffg/left/filtered_joint_state` | `JointState` | 21 | 有有效手套profile和新帧 |
| `/retarget/left/hand_intent` | `JointState` | 22 | 有有效手套profile和新帧 |
| `/retarget/left/frame_meta` | `String(JSON)` | - | 每个新映射手套帧 |
| `/retarget/g20/left/actuation_target` | `JointState` | 16 | G20 mapper有效 |
| `/retarget/o6/left/actuation_target` | `JointState` | 6 | O6 mapper有效 |
| `/retarget/g20/left/joint_target_nominal` | `JointState` | 16 | G20 mapper有效 |
| `/retarget/o6/left/joint_target_nominal` | `JointState` | 6 | O6 mapper有效 |
| `/retarget/g20/left/cmd_u8_preview` | `JointState` | 20 | G20 mapper有效 |
| `/retarget/o6/left/cmd_u8_preview` | `JointState` | 6 | O6 mapper有效 |
| `/g20/cb_left_hand_control_cmd` | `JointState` | 20 | G20已使能 |
| `/o6/cb_left_hand_control_cmd` | `JointState` | 6 | O6已使能 |
| `/ffg_dual_retarget/status` | `String(JSON)` | - | 1 Hz |
节点订阅:
| 话题 | 说明 |
|---|---|
| `raw_input_topic` | `input_mode=topic`时的FFG输入,默认 `/ffg/left/raw_joint_state` |
| `/g20/cb_left_hand_state` | G20驱动状态心跳 |
| `/o6/cb_left_hand_state` | O6驱动状态心跳 |
topic输入模式不会再次向raw话题发布收到的消息,避免默认同名话题形成反馈。
驱动状态的名称必须已经按profile `command_names`规范顺序排列;这里与FFG topic输入不同,
不会对驱动状态按集合重排。
### 10.2 帧元数据
`frame_meta`包含:
```json
{
"timestamp_ns": 0,
"sequence": 0,
"calibration": "provisional",
"q_gt": null
}
```
一个新手套帧产生的filtered、intent、frame_meta及该次定时周期的型号目标共用ROS时间戳。
定时器复用旧intent时,型号目标使用新的当前时间戳,但不会重复发布intent和frame_meta。
### 10.3 状态诊断
1 Hz状态JSON包含:
- 当前输入模式、FFG帧序号和数据年龄;
- G20/O6分别是否使能;
- glove、G20、O6的批准状态和SHA-256
- `mapping_mode``simulation_mapping_mode`
- 输入和命令滤波配置;
- 当前捏合/握持局部分支权重 `anchor_weights`
- wear-check有效性;
- 驱动状态年龄;
- profile和URDF哈希;
- profile加载错误、最近故障和p95调度延迟。
`latency_p95_ms`以本机接收手套数据的单调时钟为起点,表示接收至映射调度的延迟,
不是基于设备硬件时间戳的端到端链路延迟。
### 10.4 服务与状态转换
| 服务 | 类型 | 作用 |
|---|---|---|
| `~/enable_g20` | `SetBool` | 单独申请/撤销G20实机控制 |
| `~/enable_o6` | `SetBool` | 单独申请/撤销O6实机控制 |
| `~/enable_all` | `SetBool` | 原子检查两台后同时使能,或同时撤销 |
| `~/emergency_stop` | `Trigger` | 立即撤销两个型号的命令发布权限 |
```text
显式SetBool(true)且全部检查通过
┌──────────────────────────────────────────┐
│ ▼
PREVIEW / DISABLED ENABLED(model)
▲ │
└──────────────────────────────────────────┘
SetBool(false)、超时、映射异常或软件急停
```
撤销使能的含义是停止向SDK命令话题发布新命令,不会主动发送张手、零位或其他安全姿势。
驱动/固件将保持最后命令相关行为;物理急停仍应由系统级安全链路负责。
## 11. 实机使能门控
某个型号从PREVIEW进入ENABLED前依次检查:
1. 手套profile已加载且 `approved_for_runtime=true`
2. wear-check已通过启动时校验;
3. 机械手profile已加载且 `approved_for_control=true`
4. 启动参数中的期望SN非空,并与profile SN完全一致;
5. profile CAN接口与启动配置一致;
6. mapper为 `factorized_paired_v2`
7. FFG最近一帧未超过 `glove_timeout`,默认0.35秒;
8. 对应SDK状态名称、长度、数值和范围有效,且未超过 `driver_timeout`,默认1秒。
运行时身份门控比较SN和CAN接口,不比较profile中的固件版本。固件兼容性目前依赖操作
流程和数采manifest的可选校验;若固件变更会改变电机响应,应重新标定机械手profile。
`enable_all`先检查G20和O6两者,任意一个失败都不会使能任何一个。单型号服务互相独立。
## 12. Watchdog与故障策略
watchdog周期为50 ms
| 故障 | 动作 |
|---|---|
| 任意型号已使能且FFG超时 | 同时撤销G20和O6 |
| 某型号SDK状态无效或超时 | 只撤销该型号,另一型号保持 |
| 映射计算出现数值/形状错误 | 同时撤销G20和O6 |
| 软件急停 | 同时撤销G20和O6 |
| profile加载失败 | 启动时降级;不创建对应mapper或只保留raw |
故障恢复不会自动重新使能。排除原因后必须再次调用对应使能服务。
节点启动和profile加载采用fail-closed策略:
- 无手套profile:只发布原始FFG
- 手套profile有效但未获运行批准:允许完整预览,拒绝实机;
- seed机械手profile `approved_for_control=false`:允许预览,拒绝实机;
- 单个型号profile无效:另一个有效型号仍可生成目标和独立使能。
## 13. 主要ROS参数
### 13.1 FFG输入
| 参数 | 默认值 | 说明 |
|---|---|---|
| `input_mode` | `serial` | `serial``topic` |
| `raw_input_topic` | `/ffg/left/raw_joint_state` | topic模式输入 |
| `serial_port` | 空 | 指定串口;为空时可自动扫描 |
| `baudrate` | `0` | 大于0时优先尝试该波特率 |
| `baudrates` | `[2000000,460800,1000000,921600]` | 探测候选 |
| `auto_scan` | `true` | 指定端口失败或为空时扫描 |
| `serial_debug` | `false` | 串口调试日志 |
### 13.2 Profile与身份
| 参数 | 默认值 | 说明 |
|---|---|---|
| `glove_profile` | 空 | FFG完整标定profile |
| `wear_check` | 空 | 快速佩戴检查凭据 |
| `wear_check_max_age_hours` | `12.0` | 启动加载时允许的最大年龄 |
| `g20_profile` / `o6_profile` | 空 | 单机机械手profile |
| `g20_serial_number` / `o6_serial_number` | 空 | 运行期望SN |
| `g20_can_interface` | `can0` | G20身份核对 |
| `o6_can_interface` | `can1` | O6身份核对 |
### 13.3 时序和滤波
| 参数 | 默认值 | 说明 |
|---|---:|---|
| `publish_rate` | `30.0` | 固定映射/命令心跳Hz |
| `glove_timeout` | `0.35` | FFG超时秒数 |
| `driver_timeout` | `1.0` | SDK状态超时秒数 |
| `input_filter_enabled` | `false` | 是否启用逐维Kalman |
| `input_filter_process_variance` | `1e-5` | Kalman过程噪声 |
| `input_filter_measurement_variance` | `5e-4` | Kalman测量噪声 |
| `input_filter_reset_gap` | `0.35` | 断流重置阈值 |
| `command_filter_mode` | `passthrough` | 输出滤波模式 |
| `command_filter_ema_alpha` | `1.0` | EMA/限加速度目标增益 |
| `command_filter_max_step_u8` | `255.0` | 每帧最大速度尺度 |
| `command_filter_deadband_u8` | `0.0` | 小于该差值时保持上一目标 |
启动文件还负责创建两个SDK节点,并设置启动速度、力矩、状态轮询和G20控制期间延迟状态
读取等驱动参数;这些不是 `ffg_dual_retarget`自身参数。
## 14. Profile校验与可追踪性
### 14.1 严格加载
`profiles.py`在构造mapper之前检查:
- schema、profile类型、型号和左手侧;
- 固定的FFG关节名或机械手命令名;
- 所有数组长度、有限值和范围;
- 特征权重非负且和为1
- 分段曲线激活量、命令范围和单调性;
- 主动通道与保留通道完整覆盖命令向量;
- 仿真名称顺序、端点、限位和URDF哈希格式;
- 已批准profile的设备身份、人工确认、状态和名称完整性。
profile错误不会被静默修正为另一种型号或旧映射策略。
### 14.2 规范化哈希
profile哈希不是原文件字节哈希,而是:
1. 排除加载器添加的 `_profile_path``_profile_sha256`
2. JSON key排序;
3. 使用紧凑分隔符和UTF-8
4. 计算SHA-256。
因此仅缩进或JSON键顺序变化不会改变profile身份,持久字段变化会改变哈希。
保存使用同目录临时文件加原子替换,避免中途退出留下半个JSON。
### 14.3 相关运行文件
| 文件 | 技术作用 |
|---|---|
| glove profile | 原始帧、鲁棒统计、特征参数和捏合锚点 |
| wear-check | 绑定glove profile哈希的短期佩戴凭据 |
| robot profile | 设备身份、姿势、通道曲线、安全范围和仿真端点 |
| checkpoint | 绑定seed和设备元数据的可恢复捕获进度 |
| verification | 绑定robot profile哈希的独立人工复核结果 |
| session manifest | 绑定profile、wear-check、URDF、设备和rosbag话题 |
`session_manifest.py`当前要求G20和O6都已批准,并按 `can0/can1`核对;它适用于标准双手
型号数采拓扑,不是任意单型号或任意CAN配置的通用manifest生成器。
## 15. 离线质量检查
`retarget_profile_check`不启动ROS、不连接机械手,直接回放profile中的原始帧和姿势。
### 15.1 静态复现
对每个共有姿势,只比较该姿势真正定义的相关通道:
- 捏合:拇指和目标手指;
- 拇指姿势:拇指通道;
- 并拢:侧摆通道;
- 桌面/钩拳:普通四指屈伸通道;
- 握拳:除普通侧摆外的通道。
任一相关通道最大误差大于5个u8单位,记为hard failure。
### 15.2 捏合混淆
四个捏合中位数必须:
- 竞争winner等于目标手指;
- 目标门控不小于0.95。
否则记为hard failure。
### 15.3 动态连续性和局部性
每组动态重复记录:
- 目标通道跨度;
- 非目标通道跨度;
- 原始浮点命令帧间步长p95和最大值;
- 取整后整帧不变比例;
- 应用profile执行滤波后的同类指标。
普通手指屈伸轨迹中,若非目标通道跨度中位数大于
`max(15, 0.2 * target_span)`,生成warning。四指开合轨迹中若任一屈伸语义范围中位数
大于0.5,也生成warning。
动态步长目前只报告统计量,没有统一hard-failure阈值;应结合采样率、动作速度和设备
允许步长分析。
离线通过只证明profile内部复现和分解逻辑满足这些判据,不证明实机物理角度精度。
## 16. 实机姿势复核实现
`hand_pose_verify`加载已批准机械手profile后:
1. 查询命令话题是否已有其他发布者,有则拒绝开始;
2. 要求显式输入安全确认;
3. 从最新SDK状态而不是上一次目标开始;
4. 将目标分成每通道步长不超过 `max_step_u8` 的线性序列;
5. 默认30 Hz发送,运动中周期检查新竞争发布者;
6. 稳定后读取命名状态并计算设备空间绝对误差;
7. 保存人工通过/失败、备注、目标、状态和误差摘要。
默认 `max_step_u8=4`,CLI硬限制不超过8。复核报告明确记录
`state_is_angle_ground_truth=false`,并且不修改原机械手profile。
## 17. 扩展和维护约束
### 17.1 增加新的机械手型号
至少需要:
1.`MODEL_COMMAND_LENGTHS`登记型号和命令长度;
2. 定义唯一、稳定的 `command_names`
3. 创建seed profile,包括姿势、安全范围、主动通道、保留通道和仿真端点;
4. 明确每个执行通道的解剖语义源;
5. 扩展profile校验器的必需姿势集合;
6. 扩展节点的话题、身份参数、状态和服务;
7. 增加静态复现、局部性、限位和名称合同测试。
不要通过复制G20的裸下标映射来接入新型号;名称、主动通道和保留通道必须显式定义。
### 17.2 增加新的手套语义
需要同步更新:
- `BASE_INTENT_NAMES`或派生语义列表;
- 标定姿势标签和原始下标;
- glove profile生成及严格校验;
- `HandIntentExtractor.extract()`输出顺序;
- 使用该语义的机械手seed和测试;
- rosbag/下游消费者的数据合同。
修改名称或顺序会影响profile兼容性,应升级schema而不是让旧profile静默通过。
### 17.3 修改手势或阈值
捏合和握持阈值由当前手套profile自动推导。优先修复标定数据或距离定义,不要增加隐藏
的全局常量绕过竞争判据。若确需改变门控公式,应同时更新:
- 正/负样本定义;
- 连续性和混淆测试;
- 离线质量报告;
- `mapping_mode`或schema版本,以便数据可追踪。
### 17.4 线程与实时性
- 串口读取在线程中更新带锁快照;
- ROS节点定时器只消费最新快照,不阻塞等待串口;
- 运行时没有无界命令队列;
- 标定和复核CLI可使用后台executor线程,因为它们包含交互式终端等待;
- 映射主要是小向量NumPy运算,不包含在线优化或模型推理。
## 18. 测试与验收建议
核心单元/集成测试集中在
[`test_linkerforce_v2.py`](../src/linkerhand_retarget/test/test_linkerforce_v2.py),覆盖:
- 鲁棒统计和profile完整性;
- 根部/末端解耦和曲线内部无平台;
- 捏合局部性、连续切换和无历史锁存;
- 普通手指对拇指动作的独立性;
- 动态侧摆串扰补偿;
- 仿真目标与电机残差隔离;
- 滤波步长、加速度、重置和默认直通行为;
- profile身份、checkpoint、安全范围和命名合同;
- 重使能重基准和超时撤权;
- G20/O6输出长度、名称和限位。
修改核心算法后至少执行:
```bash
source /opt/ros/jazzy/setup.bash
source install/setup.bash
python3 -m pytest -q \
src/linkerhand_retarget/test/test_linkerforce_v2.py
```
完成profile标定后再分别执行G20和O6离线质量回放。涉及ROS接口、SDK命名或launch参数的
修改,还应在PREVIEW状态检查实际话题长度、名称、频率和status JSON,再进入低速实机
验收。
## 19. 已知边界
- 当前只支持左手FFG到左手G20/O6;
- `provisional`不提供真实物理关节角精度声明;
- `cmd_u8`和SDK `state_u8`不能转换成可靠的真实关节弧度;
- 仿真目标只代表语义—URDF名义映射,不是实机测量;
- 捏合竞争无时间滞回,连续性依赖当前帧证据质量和可选输入滤波;
- wear-check只在节点加载时验证,不在长时间运行中自动过期撤权;
- 运行时身份门控不核对固件版本;
- 软件急停只撤销发布权限,不替代硬件急停或独立安全控制器;
- 默认双型号launch会同时创建两个SDK驱动;单型号系统可直接启动所需驱动和
`ffg_dual_retarget`节点。
这些边界应保留在数据报告、实验结论和对外精度声明中。
@@ -0,0 +1,731 @@
# FFG多手势标定映射与遥操操作说明
## 1. 文档目的
本文档说明当前 `linkerforce_v2` 无Marker方案的工作原理、标定流程和实机操作方法。
当前主要使用场景是一只左手FFG控制左手G20,也支持在配置对应profile后同时生成O6目标。
当前方案属于 `provisional` 阶段:
- 可以验证手套语义、机械手通道、方向、动作范围和连续性;
- 可以用于演示和临时数采;
- 不能把G20/O6的 `0255` 电机命令当作真实关节角;
- 没有Marker、编码器或独立角度传感器时,不能给出实机与仿真的真实角度误差。
旧入口 `handretarget` 仍然保留;本文档只描述新入口 `ffg_dual_retarget`
## 2. 当前映射架构
```text
FFG左手套21维原始数据
hand_intent:模型无关的人手语义(0~1)
├─→ G20 actuation_target
│ ├─→ G20单机profile → 20维cmd_u8 → G20实机
│ └─→ G20名义URDF范围 → q_nominal → 仿真
└─→ O6 actuation_target
├─→ O6单机profile → 6维cmd_u8 → O6实机
└─→ O6名义URDF范围 → q_nominal → 仿真
```
实机命令和仿真目标是两条独立标尺:
- `cmd_u8`:设备电机空间命令,范围为0~255;
- `q_nominal`:根据URDF名义限位生成的仿真弧度目标;
- `state_u8`:SDK返回的设备状态,只用于运行诊断,不是真实关节角GT。
仿真不应直接把 `cmd_u8` 当作真实角度。需要接近实机外观时,可以使用同一
`actuation_target`,再通过实测角度标定完善仿真标尺。
## 3. 多手势标定解决什么问题
### 3.1 FFG静态姿势
完整手套标定采集11个静态姿势:
1. 五指自然张开、自然分开;
2. 五指伸直并拢;
3. 桌面手势:四指根部弯曲、末端伸直;
4. 钩拳:四指根部伸直、末端弯曲;
5. 自然握拳;
6. 拇指最大外展;
7. 拇指横跨掌心;
8. 拇指—食指捏合;
9. 拇指—中指捏合;
10. 拇指—无名指捏合;
11. 拇指—小指捏合。
每个静态姿势默认采集2秒、重复3次,每次至少50帧。profile保留全部原始帧、
每次中位数、MAD和有效帧数,而不是只保存一个平均值。
### 3.2 FFG动态轨迹
完整标定还采集7组短时往返轨迹:
- 食指独立弯曲往返;
- 中指独立弯曲往返;
- 无名指独立弯曲往返;
- 小指独立弯曲往返;
- 拇指弯曲往返;
- 拇指对掌往返;
- 四指开合往返。
动态轨迹主要用于发现和补偿同一手指屈伸对侧摆语义的传感器串扰,并检查非目标
手指是否跟随运动。它们不是额外的离散手势模板。
### 3.3 连续映射原则
当前运行时不会把整只手吸附到“最相似的标定手势”:
- 每根普通手指只读取自身的根部、末端和侧摆语义;
- 张手、桌面、钩拳和握拳构成四指根部—末端标定面,连续解耦传感器串扰;
- 普通屈伸映射保持连续,不在曲线内部加入停止平台;
- 四种捏合分别进行竞争判断;
- 捏合只局部修正拇指和目标手指,不替换整只手命令;
- 捏合证据不明确时,连续退回普通逐关节映射;
- 握拳只增加必要的拇指协调,不把相似动作强制变成握拳模板。
### 3.4 当前实时执行策略
当前默认执行节奏与旧版左手G20的有效路径一致:
```text
publish_rate=30Hz
input_filter_enabled=false
command_filter_mode=passthrough
command_filter_ema_alpha=1.0
command_filter_max_step_u8=255
command_filter_deadband_u8=0
repeat_position_commands=true
```
也就是直接使用最新手套帧,将连续电机目标交给G20固件插值,并在每个30Hz控制心跳
重复发送最新目标。待发送队列深度为1,来不及发送时只保留最新目标,不重放旧命令。
Kalman和EMA仍然可以显式启用,但不要同时启用两层滤波。两层滤波叠加后再进行整数
取整,容易表现为小幅运动停顿、累计后跳变。
## 4. profile与运行文件
| 文件 | 内容 | 是否提交Git |
|---|---|---|
| `glove_<ID>_left_<operator>.json` | 个人佩戴下的FFG完整标定 | 否 |
| `*.wear_check.json` | 绑定手套profile哈希的快速佩戴检查 | 否 |
| `hand_G20_left_<SN>_provisional.json` | 指定G20实机的姿势命令profile | 否 |
| `*.checkpoint.json` | 实机姿势标定中断续标检查点 | 否 |
| `*.verification.json` | 实机姿势人工复核报告 | 否 |
| `g20_seed_profile.json` | G20 GUI安全初值和标定结构 | 是 |
| `o6_seed_profile.json` | O6 GUI安全初值和标定结构 | 是 |
根目录 `profiles/` 已加入 `.gitignore`。可复现的种子profile位于:
```text
src/linkerhand_retarget/resource/linkerforce_v2/profiles/
```
手套profile与操作者、手套和佩戴方式相关;机械手profile与型号、序列号、固件和CAN
接口相关。换手套、换操作者、明显改变佩戴位置时,应重新完整标定FFG。换机械手本体
或固件导致电机响应变化时,应重新标定机械手profile。
## 5. 构建与环境准备
在每个新终端中都要加载ROS和工作空间:
```bash
cd /home/lxp/projects/linkerhand_retarget_ros2
source /opt/ros/jazzy/setup.bash
source install/setup.bash
```
源码修改后重新构建:
```bash
cd /home/lxp/projects/linkerhand_retarget_ros2
source /opt/ros/jazzy/setup.bash
colcon build --symlink-install --packages-select \
linker_hand_ros2_sdk gui_control linkerhand_retarget
source install/setup.bash
```
如果出现 `Package 'linkerhand_retarget' not found`,通常是当前终端没有执行上述两个
`source`,或者源码修改后还没有构建。
## 6. 完整标定FFG左手套
### 6.1 启动FFG原始数据发布
终端A启动只读FFG节点。只连接左手套即可,不要求右手套存在:
```bash
cd /home/lxp/projects/linkerhand_retarget_ros2
source /opt/ros/jazzy/setup.bash
source install/setup.bash
ros2 run linkerhand_retarget ffg_dual_retarget --ros-args \
-p serial_port:=/dev/ttyUSB0 \
-p baudrate:=2000000 \
-p auto_scan:=true
```
确认原始话题有数据:
```bash
ros2 topic hz /ffg/left/raw_joint_state
```
如果标定提示“0个有效帧”,不要继续重复按Enter。先确认:
- 终端A仍在运行;
- 日志显示左手FFG已连接;
- `/dev/ttyUSB0`没有被另一个FFG进程占用;
- `/ffg/left/raw_joint_state`有稳定数据。
### 6.2 执行完整手套标定
终端B执行:
```bash
cd /home/lxp/projects/linkerhand_retarget_ros2
source /opt/ros/jazzy/setup.bash
source install/setup.bash
ros2 run linkerhand_retarget ffg_calibrate -- \
--glove-id FFG_LEFT_SN \
--operator lxp \
--firmware 2.1.4 \
--output /home/lxp/projects/linkerhand_retarget_ros2/profiles/glove_FFG_LEFT_SN_left_lxp.json
```
静态姿势的正确操作:
1. 先摆好终端提示的固定姿势;
2. 姿势稳定后按Enter
3. 按Enter后继续保持不动约2秒;
4. 终端显示保存帧数后再放松;
5. 同一姿势按相同方法重复3次。
动态往返轨迹的正确操作:
1. 先回到该动作的自然起始位置;
2. 按Enter后立即开始连续、缓慢地往返运动;
3. 在默认3秒采集窗口内完成若干次完整往返;
4. 非目标手指尽量保持稳定;
5. 不要先弯好后全程静止,否则采不到动态关系。
标定成功应显示:
```text
approved_for_runtime=True
sha256=<手套profile哈希>
```
不要使用 `--skip-dynamic` 生成正式运行profile。该参数只适合调试。
## 7. 快速佩戴检查
每次正式实机启动前,对当前准备使用的手套profile执行快速检查:
```bash
ros2 run linkerhand_retarget ffg_calibrate -- \
--quick-check /home/lxp/projects/linkerhand_retarget_ros2/profiles/glove_FFG_LEFT_SN_left_lxp.json \
--quick-output /home/lxp/projects/linkerhand_retarget_ros2/profiles/glove_FFG_LEFT_SN_left_lxp.wear_check.json
```
依次检查张手、握拳和食指捏合。每个姿势也是“先摆好,再按Enter,然后保持2秒”。
快速检查凭据:
- 默认12小时有效;
- 必须显示 `passed=true`
- 必须与启动时使用的手套profile SHA-256完全一致;
- 切换v4、v5等手套profile时,必须同时切换到对应的wear-check文件。
快速检查失败时,先重新调整手套佩戴位置并重试。如果多次失败,说明当前佩戴与完整
标定差异过大,应重新完整标定,不要通过增大阈值静默放行实机。
## 8. 标定G20实机姿势
### 8.1 安全要求
- 标定时使用低速、低力矩;
- 配备软件急停,并保证机械手周围无障碍物;
- 不得使用旧手套映射把机械手带到标定姿势;
- 不得在带电状态强行手掰;
- 使用GUI逐通道调整,并确认通道方向正确。
### 8.2 启动G20 SDK
终端A
```bash
cd /home/lxp/projects/linkerhand_retarget_ros2
source /opt/ros/jazzy/setup.bash
source install/setup.bash
ros2 run linker_hand_ros2_sdk linker_hand_sdk --ros-args \
-p hand_type:=left \
-p hand_joint:=G20 \
-p is_touch:=false \
-p can:=can0 \
-p modbus:=None \
-p topic_prefix:=/g20 \
-p startup_speed:=30 \
-p startup_torque:=80 \
-p move_on_startup:=false \
-p state_poll_rate:=10.0
```
### 8.3 启动G20标定GUI
终端B
```bash
cd /home/lxp/projects/linkerhand_retarget_ros2
source /opt/ros/jazzy/setup.bash
source install/setup.bash
ros2 run gui_control gui_control --ros-args \
-r __node:=g20_calibration_gui \
-p hand_type:=left \
-p hand_joint:=G20 \
-p topic_prefix:=/g20
```
### 8.4 捕获11个G20姿势
终端C
```bash
cd /home/lxp/projects/linkerhand_retarget_ros2
source /opt/ros/jazzy/setup.bash
source install/setup.bash
ros2 run linkerhand_retarget hand_pose_capture -- \
--model G20 \
--seed /home/lxp/projects/linkerhand_retarget_ros2/install/linkerhand_retarget/share/linkerhand_retarget/linkerforce_v2/profiles/g20_seed_profile.json \
--serial-number G20_LEFT_001 \
--operator lxp \
--firmware unknown \
--can-interface can0 \
--output /home/lxp/projects/linkerhand_retarget_ros2/profiles/hand_G20_left_G20_LEFT_001_provisional.json
```
每个姿势的操作:
1. 用GUI低速调整机械手;
2. 目视确认目标手指、通道方向和最终姿势;
3. 等待实机稳定;
4. 点击GUI“保存当前标定姿势”;
5. CLI收到快照后选择姿势状态。
状态含义:
- `exact`:机械手能够准确实现该姿势;
- `approximate`:受机构自由度限制,只能实现最佳近似;
- `unsupported`:该型号不能可靠实现,不用于对应姿势约束;
- 输入 `r`:放弃刚才的快照,重新调整和保存。
CLI每完成一个姿势都会立即写入 `*.checkpoint.json`。程序中断后,使用完全相同的
命令会自动恢复并跳过已保存姿势,不需要从头开始。
只有确实要放弃原进度时才增加:
```text
--fresh
```
全部姿势完成后,只有输入 `y` 批准,最终profile才会包含:
```text
approved_for_control=True
```
O6操作相同,但使用 `--model O6`、O6 seed、`can1``/o6` 命名空间。O6自由度较少,
桌面、钩拳及部分捏合通常应标为 `approximate`
## 9. 复核机械手profile
姿势复核可以避免手工把JSON中的20维命令复制到GUI。
复核前:
- 停止 `ffg_dual_retarget`
- 停止GUI,避免命令话题存在其他发布者;
- 只保留低速、低力矩的G20 SDK。
执行:
```bash
ros2 run linkerhand_retarget hand_pose_verify -- \
--profile /home/lxp/projects/linkerhand_retarget_ros2/profiles/hand_G20_left_G20_LEFT_001_provisional.json \
--operator lxp \
--topic-prefix /g20
```
只复核单个姿势:
```bash
ros2 run linkerhand_retarget hand_pose_verify -- \
--profile /home/lxp/projects/linkerhand_retarget_ros2/profiles/hand_G20_left_G20_LEFT_001_provisional.json \
--operator lxp \
--topic-prefix /g20 \
--pose pinch_index
```
按照提示输入 `VERIFY``MOVE`,再选择:
- `p`:目视通过;
- `f`:目视未通过;
- `r`:重放;
- `s`:跳过。
工具会低速平滑过渡,并保存独立的 `*.verification.json`,不会修改原始机械手profile。
## 10. 离线检查映射质量
不连接实机即可回放profile中的静态与动态数据:
```bash
ros2 run linkerhand_retarget retarget_profile_check -- \
--glove /home/lxp/projects/linkerhand_retarget_ros2/profiles/glove_FFG_LEFT_SN_left_lxp.json \
--robot /home/lxp/projects/linkerhand_retarget_ros2/profiles/hand_G20_left_G20_LEFT_001_provisional.json \
--model G20 \
--output /home/lxp/projects/linkerhand_retarget_ros2/profiles/g20_mapping_quality.json
```
重点查看:
- `passed``hard_failures`
- 静态有效通道复现误差;
- 四种捏合的winner和gate
- 四指动态轨迹的非目标通道跨度;
- 小指、侧摆等动作是否有明显串扰;
- 帧间命令变化是否存在异常突跳。
离线检查通过不等于实机角度准确,只表示profile内部逻辑一致。
## 11. 启动G20正式遥操
正式启动前,停止旧SDK、标定GUI、姿势捕获工具和占用FFG串口的只读节点。每种节点
只保留一个实例。
### 11.1 终端A:启动G20驱动
```bash
cd /home/lxp/projects/linkerhand_retarget_ros2
source /opt/ros/jazzy/setup.bash
source install/setup.bash
ros2 run linker_hand_ros2_sdk linker_hand_sdk --ros-args \
-p hand_type:=left \
-p hand_joint:=G20 \
-p is_touch:=false \
-p can:=can0 \
-p modbus:=None \
-p topic_prefix:=/g20 \
-p startup_speed:=255 \
-p startup_torque:=80 \
-p move_on_startup:=false \
-p state_poll_rate:=10.0 \
-p velocity_poll_rate:=10.0 \
-p defer_state_reads_while_commanding:=true \
-p repeat_position_commands:=true
```
`startup_speed`控制电机最大运动速度;`startup_torque`控制最大输出力矩。提高力矩不会
解决映射卡顿。建议先使用80,在确有负载需要并完成安全评估后再提高。
确认状态话题已有发布者:
```bash
ros2 topic info /g20/cb_left_hand_state
```
应至少显示:
```text
Publisher count: 1
```
### 11.2 终端B:启动FFG映射节点
以下示例使用当前v4手套profile:
```bash
cd /home/lxp/projects/linkerhand_retarget_ros2
source /opt/ros/jazzy/setup.bash
source install/setup.bash
ros2 run linkerhand_retarget ffg_dual_retarget --ros-args \
-p serial_port:=/dev/ttyUSB0 \
-p baudrate:=2000000 \
-p auto_scan:=true \
-p publish_rate:=30.0 \
-p input_filter_enabled:=false \
-p command_filter_mode:=passthrough \
-p command_filter_ema_alpha:=1.0 \
-p command_filter_max_step_u8:=255.0 \
-p command_filter_deadband_u8:=0.0 \
-p glove_profile:=/home/lxp/projects/linkerhand_retarget_ros2/profiles/glove_FFG_LEFT_SN_left_lxp_v4.json \
-p wear_check:=/home/lxp/projects/linkerhand_retarget_ros2/profiles/glove_FFG_LEFT_SN_left_lxp_v4.wear_check.json \
-p g20_profile:=/home/lxp/projects/linkerhand_retarget_ros2/profiles/hand_G20_left_G20_LEFT_001_provisional.json \
-p g20_serial_number:=G20_LEFT_001 \
-p g20_can_interface:=can0
```
正常日志应包含:
```text
FFG profile已加载
三姿势快速佩戴检查有效
G20 profile已加载(可申请实机使能)
G20映射=factorized_paired_v2
执行滤波=passthrough alpha=1.0, max_step=255.0
左手FFG已连接
```
节点启动后默认处于PREVIEW,不会立即控制实机。
### 11.3 PREVIEW检查
在使能前观察:
```text
/ffg/left/raw_joint_state
/ffg/left/filtered_joint_state
/retarget/left/hand_intent
/retarget/g20/left/actuation_target
/retarget/g20/left/joint_target_nominal
/retarget/g20/left/cmd_u8_preview
```
检查要求:
- 张手、半握、握拳过程中命令连续;
- 弯曲一根手指时,主要变化的是对应手指通道;
- 食指捏合主要影响拇指与食指;
- 不应锁定在某个历史捏合模板;
- 不应出现明显越限或突跳。
### 11.4 使能G20
终端C
```bash
cd /home/lxp/projects/linkerhand_retarget_ros2
source /opt/ros/jazzy/setup.bash
source install/setup.bash
ros2 service call /ffg_dual_retarget/enable_g20 \
std_srvs/srv/SetBool "{data: true}"
```
成功返回:
```text
success=True
message='G20已使能'
```
停用G20
```bash
ros2 service call /ffg_dual_retarget/enable_g20 \
std_srvs/srv/SetBool "{data: false}"
```
软件急停:
```bash
ros2 service call /ffg_dual_retarget/emergency_stop \
std_srvs/srv/Trigger "{}"
```
急停、FFG断开、驱动状态超时或profile错误后,都需要排除问题并重新显式使能。
## 12. ROS话题说明
| 话题 | 说明 |
|---|---|
| `/ffg/left/raw_joint_state` | 21维FFG原始数据 |
| `/ffg/left/filtered_joint_state` | 实际送入语义提取器的数据;默认与raw相同 |
| `/retarget/left/hand_intent` | 模型无关的01人手语义 |
| `/retarget/g20/left/actuation_target` | G20归一化目标 |
| `/retarget/g20/left/joint_target_nominal` | G20仿真名义弧度目标 |
| `/retarget/g20/left/cmd_u8_preview` | 未使能时也持续发布的20维预览命令 |
| `/g20/cb_left_hand_control_cmd` | 使能后发送给G20 SDK的20维命令 |
| `/g20/cb_left_hand_state` | G20 SDK状态心跳 |
| `/ffg_dual_retarget/status` | profile、使能、超时、滤波和延迟诊断 |
所有向量都带名称。仿真桥和其他消费者必须按 `JointState.name` 匹配,不得依赖裸下标。
## 13. 安全与自动停用
- 默认PREVIEW,必须通过服务显式使能;
- FFG超过0.35秒没有新帧:撤销全部实机使能;
- 对应驱动状态超过1秒未更新:只撤销该型号;
- profile缺失、未批准、SN/CAN不匹配:拒绝实机使能;
- wear-check缺失、失败、过期或哈希不匹配:拒绝实机使能;
- 所有命令检查长度、名称、有限值和0~255范围;
- G20四个保留通道保持安全固定值;
- 带电机械手不得强行手掰。
## 14. 常见问题排查
### 14.1 `Package 'linkerhand_retarget' not found`
```bash
source /opt/ros/jazzy/setup.bash
source /home/lxp/projects/linkerhand_retarget_ros2/install/setup.bash
```
如果仍然找不到,重新执行第5节的构建命令。
### 14.2 手套标定只有0个有效帧
原因通常是没有单独启动FFG原始数据发布节点。检查:
```bash
ros2 topic info /ffg/left/raw_joint_state
ros2 topic hz /ffg/left/raw_joint_state
```
### 14.3 快速佩戴检查失败
- 确认使用的是正确版本profile;
- 调整手套位置、腕带和手指传感器;
- 每个姿势先摆好再按Enter
- 按Enter后保持不动2秒;
- 多次失败则重新完整标定。
### 14.4 `三姿势快速佩戴检查缺失、失败或过期`
重新对启动时使用的同一个手套profile执行第7节命令。不能拿v5的wear-check启动v4。
### 14.5 `G20驱动状态无效或已超时`
先执行:
```bash
ros2 topic info /g20/cb_left_hand_state
```
如果 `Publisher count: 0`,说明G20 SDK未启动或没有使用 `/g20` 命名空间。按第11.1节
启动驱动。如果有发布者,再检查:
```bash
ros2 topic echo /g20/cb_left_hand_state --once
```
状态必须是20维、名称与G20 profile一致、数值有限且位于0255。
### 14.6 服务一直显示 `waiting for service`
检查:
```bash
ros2 node list
ros2 service list | grep ffg_dual_retarget
```
常见原因是 `ffg_dual_retarget`没有启动、当前终端未source,或者服务名称中多写了反斜杠。
### 14.7 服务成功但机械手不动
检查命令话题:
```bash
ros2 topic info /g20/cb_left_hand_control_cmd
ros2 topic hz /g20/cb_left_hand_control_cmd
```
使能后应同时存在发布者和订阅者,并接近30Hz。还要检查SDK终端是否报告CAN错误。
### 14.8 机械手运动卡顿
确认运行参数:
```bash
ros2 param get /ffg_dual_retarget input_filter_enabled
ros2 param get /ffg_dual_retarget command_filter_mode
ros2 param get /linker_hand_sdk repeat_position_commands
```
当前推荐结果:
```text
False
passthrough
True
```
同时检查:
- 只运行一个G20 SDK和一个映射节点;
- 命令话题稳定接近30Hz
- CAN状态查询在遥操期间已延后;
- 不要用提高力矩解决卡顿;
- 如果只有某一根手指异常,运行离线profile质量检查,重点看该手指动态轨迹。
### 14.9 某根手指张手不到位或发生串指
依次比较:
```text
raw_joint_state
→ hand_intent
→ cmd_u8_preview
→ cb_left_hand_state
```
- raw异常:佩戴或FFG采集问题;
- hand_intent异常:手套标定/语义解耦问题;
- intent正确但preview异常:映射/profile问题;
- preview正确但实机异常:机械手profile、驱动、固件或机构问题。
不要直接通过修改某个写死系数掩盖问题。
### 14.10 实机姿势标定中断
使用完全相同的 `hand_pose_capture` 命令重新启动,会自动读取检查点并跳过已保存姿势。
不要增加 `--fresh`,除非明确要删除当前标定进度并从头开始。
## 15. 当前精度边界与后续优化
当前多手势方案可以继续优化:
- 重采质量较差的小指、侧摆或拇指动态轨迹;
- 改善21维传感器到人体语义的连续解耦;
- 完善拇指对掌和四种捏合的局部连续映射;
- 为每台机械手建立方向相关、非线性的电机命令曲线;
- 记录输入、映射、发布、CAN和状态时间戳,量化延迟与丢帧。
要得到可量化的真实角度精度,仍需增加Marker、编码器或独立角度传感器,建立:
```text
hand_intent
→ 实机真实关节角q_target
→ 单机关节角—cmd_u8标定
```
在此之前,验收结论只能是动作语义、通道、连续性和外观接近程度,不能声明实机与仿真
达到某个真实关节角误差。
## 16. 正式运行前检查清单
- [ ] ROS与工作空间已source
- [ ] 当前只有一个FFG读取进程;
- [ ] 当前只有一个G20 SDK,使用`can0``/g20`
- [ ] FFG profile显示`approved_for_runtime=True`
- [ ] wear-check通过、未过期且哈希匹配;
- [ ] G20 profile显示`approved_for_control=True`
- [ ] profile中的SN和CAN与启动参数一致;
- [ ] `/g20/cb_left_hand_state`有有效发布者;
- [ ] PREVIEW下逐指、握拳和四种捏合动作正确;
- [ ] 默认实时参数为关闭输入滤波、直通命令、30Hz重复目标;
- [ ] 周围安全、急停可用;
- [ ] 最后才调用`enable_g20`
Submodule src/agillink_omnihand_sdk added at 026740d9fd
+79 -11
View File
@@ -9,6 +9,20 @@ class HandConfig:
joint_names_en: Optional[List[str]] = None
init_pos: List[int] = field(default_factory=list)
preset_actions: Optional[Dict[str, List[int]]] = None
preset_action_overrides: Optional[Dict[str, Dict[str, List[int]]]] = None
# Integer GUI values are divided by this scale before publishing.
# Legacy hands use raw u8 values (scale=1); O12 uses milliradians.
position_scale: int = 1
position_unit: str = "u8"
def get_preset_actions(self, hand_type: str) -> Dict[str, List[int]]:
"""Return preset actions with hand-specific values applied."""
actions = dict(self.preset_actions or {})
if self.preset_action_overrides:
actions.update(
self.preset_action_overrides.get(hand_type.lower(), {})
)
return actions
# ------------------------------------------------------------------
# 常量字典(仅构建一次)
@@ -75,10 +89,10 @@ _HAND_CONFIGS: Dict[str, HandConfig] = {
"点赞": [255, 0, 0, 0, 0, 255, 162, 162, 144, 100, 210, 255, 255, 255, 255, 255, 0, 0, 0, 0],
"握拳": [96, 0, 0, 0, 0, 0, 193, 158, 128, 91, 132, 255, 255, 255, 255, 144, 0, 0, 0, 0],
"张开": [255, 255, 255, 255, 255, 255, 193, 148, 105, 42, 245, 255, 255, 255, 255, 255, 255, 255, 255, 255],
"OK": [148, 110, 255, 255, 255, 44, 164, 100, 114, 127, 178, 255, 255, 255, 255, 94, 71, 255, 255, 255],
"拇指对中指": [191, 255, 55, 255, 255, 96, 95, 100, 114, 127, 105, 255, 255, 255, 255, 94, 255, 108, 255, 255],
"拇指对无名指": [191, 255, 255, 72, 255, 115, 95, 100, 114, 127, 60, 255, 255, 255, 255, 94, 255, 255, 97, 255],
"拇指对小指": [191, 255, 255, 255, 55, 0, 95, 100, 114, 121, 70, 255, 255, 255, 255, 94, 255, 255, 255, 100],
"OK": [0, 0, 255, 255, 255, 138, 147, 148, 105, 42, 109, 255, 255, 255, 255, 255, 211, 255, 255, 255],
"拇指对中指": [0, 255, 0, 255, 255, 107, 149, 148, 105, 42, 109, 255, 255, 255, 255, 255, 225, 202, 255, 255],
"拇指对无名指": [0, 255, 255, 0, 255, 88, 171, 148, 105, 42, 59, 255, 255, 255, 255, 255, 255, 255, 206, 254],
"拇指对小指": [0, 255, 255, 255, 0, 32, 170, 148, 105, 42, 109, 255, 255, 255, 255, 255, 255, 255, 255, 203],
"准备1": [255, 0, 0, 0, 0, 255, 162, 162, 144, 100, 210, 255, 255, 255, 255, 255, 0, 0, 0, 0],
"": [96, 255, 0, 0, 0, 0, 190, 161, 127, 80, 68, 255, 255, 255, 255, 144, 255, 0, 0, 0],
"": [96, 255, 255, 0, 0, 0, 190, 66, 127, 80, 68, 255, 255, 255, 255, 144, 255, 255, 0, 0],
@@ -100,8 +114,6 @@ _HAND_CONFIGS: Dict[str, HandConfig] = {
"动作9": [255, 255, 255, 255, 255, 255, 125, 129, 125, 130, 210, 255, 255, 255, 255, 255, 255, 255, 255, 255],
"根部1": [0, 0, 0, 0, 0, 255, 125, 129, 125, 130, 245, 255, 255, 255, 255, 255, 255, 255, 255, 255],
"根部2": [255, 255, 255, 255, 255, 255, 125, 129, 125, 130, 245, 255, 255, 255, 255, 255, 255, 255, 255, 255],
"根部1": [0, 0, 0, 0, 0, 255, 125, 129, 125, 130, 245, 255, 255, 255, 255, 255, 255, 255, 255, 255],
"根部2": [255, 255, 255, 255, 255, 255, 125, 129, 125, 130, 245, 255, 255, 255, 255, 255, 255, 255, 255, 255],
"末端1": [6, 0, 0, 0, 0, 255, 125, 129, 125, 130, 219, 255, 255, 255, 255, 125, 0, 0, 0, 0],
"末端2": [6, 0, 0, 0, 0, 255, 125, 129, 125, 130, 219, 255, 255, 255, 255, 255, 255, 255, 255, 255],
"末端3": [6, 0, 0, 0, 0, 255, 125, 129, 125, 130, 219, 255, 255, 255, 255, 125, 0, 0, 0, 0],
@@ -224,13 +236,20 @@ _HAND_CONFIGS: Dict[str, HandConfig] = {
"": [92, 87, 255, 255, 255, 0],
"": [92, 87, 255, 255, 255, 255],
"": [255, 255, 255, 255, 255, 255],
"OK": [139, 91, 103, 250, 250, 250],
"OK": [95, 75, 116, 255, 255, 255],
"拇指对中指": [88, 2, 255, 114, 255, 255],
"点赞": [250, 79, 0, 0, 0, 0],
"握拳": [102, 18, 0, 0, 0, 0],
}
},
preset_action_overrides={
"right": {
"OK": [95, 83, 122, 255, 255, 255],
"拇指对中指": [95, 8, 255, 114, 255, 255],
},
},
),
"L6": HandConfig(
joint_names_en=["thumb_cmc_pitch", "thumb_cmc_yaw", "index_mcp_pitch", "middle_mcp_pitch", "pinky_mcp_pitch", "ring_mcp_pitch"],
joint_names_en=["thumb_cmc_pitch", "thumb_cmc_roll", "index_mcp_pitch", "middle_mcp_pitch", "ring_mcp_pitch", "pinky_mcp_pitch"],
joint_names=["大拇指弯曲", "大拇指横摆", "食指弯曲", "中指弯曲", "无名指弯曲", "小拇指弯曲"],
init_pos=[250] * 6,
preset_actions={
@@ -240,7 +259,7 @@ _HAND_CONFIGS: Dict[str, HandConfig] = {
"": [0, 39, 255, 255, 255, 0],
"": [0, 0, 255, 255, 255, 255],
"": [255, 255, 255, 255, 255, 255],
"OK": [74, 13, 153, 255, 255, 255],
"OK": [62, 5, 151, 255, 255, 255],
"点赞": [255, 255, 0, 0, 0, 0],
"握拳": [79, 11, 0, 0, 0, 0],
"序列动作1": [250, 250, 250, 250, 250, 250],
@@ -256,7 +275,56 @@ _HAND_CONFIGS: Dict[str, HandConfig] = {
"拇指压感准备1": [139, 18, 130, 0, 0, 0],
"拇指压感测试": [39, 30, 122, 250, 250, 250],
"拇指压感准备2": [139, 18, 130, 0, 0, 0]
}
},
preset_action_overrides={
"right": {
"OK": [58, 6, 153, 255, 255, 255],
},
},
),
"O12": HandConfig(
# O12 active-angle API order. Keep this independent from the
# physical motor order exposed by some SDK metadata.
joint_names_en=[
"thumb_roll", "thumb_abad", "thumb_mcp", "thumb_pip",
"index_abad", "index_mcp", "index_pip",
"middle_abad", "middle_mcp", "middle_pip",
"ring_mcp", "pinky_mcp",
],
joint_names=[
"拇指旋转", "拇指侧摆", "拇指根部", "拇指中部",
"食指侧摆", "食指根部", "食指中部",
"中指侧摆", "中指根部", "中指中部",
"无名指根部", "小指根部",
],
# Values are milliradians so QSlider can retain 0.001 rad resolution.
init_pos=[0] * 12,
preset_actions={
"展开": [0] * 12,
"食指轻弯": [0, 0, 0, 0, 0, 300, 400, 0, 0, 0, 0, 0],
"半握": [
300, -200, -200, -500,
0, 650, 700, 0, 650, 800, 700, 700,
],
},
preset_action_overrides={
"right": {
"拇指对食指": [
245, -766, -818, 0, 0, 1046, 0, 0, 0, 0, 0, 0,
],
"拇指对中指": [
489, -843, -685, 0, 0, 0, 0, 0, 1147, 0, 0, 0,
],
"拇指对无名指": [
699, -779, -827, 0, 0, 0, 0, 0, 0, 0, 690, 0,
],
"拇指对小指": [
786, -1065, -827, 0, 0, 0, 0, 0, 0, 0, 0, 645,
],
},
},
position_scale=1000,
position_unit="rad",
),
}
HAND_CONFIGS = MappingProxyType(_HAND_CONFIGS)
+273 -33
View File
@@ -1,5 +1,6 @@
import sys
import time, json
import math
import threading
from dataclasses import dataclass
from typing import List, Dict
@@ -19,14 +20,72 @@ from .utils.mapping import *
from .config.constants import _HAND_CONFIGS
LOOP_TIME = 1000 # 循环动作间隔时间 毫秒
_CANONICAL_COMMAND_NAMES = {
"G20": [
"thumb_cmc_pitch", "index_mcp_pitch", "middle_mcp_pitch",
"ring_mcp_pitch", "pinky_mcp_pitch", "thumb_cmc_roll",
"index_mcp_roll", "middle_mcp_roll", "ring_mcp_roll",
"pinky_mcp_roll", "thumb_cmc_yaw", "reserved_11",
"reserved_12", "reserved_13", "reserved_14", "thumb_mcp",
"index_pip", "middle_pip", "ring_pip", "pinky_pip",
],
"O6": [
"thumb_cmc_pitch", "thumb_cmc_yaw", "index_mcp_pitch",
"middle_mcp_pitch", "ring_mcp_pitch", "pinky_mcp_pitch",
],
"L6": [
"thumb_cmc_pitch", "thumb_cmc_roll", "index_mcp_pitch",
"middle_mcp_pitch", "ring_mcp_pitch", "pinky_mcp_pitch",
],
"O12": [
"thumb_roll", "thumb_abad", "thumb_mcp", "thumb_pip",
"index_abad", "index_mcp", "index_pip",
"middle_abad", "middle_mcp", "middle_pip",
"ring_mcp", "pinky_mcp",
],
}
_CANONICAL_COMMAND_BOUNDS = {
"G20": [
*[(0, 255)] * 10,
(0, 255),
*[(255, 255)] * 4,
*[(0, 255)] * 5,
],
"O6": [(0, 255)] * 6,
"L6": [(0, 255)] * 6,
}
# Integer slider bounds in milliradians, derived from the O12 active-angle
# limits. Right and left thumb signs differ; all other limits are shared.
_O12_COMMAND_BOUNDS_MRAD = {
"right": [
(0, 942), (-1387, 0), (-827, 0), (-1291, 0),
(-261, 261), (0, 1352), (0, 1530), (-261, 261),
(0, 1357), (0, 1815), (0, 1535), (0, 1535),
],
"left": [
(-942, 0), (0, 1387), (-827, 0), (-1291, 0),
(-261, 261), (0, 1352), (0, 1530), (-261, 261),
(0, 1357), (0, 1815), (0, 1535), (0, 1535),
],
}
class ROS2NodeManager(QObject):
"""ROS2节点管理器,处理ROS通信"""
status_updated = pyqtSignal(str, str) # 状态类型, 消息内容
feedback_updated = pyqtSignal(object)
def __init__(self, node_name: str = "hand_control_node"):
super().__init__()
self.node = None
self.publisher = None
self.publisher_arc = None
self.state_subscription = None
self.speed_pub = None
self.torque_pub = None
self.joint_state = JointState()
self.joint_state.header = Header()
@@ -45,28 +104,56 @@ class ROS2NodeManager(QObject):
self.node.declare_parameter('hand_joint', 'L10')
self.node.declare_parameter('topic_hz', 30)
self.node.declare_parameter('is_arc', False)
self.node.declare_parameter('topic_prefix', '')
# 获取参数
self.hand_type = self.node.get_parameter('hand_type').value
self.hand_joint = self.node.get_parameter('hand_joint').value
self.hz = self.node.get_parameter('topic_hz').value
self.is_arc = self.node.get_parameter('is_arc').value
self.topic_prefix = self.normalize_topic_prefix(
self.node.get_parameter('topic_prefix').value
)
if self.is_arc == True:
if self.hand_joint == "O12":
topic_base = f'/o12/{self.hand_type}'
self.publisher = self.node.create_publisher(
JointState, f'{topic_base}/joint_cmd', 10
)
self.state_subscription = self.node.create_subscription(
JointState,
f'{topic_base}/joint_states',
self._o12_feedback_callback,
10,
)
elif self.is_arc == True:
# 创建发布者
self.publisher_arc = self.node.create_publisher(
JointState, f'/cb_{self.hand_type}_hand_control_cmd_arc', 10
JointState,
self.topic(f'/cb_{self.hand_type}_hand_control_cmd_arc'),
10
)
# 创建发布者
self.publisher = self.node.create_publisher(
JointState, f'/cb_{self.hand_type}_hand_control_cmd', 10
# 创建旧型号发布者
if self.hand_joint != "O12":
self.publisher = self.node.create_publisher(
JointState,
self.topic(f'/cb_{self.hand_type}_hand_control_cmd'),
10
)
self.snapshot_publisher = self.node.create_publisher(
JointState, self.topic('/calibration_pose_snapshot'), 10
)
# 新增 speed / torque 发布者
self.speed_pub = self.node.create_publisher(
String, f'/cb_hand_setting_cmd', 10)
self.torque_pub = self.node.create_publisher(
String, f'/cb_hand_setting_cmd', 10)
self.status_updated.emit("info", f"ROS2节点初始化成功: {self.hand_type} {self.hand_joint}")
if self.hand_joint != "O12":
self.speed_pub = self.node.create_publisher(
String, self.topic('/cb_hand_setting_cmd'), 10)
self.torque_pub = self.node.create_publisher(
String, self.topic('/cb_hand_setting_cmd'), 10)
self.status_updated.emit(
"info",
f"ROS2节点初始化成功: {self.topic_prefix or '/'} "
f"{self.hand_type} {self.hand_joint}"
)
# 启动ROS2自旋线程
self.spin_thread = threading.Thread(target=self.spin_node, daemon=True)
@@ -75,11 +162,55 @@ class ROS2NodeManager(QObject):
self.status_updated.emit("error", f"ROS2初始化失败: {str(e)}")
raise
@staticmethod
def normalize_topic_prefix(prefix: str) -> str:
prefix = str(prefix).strip()
if not prefix or prefix == '/':
return ''
if not prefix.startswith('/'):
prefix = '/' + prefix
return prefix.rstrip('/')
def topic(self, absolute_topic: str) -> str:
if not absolute_topic.startswith('/'):
raise ValueError('base topic must be absolute')
return self.topic_prefix + absolute_topic
def spin_node(self):
"""运行ROS2节点自旋循环"""
while rclpy.ok() and self.node:
rclpy.spin_once(self.node, timeout_sec=0.1)
def _o12_feedback_callback(self, msg: JointState):
"""Forward 12-channel O12 radian feedback safely into the Qt thread."""
if len(msg.position) != 12:
self.status_updated.emit(
"error", f"O12位置反馈长度错误: {len(msg.position)},应为12"
)
return
self.feedback_updated.emit([float(value) for value in msg.position])
def command_bounds(self, count: int):
if self.hand_joint == "O12":
bounds = _O12_COMMAND_BOUNDS_MRAD.get(self.hand_type)
else:
bounds = _CANONICAL_COMMAND_BOUNDS.get(self.hand_joint)
if not bounds or len(bounds) != count:
return [(0, 255)] * count
return bounds
def bound_positions(self, positions: List[int]) -> List[int]:
bounds = self.command_bounds(len(positions))
return [
max(minimum, min(maximum, int(value)))
for value, (minimum, maximum) in zip(positions, bounds)
]
def wire_positions(self, positions: List[int]) -> List[float]:
positions = self.bound_positions(positions)
scale = _HAND_CONFIGS[self.hand_joint].position_scale
return [float(value) / float(scale) for value in positions]
def publish_joint_state(self, positions: List[int]):
"""发布关节状态消息"""
if not self.publisher or not self.node:
@@ -87,21 +218,31 @@ class ROS2NodeManager(QObject):
return
try:
positions = self.bound_positions(positions)
wire_positions = self.wire_positions(positions)
self.joint_state.header.stamp = self.node.get_clock().now().to_msg()
self.joint_state.position = [float(pos) for pos in positions]
self.joint_state.position = wire_positions
# self.joint_state.velocity = [0.1] * len(positions)
# self.joint_state.effort = [0.01] * len(positions)
# 如果有关节名称,添加到消息中
#hand_config = HandConfig.from_hand_type(self.hand_joint)
hand_config = _HAND_CONFIGS[self.hand_joint]
if len(hand_config.joint_names) == len(positions):
canonical_names = _CANONICAL_COMMAND_NAMES.get(self.hand_joint)
# O12 SDK 1.1.8 may expose names in physical motor order while the
# position array uses active-angle order. An empty name array makes
# the documented position order unambiguous to the driver.
if self.hand_joint == "O12":
self.joint_state.name = []
elif canonical_names and len(canonical_names) == len(positions):
self.joint_state.name = canonical_names
elif len(hand_config.joint_names) == len(positions):
if hand_config.joint_names_en != None:
self.joint_state.name = hand_config.joint_names_en
else:
self.joint_state.name = hand_config.joint_names
self.publisher.publish(self.joint_state)
if self.is_arc == True:
if self.is_arc == True and self.hand_joint != "O12":
if self.hand_joint == "O6":
if self.hand_type == "left":
pose = range_to_arc_left(positions,self.hand_joint)
@@ -131,9 +272,27 @@ class ROS2NodeManager(QObject):
except Exception as e:
self.status_updated.emit("error", f"发布失败: {str(e)}")
def publish_pose_snapshot(self, positions: List[int]):
"""发布带名称的当前标定姿势快照。"""
positions = self.bound_positions(positions)
self.publish_joint_state(positions)
self.joint_state.header.stamp = self.node.get_clock().now().to_msg()
self.joint_state.position = self.wire_positions(positions)
canonical_names = _CANONICAL_COMMAND_NAMES.get(self.hand_joint)
if canonical_names and self.hand_joint != "O12":
self.joint_state.name = canonical_names
elif self.hand_joint == "O12":
self.joint_state.name = []
self.snapshot_publisher.publish(self.joint_state)
self.status_updated.emit("info", "当前标定姿势快照已发布")
def publish_speed(self, val: int):
joint_len = 0
if (self.hand_joint.upper() == "O6" or self.hand_joint.upper() == "L6"):
if self.hand_joint == "O12":
self.status_updated.emit(
"warning", "O12不使用旧版u8速度接口;请通过角度轨迹限制速度"
)
return
if self.hand_joint.upper() in ("O6", "L6"):
joint_len = 6
elif self.hand_joint == "L7":
joint_len = 7
@@ -152,8 +311,12 @@ class ROS2NodeManager(QObject):
self.speed_pub.publish(msg)
def publish_torque(self, val: int):
joint_len = 0
if (self.hand_joint.upper() == "O6" or self.hand_joint.upper() == "L6"):
if self.hand_joint == "O12":
self.status_updated.emit(
"warning", "O12不使用旧版u8扭矩接口;位置GUI已禁用该设置"
)
return
if self.hand_joint.upper() in ("O6", "L6"):
joint_len = 6
elif self.hand_joint == "L7":
joint_len = 7
@@ -194,11 +357,17 @@ class HandControlGUI(QWidget):
# 设置ROS管理器
self.ros_manager = ros_manager
self.ros_manager.status_updated.connect(self.update_status)
self.ros_manager.feedback_updated.connect(self.on_feedback_updated)
# 获取手部配置
self.hand_joint = self.ros_manager.hand_joint
self.hand_type = self.ros_manager.hand_type
self.hand_config = _HAND_CONFIGS[self.hand_joint]
self.preset_actions = self.hand_config.get_preset_actions(self.hand_type)
self.feedback_positions = None
# O12 only publishes after an explicit slider/preset action. This
# prevents opening or closing the real hand merely by launching GUI.
self.command_dirty = self.hand_joint != "O12"
# 初始化UI
self.init_ui()
@@ -209,10 +378,27 @@ class HandControlGUI(QWidget):
self.publish_timer.timeout.connect(self.publish_joint_state)
self.publish_timer.start()
def format_position(self, value: int) -> str:
"""Format an integer GUI value in the hand's actual command unit."""
scale = self.hand_config.position_scale
if self.hand_config.position_unit == "rad":
radians = float(value) / float(scale)
return f"{radians:.3f} rad / {math.degrees(radians):.1f}°"
return str(value)
def command_positions(self) -> List[float]:
"""Return current slider values converted to wire units."""
return self.ros_manager.wire_positions(
[slider.value() for slider in self.sliders]
)
def init_ui(self):
"""初始化用户界面"""
# 设置窗口属性
self.setWindowTitle(f'灵巧手控制界面 - {self.hand_type} {self.hand_joint}')
self.setWindowTitle(
f'灵巧手控制界面 - {self.ros_manager.topic_prefix or "/"} '
f'{self.hand_type} {self.hand_joint}'
)
self.setMinimumSize(1200, 900)
# 设置样式
@@ -383,13 +569,16 @@ class HandControlGUI(QWidget):
for i, (name, value) in enumerate(zip(
self.hand_config.joint_names, self.hand_config.init_pos
)):
bounds = self.ros_manager.command_bounds(len(self.hand_config.init_pos))
minimum, maximum = bounds[i]
value = max(minimum, min(maximum, int(value)))
# 创建标签
label = QLabel(f"{name}: {value}")
label = QLabel(f"{name}: {self.format_position(value)}")
label.setMinimumWidth(120)
# 创建滑动条
slider = QSlider(Qt.Horizontal)
slider.setRange(0, 255)
slider.setRange(minimum, maximum)
slider.setValue(value)
slider.valueChanged.connect(
lambda val, idx=i: self.on_slider_value_changed(idx, val)
@@ -435,6 +624,11 @@ class HandControlGUI(QWidget):
self.stop_button.setProperty("category", "danger")
self.stop_button.clicked.connect(self.on_stop_clicked)
actions_layout.addWidget(self.stop_button)
self.save_pose_button = QPushButton("保存当前标定姿势")
self.save_pose_button.setProperty("category", "action")
self.save_pose_button.clicked.connect(self.on_save_pose_clicked)
actions_layout.addWidget(self.save_pose_button)
layout.addLayout(actions_layout)
@@ -443,9 +637,9 @@ class HandControlGUI(QWidget):
def create_system_preset_buttons(self, parent_layout):
"""创建系统预设动作按钮"""
self.preset_buttons = [] # 清空按钮列表
if self.hand_config.preset_actions:
if self.preset_actions:
buttons = []
for idx, (name, positions) in enumerate(self.hand_config.preset_actions.items()):
for idx, (name, positions) in enumerate(self.preset_actions.items()):
button = QPushButton(name)
button.setProperty("category", "preset")
button.clicked.connect(
@@ -472,6 +666,9 @@ class HandControlGUI(QWidget):
# —— 2. 新增:速度与扭矩设置(每行一个)——
quick_set_gb = QGroupBox("快速设置")
if self.hand_joint == "O12":
quick_set_gb.setEnabled(False)
quick_set_gb.setToolTip("O12不使用旧版0-255速度/扭矩设置接口")
qv_layout = QVBoxLayout(quick_set_gb)
# 速度行
@@ -600,7 +797,10 @@ class HandControlGUI(QWidget):
"""滑动条值改变事件处理"""
if 0 <= index < len(self.slider_labels):
joint_name = self.hand_config.joint_names[index]
self.slider_labels[index].setText(f"{joint_name}: {value}")
self.slider_labels[index].setText(
f"{joint_name}: {self.format_position(value)}"
)
self.command_dirty = True
# 更新数值显示
self.update_value_display()
@@ -608,10 +808,23 @@ class HandControlGUI(QWidget):
def update_value_display(self):
"""更新数值显示面板内容"""
# 获取所有滑动条的当前值
values = [slider.value() for slider in self.sliders]
# 格式化显示为列表形式
self.value_display.setText(f"{values}")
command = self.command_positions()
command_text = ", ".join(f"{value:.3f}" for value in command)
lines = [f"目标({self.hand_config.position_unit}): [{command_text}]"]
if self.feedback_positions is not None:
feedback_text = ", ".join(
f"{value:.3f}" for value in self.feedback_positions
)
lines.append(f"反馈(rad): [{feedback_text}]")
self.value_display.setText("\n".join(lines))
def on_feedback_updated(self, positions):
"""Display actual O12 joint feedback without moving command sliders."""
self.feedback_positions = list(positions)
self.update_value_display()
if hasattr(self, "connection_status"):
self.connection_status.setText("O12右手已连接并收到位置反馈")
self.connection_status.setObjectName("StatusInfo")
def on_preset_action_clicked(self, positions: List[int]):
"""预设动作按钮点击事件处理"""
@@ -650,11 +863,34 @@ class HandControlGUI(QWidget):
self.cycle_button.setText("循环运行预设动作")
self.reset_preset_buttons_color()
self.status_updated.emit("warning", "已停止所有动作")
if self.hand_joint == "O12" and self.feedback_positions is not None:
# Hold the latest measured pose instead of merely stopping topic
# publication while a previous position target is still active.
scale = self.hand_config.position_scale
hold_values = [
int(round(value * scale)) for value in self.feedback_positions
]
hold_values = self.ros_manager.bound_positions(hold_values)
for slider, value in zip(self.sliders, hold_values):
slider.blockSignals(True)
slider.setValue(value)
slider.blockSignals(False)
self.command_dirty = True
self.publish_joint_state()
self.update_value_display()
self.status_updated.emit("warning", "已下发当前反馈位置并保持")
else:
self.command_dirty = False
self.status_updated.emit("warning", "已停止连续动作和新的位置发布")
def on_save_pose_clicked(self):
"""发布当前滑块姿势,供hand_pose_capture写入profile。"""
positions = [slider.value() for slider in self.sliders]
self.ros_manager.publish_pose_snapshot(positions)
def on_cycle_clicked(self):
"""循环运行预设动作按钮点击事件处理"""
if not self.hand_config.preset_actions:
if not self.preset_actions:
QMessageBox.warning(self, "无预设动作", "当前手部型号没有预设动作可循环运行")
return
@@ -677,19 +913,19 @@ class HandControlGUI(QWidget):
def run_next_action(self):
"""运行下一个预设动作"""
if not self.hand_config.preset_actions:
if not self.preset_actions:
return
# 重置所有按钮颜色
self.reset_preset_buttons_color()
# 计算下一个动作索引
self.current_action_index = (self.current_action_index + 1) % len(self.hand_config.preset_actions)
self.current_action_index = (self.current_action_index + 1) % len(self.preset_actions)
# 获取下一个动作
action_names = list(self.hand_config.preset_actions.keys())
action_names = list(self.preset_actions.keys())
action_name = action_names[self.current_action_index]
action_positions = self.hand_config.preset_actions[action_name]
action_positions = self.preset_actions[action_name]
# 执行动作
self.on_preset_action_clicked(action_positions)
@@ -714,6 +950,7 @@ class HandControlGUI(QWidget):
"""关节类型改变事件处理"""
self.hand_joint = joint_type
self.hand_config = _HAND_CONFIGS[self.hand_joint]
self.preset_actions = self.hand_config.get_preset_actions(self.hand_type)
# 更新手部信息
info_text = f"""手部类型: {self.hand_type}
@@ -732,8 +969,11 @@ class HandControlGUI(QWidget):
def publish_joint_state(self):
"""发布当前关节状态"""
if self.hand_joint == "O12" and not self.command_dirty:
return
positions = [slider.value() for slider in self.sliders]
self.ros_manager.publish_joint_state(positions)
self.command_dirty = False
def update_status(self, status_type: str, message: str):
"""更新状态显示"""
@@ -0,0 +1,42 @@
#!/usr/bin/env python3
"""Start the HCAN-connected O12 right hand and its safe position GUI."""
from launch import LaunchDescription
from launch.actions import IncludeLaunchDescription, TimerAction
from launch.launch_description_sources import PythonLaunchDescriptionSource
from launch.substitutions import PathJoinSubstitution
from launch_ros.actions import Node
from launch_ros.substitutions import FindPackageShare
def generate_launch_description():
o12_driver = IncludeLaunchDescription(
PythonLaunchDescriptionSource(
PathJoinSubstitution([
FindPackageShare("omnihand_node"),
"launch",
"omnihand_pro_2025_node.launch.py",
])
)
)
o12_gui = Node(
package="gui_control",
executable="gui_control",
name="o12_right_gui_control",
output="screen",
parameters=[{
"hand_type": "right",
"hand_joint": "O12",
# Commands are coalesced, so dragging a slider sends at most 10 Hz.
"topic_hz": 10,
"is_arc": False,
"topic_prefix": "",
}],
)
return LaunchDescription([
o12_driver,
# Give the CANFD driver a moment to claim and initialise the adapter.
TimerAction(period=1.5, actions=[o12_gui]),
])
+1
View File
@@ -13,6 +13,7 @@
<test_depend>python3-pytest</test_depend>
<exec_depend>linker_hand_ros2_sdk</exec_depend>
<exec_depend>omnihand_node</exec_depend>
<export>
<build_type>ament_python</build_type>
+23
View File
@@ -0,0 +1,23 @@
from gui_control.config.constants import HAND_CONFIGS
def test_l6_gui_uses_the_sdk_channel_order() -> None:
assert HAND_CONFIGS["L6"].joint_names_en == [
"thumb_cmc_pitch",
"thumb_cmc_roll",
"index_mcp_pitch",
"middle_mcp_pitch",
"ring_mcp_pitch",
"pinky_mcp_pitch",
]
def test_l6_ok_preset_is_hand_specific() -> None:
config = HAND_CONFIGS["L6"]
assert config.get_preset_actions("left")["OK"] == [
62, 5, 151, 255, 255, 255,
]
assert config.get_preset_actions("right")["OK"] == [
58, 6, 153, 255, 255, 255,
]
+46
View File
@@ -0,0 +1,46 @@
from gui_control.config.constants import HAND_CONFIGS
def test_o12_gui_uses_active_angle_order_and_milliradians() -> None:
config = HAND_CONFIGS["O12"]
assert config.joint_names_en == [
"thumb_roll", "thumb_abad", "thumb_mcp", "thumb_pip",
"index_abad", "index_mcp", "index_pip",
"middle_abad", "middle_mcp", "middle_pip",
"ring_mcp", "pinky_mcp",
]
assert config.position_scale == 1000
assert config.position_unit == "rad"
assert len(config.init_pos) == 12
assert all(len(pose) == 12 for pose in config.preset_actions.values())
def test_o12_right_fingertip_preset_actions_use_requested_radians() -> None:
config = HAND_CONFIGS["O12"]
actions = config.get_preset_actions("right")
assert actions["拇指对食指"] == [
245, -766, -818, 0, 0, 1046, 0, 0, 0, 0, 0, 0,
]
assert actions["拇指对中指"] == [
489, -843, -685, 0, 0, 0, 0, 0, 1147, 0, 0, 0,
]
assert actions["拇指对无名指"] == [
699, -779, -827, 0, 0, 0, 0, 0, 0, 0, 690, 0,
]
assert actions["拇指对小指"] == [
786, -1065, -827, 0, 0, 0, 0, 0, 0, 0, 0, 645,
]
assert all(len(actions[name]) == 12 for name in (
"拇指对食指", "拇指对中指", "拇指对无名指", "拇指对小指",
))
def test_o12_fingertip_preset_actions_are_right_hand_only() -> None:
actions = HAND_CONFIGS["O12"].get_preset_actions("left")
assert "拇指对食指" not in actions
assert "拇指对中指" not in actions
assert "拇指对无名指" not in actions
assert "拇指对小指" not in actions
+18
View File
@@ -0,0 +1,18 @@
from gui_control.config.constants import HAND_CONFIGS
def test_o6_ok_preset_is_hand_specific() -> None:
config = HAND_CONFIGS["O6"]
assert config.get_preset_actions("left")["OK"] == [
95, 75, 116, 255, 255, 255,
]
assert config.get_preset_actions("right")["OK"] == [
95, 83, 122, 255, 255, 255,
]
assert config.get_preset_actions("left")["拇指对中指"] == [
88, 2, 255, 114, 255, 255,
]
assert config.get_preset_actions("right")["拇指对中指"] == [
95, 8, 255, 114, 255, 255,
]
-31
View File
@@ -1,31 +0,0 @@
from launch import LaunchDescription
from launch_ros.actions import Node
def generate_launch_description():
return LaunchDescription([
Node(
package='linker_hand_ros2_sdk',
executable='linker_hand_sdk',
name='linker_hand_sdk_left',
output='screen',
parameters=[{
'hand_type': 'left',
'hand_joint': "L10", # 这里需要修改为实际Linker Hand的型号 L7、L10、L20、L21、L25
'is_touch': True, # 是否带有压力传感器
'can': 'can0', # 这里需要修改为实际的CAN总线名称
}],
),
Node(
package='linker_hand_ros2_sdk',
executable='linker_hand_sdk',
name='linker_hand_sdk_right',
output='screen',
parameters=[{
'hand_type': 'right',
'hand_joint': "L10", # 这里需要修改为实际Linker Hand的型号 L7、L10、L20、L21、L25
'is_touch': True, # 是否带有压力传感器
'can': 'can0', # 这里需要修改为实际的CAN总线名称
}],
),
])
@@ -7,6 +7,7 @@ from enum import Enum
from utils.open_can import OpenCan
from can.exceptions import CanError
from utils.color_msg import ColorMsg
from utils.feedback import ReceivedPositionFrames
current_dir = os.path.dirname(os.path.abspath(__file__))
target_dir = os.path.abspath(os.path.join(current_dir, ".."))
sys.path.append(target_dir)
@@ -178,6 +179,10 @@ class LinkerHandG20Can:
# 串联控制数据存储
self.x41, self.x42, self.x43, self.x44, self.x45 = [], [], [], [], []
frame_layout = [[(frame, index) for index in range(6)] for frame in range(0x41, 0x46)]
layout = self.joint_state_to_cmd_state(frame_layout)
self.position_feedback = ReceivedPositionFrames(
{frame: 6 for frame in range(0x41, 0x46)}, [item if item != 0 else None for item in layout])
self.x49, self.x4A, self.x4B, self.x4C, self.x4D = [0] * 6, [0] * 6, [0] * 6, [0] * 6, [0] * 6
self.x51, self.x52, self.x53, self.x54, self.x55 = [], [], [], [], []
self.x59, self.x5A, self.x5B, self.x5C, self.x5D = [], [], [], [], []
@@ -354,6 +359,8 @@ class LinkerHandG20Can:
response_data = msg.data[1:]
if len(list(response_data)) == 0:
return
if frame_type in range(0x41, 0x46):
self.position_feedback.accept(frame_type, response_data, msg.timestamp)
# 并联控制指令响应
if frame_type == 0x01: self.x01 = list(response_data)
elif frame_type == 0x02: self.x02 = list(response_data)
@@ -1030,6 +1037,19 @@ class LinkerHandG20Can:
cmd_state = self.joint_state_to_cmd_state(state=s)
return cmd_state
def get_cached_current_status(self):
"""Return the latest received five-finger state without CAN queries."""
state = [self.x41, self.x42, self.x43, self.x44, self.x45]
if not all(
isinstance(finger, (list, tuple)) and len(finger) == 6
for finger in state
):
return None
return self.joint_state_to_cmd_state(state=state)
def get_feedback_snapshot(self):
return self.position_feedback.snapshot()
def get_current_pub_status(self):
"""API接口:获取手指当前状态"""
self.get_current_status()
@@ -1266,4 +1286,25 @@ class LinkerHandG20Can:
except:
return "-1"
def get_finger_order(self):
return ["Thumb Base", "Index Finger Base", "Middle Finger Base", "Ring Finger Base", "Pinky Finger Base", "Thumb Abduction", "Index Finger Abduction", "Middle Finger Abduction", "Ring Finger Abduction", "Pinky Finger Abduction", "Thumb Horizontal Abduction", "Reserved", "Reserved", "Reserved", "Reserved", "Thumb Tip", "Index Finger Tip", "Middle Finger Tip", "Ring Finger Tip", "Pinky Finger Tip"]
return [
"thumb_cmc_pitch",
"index_mcp_pitch",
"middle_mcp_pitch",
"ring_mcp_pitch",
"pinky_mcp_pitch",
"thumb_cmc_roll",
"index_mcp_roll",
"middle_mcp_roll",
"ring_mcp_roll",
"pinky_mcp_roll",
"thumb_cmc_yaw",
"reserved_11",
"reserved_12",
"reserved_13",
"reserved_14",
"thumb_mcp",
"index_pip",
"middle_pip",
"ring_pip",
"pinky_pip",
]
@@ -1,9 +1,12 @@
from collections import deque
import can
import time, sys
import threading
import numpy as np
from utils.open_can import OpenCan
from utils.color_msg import ColorMsg
from utils.feedback import ReceivedPositionFrames
from can.exceptions import CanError
@@ -15,6 +18,7 @@ class LinkerHandL6Can:
self.open_can = OpenCan(load_yaml=yaml)
self.x01 = [0] * 6 # 关节位置
self.position_feedback = ReceivedPositionFrames({0x01: 6}, [(0x01, i) for i in range(6)])
self.x02 = [-1] * 6 # 转矩限制
self.x05 = [0] * 6 # 速度
self.x07 = [-1] * 6 # 加速度
@@ -57,6 +61,14 @@ class LinkerHandL6Can:
self.normal_force, self.tangential_force, self.tangential_force_dir, self.approach_inc = [[-1] * 6 for _ in range(4)]
self.is_lock = False
self.version = None
# L6 replies to a six-byte 0x01 position command with an immediate
# byte-for-byte echo on the same CAN ID. A zero-payload 0x01 state
# query also replies on that ID, but with the measured positions.
# Keep the two transactions distinct so command echoes never enter
# the published feedback stream used by calibration.
self._position_echo_lock = threading.Lock()
self._pending_position_echoes = deque(maxlen=32)
self._position_echo_timeout_seconds = 0.02
# Start the receiving thread
self.running = True
self.receive_thread = threading.Thread(target=self.receive_response)
@@ -111,6 +123,11 @@ class LinkerHandL6Can:
frame_property_value = int(frame_property.value) if hasattr(frame_property, 'value') else frame_property
data = [frame_property_value] + [int(val) for val in data_list]
msg = can.Message(arbitration_id=self.can_id, data=data, is_extended_id=False)
if frame_property_value == 0x01 and len(data_list) == 6:
with self._position_echo_lock:
self._pending_position_echoes.append(
(time.monotonic(), tuple(int(value) for value in data_list))
)
try:
self.bus.send(msg)
except can.CanError as e:
@@ -201,7 +218,24 @@ class LinkerHandL6Can:
except:
return
if frame_type == 0x01: # 0x01
self.x01 = list(response_data)
response = tuple(int(value) for value in response_data)
now = time.monotonic()
is_position_echo = False
with self._position_echo_lock:
while (
self._pending_position_echoes
and now - self._pending_position_echoes[0][0]
> self._position_echo_timeout_seconds
):
self._pending_position_echoes.popleft()
for pending in tuple(self._pending_position_echoes):
if pending[1] == response:
self._pending_position_echoes.remove(pending)
is_position_echo = True
break
if not is_position_echo:
self.x01 = list(response)
self.position_feedback.accept(frame_type, response, msg.timestamp)
elif frame_type == 0x02: # 0x02
self.x02 = list(response_data)
elif frame_type == 0x05: # Set speed
@@ -295,6 +329,9 @@ class LinkerHandL6Can:
def get_current_pub_status(self):
return self.x01
def get_feedback_snapshot(self):
return self.position_feedback.snapshot()
def get_speed(self):
#self.send_frame(0x05, [],sleep=0.003)
#print("L6暂不支持读取实时速度")
@@ -391,7 +428,10 @@ class LinkerHandL6Can:
return self.x35
def get_finger_order(self):
return ["thumb_cmc_pitch", "thumb_cmc_yaw", "index_mcp_pitch", "middle_mcp_pitch", "ring_mcp_pitch", "pinky_mcp_pitch"]
# L6 channel 1 is the physical CMC roll actuator. Older SDK releases
# exposed the channel as ``thumb_cmc_yaw`` even though the wire order
# and mechanism have always been roll.
return ["thumb_cmc_pitch", "thumb_cmc_roll", "index_mcp_pitch", "middle_mcp_pitch", "ring_mcp_pitch", "pinky_mcp_pitch"]
def show_fun_table(self):
pass
@@ -4,6 +4,7 @@ import threading
import numpy as np
from utils.open_can import OpenCan
from utils.color_msg import ColorMsg
from utils.feedback import ReceivedPositionFrames
from can.exceptions import CanError
@@ -15,6 +16,7 @@ class LinkerHandO6Can:
self.open_can = OpenCan(load_yaml=yaml)
self.x01 = [0] * 6 # 关节位置
self.position_feedback = ReceivedPositionFrames({0x01: 6}, [(0x01, i) for i in range(6)])
self.x02 = [-1] * 6 # 转矩限制
self.x05 = [0] * 6 # 速度
self.x07 = [-1] * 6 # 加速度
@@ -224,6 +226,7 @@ class LinkerHandO6Can:
return
if frame_type == 0x01: # 0x01
self.x01 = list(response_data)
self.position_feedback.accept(frame_type, response_data, msg.timestamp)
elif frame_type == 0x02: # 0x02
self.x02 = list(response_data)
elif frame_type == 0x05: # Set speed
@@ -317,6 +320,9 @@ class LinkerHandO6Can:
def get_current_pub_status(self):
return self.x01
def get_feedback_snapshot(self):
return self.position_feedback.snapshot()
def get_speed(self):
self.send_frame(0x05, [],sleep=0.002)
#print("L6暂不支持读取实时速度")
@@ -372,7 +372,7 @@ class LinkerHandL6RS485:
return [0] * 6
def get_finger_order(self):
return ["thumb_cmc_pitch", "thumb_cmc_yaw", "index_mcp_pitch", "middle_mcp_pitch", "ring_mcp_pitch", "pinky_mcp_pitch"]
return ["thumb_cmc_pitch", "thumb_cmc_roll", "index_mcp_pitch", "middle_mcp_pitch", "ring_mcp_pitch", "pinky_mcp_pitch"]
# --------------------------------------------------
# 便捷方法
@@ -1,444 +0,0 @@
#!/usr/bin/env python3
import os
import time
from pymodbus.client import ModbusSerialClient
from typing import List, Dict
import numpy as np
_INTERVAL = 0.006 # 8 ms
class LinkerHandL6RS485:
"""L6机械手 Modbus-RTU 控制类"""
# 6个关节名称
JOINT_NAMES = ["thumb_pitch", "thumb_yaw", "index_pitch",
"middle_pitch", "ring_pitch", "little_pitch"]
# 手指名称
FINGER_NAMES = ["thumb", "index", "middle", "ring", "little"]
def __init__(self, hand_id=0x27, modbus_port="/dev/ttyUSB0", baudrate=115200):
"""
初始化L6机械手
hand_id: 右手0x27(39), 左手0x28(40)
modbus_port: 串口设备路径
baudrate: 波特率,固定115200
"""
self.slave = hand_id
self.cli = ModbusSerialClient(
port=modbus_port,
baudrate=baudrate,
bytesize=8,
parity="N",
stopbits=1,
timeout=0.05
)
# pymodbus 3.5.1 需要显式连接
self.connected = self.cli.connect()
if not self.connected:
raise ConnectionError(f"RS485连接失败,端口: {modbus_port}")
def _read_input_registers(self, address: int, count: int) -> List[int]:
"""读取输入寄存器"""
time.sleep(_INTERVAL)
result = self.cli.read_input_registers(address=address, count=count, slave=self.slave)
if result.isError():
raise RuntimeError(f"读取输入寄存器失败: address={address}, count={count}")
return result.registers
def _write_register(self, address: int, value: int):
"""写入单个寄存器"""
time.sleep(_INTERVAL)
result = self.cli.write_register(address=address, value=value, slave=self.slave)
if result.isError():
raise RuntimeError(f"写入寄存器失败: address={address}, value={value}")
def _write_registers(self, address: int, values: List[int]):
"""写入多个寄存器"""
time.sleep(_INTERVAL)
result = self.cli.write_registers(address=address, values=values, slave=self.slave)
if result.isError():
raise RuntimeError(f"写入多个寄存器失败: address={address}, values={values}")
# --------------------------------------------------
# 基础读取接口
# --------------------------------------------------
def read_angles(self) -> List[int]:
"""读取6个关节角度 (输入寄存器 0-5)"""
return self._read_input_registers(0, 6)
def read_torques(self) -> List[int]:
"""读取6个关节转矩 (输入寄存器 6-11)"""
return self._read_input_registers(6, 6)
def read_speeds(self) -> List[int]:
"""读取6个关节速度 (输入寄存器 12-17)"""
return self._read_input_registers(12, 6)
def read_temperatures(self) -> List[int]:
"""读取6个关节温度 (输入寄存器 18-23)"""
return self._read_input_registers(18, 6)
def read_error_codes(self) -> List[int]:
"""读取6个关节错误码 (输入寄存器 24-29)"""
return self._read_input_registers(24, 6)
# --------------------------------------------------
# 压力传感器接口
# --------------------------------------------------
# def _pressure(self, finger: int) -> List[int]:
# """内部:选手指 → 读压力数据"""
# # 选择手指 (保持寄存器 36)
# self._write_register(36, finger)
# time.sleep(_INTERVAL)
# # 读取压力数据 (输入寄存器 52-122)
# return np.array(self._read_input_registers(52, 71))
def _pressure(self, finger: int) -> np.ndarray:
"""
6x12 (72点) 矩阵尺寸。
Modbus 地址 60/62。
"""
rows = 12 # 12 行
cols = 6 # 6 列
finger_size = rows * cols # 72 个数据点
# modbus 地址和计数
write_address = 60 # 写入手指选择
read_address = 62 # 读取压力数据
read_count = 96 # 读取 96 个寄存器
skip_count = 10 # 跳过前 10 个校验点
# 0. 参数校验和手指写入值确定
if finger < 1 or finger > 5:
raise ValueError(f"无效的手指编号: {finger}。手指编号应在 1 到 5 之间。")
finger_write_value = finger
# 1. 写入手指选择寄存器 (地址 60)
time.sleep(0.008)
wrsp = self.cli.write_register(address=write_address, value=finger_write_value, slave=self.slave)
if wrsp.isError():
raise RuntimeError(f"写入手指选择 {finger} 到地址 {write_address} 失败: {wrsp}")
# 写入后等待片刻
time.sleep(0.008)
# 2. 读取地址 62 的数据
rrsp = self.cli.read_input_registers(address=read_address, count=read_count, slave=self.slave)
if rrsp.isError():
raise RuntimeError(f"读取地址 {read_address} 压力数据失败: {rrsp}")
registers_16bit: List[int] = rrsp.registers
# 3. 核心数据处理
# a. 提取低 8 位数据 (得到 96 个 8 位数据点)
final_data_96 = [reg_value & 255 for reg_value in registers_16bit]
# b. 跳过前 10 个校验/头部数据点 (得到 86 个有效数据点)
effective_data = np.array(final_data_96[skip_count:], dtype=np.uint8)
# c. 截取当前手指的矩阵数据 (从 86 个有效点中截取 72 个点)
start_idx = 0
end_idx = finger_size # 72
finger_data_flat = effective_data[start_idx:end_idx]
# d. 验证数据长度
if finger_data_flat.size != finger_size:
raise ValueError(
f"数据提取失败。期望 {finger_size} 点 ({rows}x{cols})"
f"但仅截取到 {finger_data_flat.size} 点。请检查协议,确认地址 62 是否一次性返回了所有手指数据。"
)
# e. 重塑为二维矩阵 (12 行 6 列)
finger_matrix = finger_data_flat.reshape((rows, cols))
return finger_matrix
def read_pressure_thumb(self) -> np.ndarray:
"""读取大拇指压力数据"""
return np.array(self._pressure(1), dtype=np.uint8)
def read_pressure_index(self) -> np.ndarray:
"""读取食指压力数据"""
return np.array(self._pressure(2), dtype=np.uint8)
def read_pressure_middle(self) -> np.ndarray:
"""读取中指压力数据"""
return np.array(self._pressure(3), dtype=np.uint8)
def read_pressure_ring(self) -> np.ndarray:
"""读取无名指压力数据"""
return np.array(self._pressure(4), dtype=np.uint8)
def read_pressure_little(self) -> np.ndarray:
"""读取小拇指压力数据"""
return np.array(self._pressure(5), dtype=np.uint8)
# --------------------------------------------------
# 版本信息接口
# --------------------------------------------------
def read_versions(self) -> Dict[str, int]:
"""读取版本信息 (输入寄存器 148-155)"""
result = self._read_input_registers(148, 8)
return {
"hand_freedom": result[0],
"hand_version": result[1],
"hand_number": result[2],
"hand_direction": result[3],
"software_version_major": result[4],
"software_version_minor": result[5] if len(result) > 5 else 0,
"software_version_revision": result[6] if len(result) > 6 else 0,
"hardware_version": result[7] if len(result) > 7 else 0
}
# --------------------------------------------------
# 写入接口
# --------------------------------------------------
def write_angles(self, vals: List[int]):
"""设置6个关节角度 (保持寄存器 0-5)"""
vals = [int(x) for x in vals]
if not self.is_valid_6xuint8(vals):
raise ValueError("需要6个0-255的整数")
self._write_registers(0, vals)
def write_torques(self, vals: List[int]):
"""设置6个关节转矩 (保持寄存器 6-11)"""
vals = [int(x) for x in vals]
if not self.is_valid_6xuint8(vals):
raise ValueError("需要6个0-255的整数")
self._write_registers(6, vals)
def write_speeds(self, vals: List[int]):
"""设置6个关节速度 (保持寄存器 12-17)"""
vals = [int(x) for x in vals]
if not self.is_valid_6xuint8(vals):
raise ValueError("需要6个0-255的整数")
self._write_registers(12, vals)
# --------------------------------------------------
# 上下文管理
# --------------------------------------------------
def close(self):
"""关闭连接"""
if self.connected:
self.cli.close()
self.connected = False
def __enter__(self):
return self
def __exit__(self, exc_type, exc_val, exc_tb):
self.close()
# --------------------------------------------------
# API固定接口函数
# --------------------------------------------------
def is_valid_6xuint8(self, lst) -> bool:
"""验证6个0-255的整数列表"""
if len(lst) != 6:
return False
return all(isinstance(x, int) and 0 <= x <= 255 for x in lst)
def set_joint_positions(self, joint_angles=None):
"""设置关节位置"""
joint_angles = joint_angles or [0] * 6
self.write_angles(joint_angles)
def set_speed(self, speed=None):
"""设置速度"""
speed = speed or [200] * 6
self.write_speeds(speed)
def set_torque(self, torque=None):
"""设置扭矩"""
torque = torque or [200] * 6
self.write_torques(torque)
def set_current(self, current=None):
"""设置电流 (L6不支持)"""
print("当前L6不支持设置电流", flush=True)
def get_version(self) -> list:
"""获取版本信息"""
versions = self.read_versions()
return [
versions.get("hand_freedom", 0),
versions.get("hand_version", 0),
versions.get("hand_number", 0),
versions.get("hand_direction", 0),
versions.get("software_version_major", 0),
versions.get("hardware_version", 0)
]
def get_current(self):
"""获取电流 (L6不支持)"""
print("当前L6不支持获取电流", flush=True)
return []
def get_state(self) -> list:
"""获取关节状态"""
return self.read_angles()
def get_state_for_pub(self) -> list:
return self.get_state()
def get_current_status(self) -> list:
return self.get_state()
def get_speed(self) -> list:
"""获取当前速度"""
return self.read_speeds()
def get_joint_speed(self) -> list:
return self.get_speed()
def get_touch_type(self) -> int:
"""获取压感类型 (2=矩阵式)"""
return 2
def get_normal_force(self) -> list:
"""获取压感数据:点式"""
return [-1] * 5
def get_tangential_force(self) -> list:
"""获取压感数据:点式"""
return [-1] * 5
def get_approach_inc(self) -> list:
"""获取压感数据:点式"""
return [-1] * 5
def get_touch(self) -> list:
return [-1] * 5
def get_thumb_matrix_touch(self,sleep_time=0):
return self._pressure(1)
def get_index_matrix_touch(self,sleep_time=0):
return self._pressure(2)
def get_middle_matrix_touch(self,sleep_time=0):
return self._pressure(3)
def get_ring_matrix_touch(self,sleep_time=0):
return self._pressure(4)
def get_little_matrix_touch(self,sleep_time=0):
return self._pressure(5)
def get_matrix_touch(self) -> list:
"""获取压感数据:矩阵式"""
return [self._pressure(1), self._pressure(2), self._pressure(3),
self._pressure(4), self._pressure(5)]
def get_matrix_touch_v2(self) -> list:
"""获取压感数据:矩阵式"""
return self.get_matrix_touch()
def get_torque(self) -> list:
"""获取当前扭矩"""
return self.read_torques()
def get_temperature(self) -> list:
"""获取当前电机温度"""
return self.read_temperatures()
def get_fault(self) -> list:
"""获取当前电机故障码"""
return self.read_error_codes()
def get_serial_number(self):
return [0] * 6
# --------------------------------------------------
# 便捷方法
# --------------------------------------------------
def relax(self):
"""所有手指伸直"""
self.set_joint_positions([255] * 6)
def fist(self):
"""所有手指握拳"""
self.set_joint_positions([0] * 6)
def dump_status(self):
"""打印状态信息"""
print("=" * 50)
print("L6机械手状态信息")
print("=" * 50)
try:
# 关节状态
angles = self.read_angles()
torques = self.read_torques()
speeds = self.read_speeds()
temps = self.read_temperatures()
errors = self.read_error_codes()
print("关节状态:")
for i, name in enumerate(self.JOINT_NAMES):
print(f" {name:15s}: 角度={angles[i]:3d}, 扭矩={torques[i]:3d}, "
f"速度={speeds[i]:3d}, 温度={temps[i]:2d}℃, 错误={errors[i]:2d}")
# 版本信息
versions = self.read_versions()
print("\n版本信息:")
for key, value in versions.items():
print(f" {key:20s}: {value}")
# 压力传感器测试
print("\n压力传感器测试:")
thumb_pressure = self.read_pressure_thumb()
print(f"大拇指压力数据长度: {len(thumb_pressure)}")
except Exception as e:
print(f"读取状态时出错: {e}")
print("=" * 50)
# ------------------- 演示程序 -------------------
if __name__ == "__main__":
# 使用示例
try:
with LinkerHandL6RS485(hand_id=0x27, modbus_port="/dev/ttyUSB0", baudrate=115200) as hand:
print("连接成功!")
# 打印状态信息
hand.dump_status()
# 测试基本控制
print("\n测试控制功能...")
print("伸直手指...")
hand.relax()
time.sleep(2)
print("握拳...")
hand.fist()
time.sleep(2)
print("恢复伸直...")
hand.relax()
# 测试压力传感器
print("\n测试压力传感器...")
thumb_matrix = hand.get_thumb_matrix_touch()
print(f"大拇指压力数据: {len(thumb_matrix)}个点")
# 获取所有手指压力数据
all_matrices = hand.get_matrix_touch()
for i, name in enumerate(hand.FINGER_NAMES):
matrix = all_matrices[i]
print(f"{name}手指压力数据长度: {len(matrix)}")
except Exception as e:
print(f"错误: {e}")
@@ -202,6 +202,19 @@ class LinkerHandApi:
'''Get current joint state'''
return self.hand.get_current_status()
def get_state_cached(self):
"""Get the latest received state without transmitting new queries."""
getter = getattr(self.hand, "get_cached_current_status", None)
return getter() if getter is not None else None
@property
def supports_feedback_timestamps(self):
return callable(getattr(self.hand, "get_feedback_snapshot", None))
def get_feedback_snapshot(self):
getter = getattr(self.hand, "get_feedback_snapshot", None)
return getter() if getter is not None else None
def get_state_for_pub(self):
return self.hand.get_current_pub_status()
@@ -0,0 +1,65 @@
"""Atomic CAN position snapshots with the original receive timestamps."""
from dataclasses import dataclass
import math
from threading import Lock
@dataclass(frozen=True)
class FeedbackSnapshot:
positions: tuple
channel_stamps_ns: tuple[int, ...]
@property
def stamp_ns(self):
return max(self.channel_stamps_ns)
def as_dict(self, names):
return {"schema": "can_feedback_v1", "name": list(names),
"position": list(self.positions), "channel_stamps_ns": list(self.channel_stamps_ns),
"stamp_ns": self.stamp_ns}
class ReceivedPositionFrames:
"""Keep receipt identity even if stationary feedback has identical values.
``layout`` lists the frame and value index for each public SDK channel.
None represents a protocol-reserved constant-zero slot, whose freshness
is bounded by the oldest actual frame. No polling/publication clock is
allowed to update a snapshot's timestamps.
"""
def __init__(self, sizes, layout):
self.sizes, self.layout = dict(sizes), tuple(layout)
self._frames = {}
self._lock = Lock()
def accept(self, frame, values, timestamp):
if frame not in self.sizes or len(values) != self.sizes[frame]:
return False
if not isinstance(timestamp, (int, float)) or not math.isfinite(timestamp) or timestamp <= 0:
return False
stamp = int(round(timestamp * 1_000_000_000))
with self._lock:
previous = self._frames.get(frame)
if previous is not None and stamp <= previous[0]:
return False
self._frames[frame] = (stamp, tuple(values))
return True
def snapshot(self):
with self._lock:
if self._frames.keys() != self.sizes.keys():
return None
oldest = min(value[0] for value in self._frames.values())
positions, stamps = [], []
for item in self.layout:
if item is None:
positions.append(0)
stamps.append(oldest)
else:
frame, index = item
stamp, values = self._frames[frame]
positions.append(values[index])
stamps.append(stamp)
return FeedbackSnapshot(tuple(positions), tuple(stamps))
@@ -37,11 +37,30 @@ def command_changed(previous, current):
return any(float(old) != float(new) for old, new in zip(previous, values))
def position_command_should_queue(previous, current, repeat=True):
"""Whether the newest position target should be written on this heartbeat."""
values = list(current)
return bool(values) and (
bool(repeat) or command_changed(previous, values)
)
def state_poll_due(last_poll_time, now, poll_period):
"""Keep slow CAN state reads off the latency-sensitive command path."""
return last_poll_time is None or now >= last_poll_time + poll_period
def state_reads_deferred(
last_command_time, now, quiet_period, enabled=True
):
"""Return whether blocking state reads must yield to active commands."""
return (
enabled
and last_command_time is not None
and now < last_command_time + quiet_period
)
class LinkerHand(Node):
def __init__(self, name):
super().__init__(name)
@@ -54,6 +73,12 @@ class LinkerHand(Node):
# -1 keeps the model's original startup speed. Camera teleoperation can
# set this to a conservative value before the startup pose is sent.
self.declare_parameter('startup_speed', -1)
# -1 keeps the model's original startup torque. The retarget v2
# launch uses a conservative value for calibration and preview.
self.declare_parameter('startup_torque', -1)
# Preserve the legacy behaviour by default. Safety-critical launch
# files can configure limits without moving to a startup pose.
self.declare_parameter('move_on_startup', True)
# Empty keeps the legacy absolute topics/startup pose. A prefix lets
# two same-side hands coexist without receiving each other's commands.
self.declare_parameter('topic_prefix', '')
@@ -63,6 +88,17 @@ class LinkerHand(Node):
# incoming position commands.
self.declare_parameter('state_poll_rate', 60.0)
self.declare_parameter('velocity_poll_rate', 60.0)
# G20 state and velocity reads each transmit five synchronous CAN
# queries. Defer them while teleoperation commands are arriving.
self.declare_parameter('defer_state_reads_while_commanding', True)
self.declare_parameter('command_quiet_period', 0.2)
# Legacy teleoperation sent the latest target on every 30 Hz callback,
# including an unchanged target. Some firmware revisions track that
# cadence more smoothly than sparse change-only updates.
self.declare_parameter('repeat_position_commands', True)
# Faults stay manually clearable through cb_hand_setting_cmd. Repeated
# automatic clears add periodic CAN traffic to the command stream.
self.declare_parameter('auto_clear_faults', False)
# ros时间获取
self.stamp_clock = Clock()
@@ -75,6 +111,12 @@ class LinkerHand(Node):
self.startup_speed = int(self.get_parameter('startup_speed').value)
if self.startup_speed < -1 or self.startup_speed > 255:
raise ValueError('startup_speed must be -1 or in the range [0, 255]')
self.startup_torque = int(self.get_parameter('startup_torque').value)
if self.startup_torque < -1 or self.startup_torque > 255:
raise ValueError('startup_torque must be -1 or in the range [0, 255]')
self.move_on_startup = bool(
self.get_parameter('move_on_startup').value
)
self.topic_prefix = self.normalize_topic_prefix(
self.get_parameter('topic_prefix').value
)
@@ -92,6 +134,22 @@ class LinkerHand(Node):
raise ValueError('velocity_poll_rate must be greater than zero')
self.velocity_poll_period = 1.0 / self.velocity_poll_rate
self.last_velocity_poll_time = None
self.defer_state_reads_while_commanding = bool(
self.get_parameter(
'defer_state_reads_while_commanding'
).value
)
self.command_quiet_period = float(
self.get_parameter('command_quiet_period').value
)
if self.command_quiet_period < 0.0:
raise ValueError('command_quiet_period must not be negative')
self.auto_clear_faults = bool(
self.get_parameter('auto_clear_faults').value
)
self.repeat_position_commands = bool(
self.get_parameter('repeat_position_commands').value
)
configured_startup_pose = self.get_parameter_or(
'startup_pose',
Parameter('startup_pose', Parameter.Type.INTEGER_ARRAY, []),
@@ -107,8 +165,11 @@ class LinkerHand(Node):
self.last_hand_eff_cmd = None # 最新手指力矩命令
self.applied_hand_post_cmd = None
self.applied_hand_vel_cmd = None
self.last_position_command_time = None
self.last_hand_state = [-1] * 10
self.last_hand_state_stamp_ns = None
self._last_timed_feedback_stamps = None
self.last_hand_vel = [-1] * 10
self.force = [[-1] * 5] * 4
self.matrix_dic = {
@@ -186,6 +247,7 @@ class LinkerHand(Node):
COMMAND_QOS,
)
self.hand_state_pub = self.create_publisher(JointState, self.topic(f'/cb_{self.hand_type}_hand_state'),10)
self.timed_state_pub = self.create_publisher(String, self.topic(f'/cb_{self.hand_type}_hand_state_timed'), 30)
self.hand_info_pub = self.create_publisher(String, self.topic(f'/cb_{self.hand_type}_hand_info'), 10)
if self.is_touch == True:
if self.modbus != "None":
@@ -244,26 +306,38 @@ class LinkerHand(Node):
pose = list(self.startup_pose)
if self.startup_speed >= 0:
speed = [self.startup_speed] * len(speed)
if self.startup_torque >= 0:
torque = [self.startup_torque] * len(torque)
if pose is not None:
for i in range(1):
self.api.set_speed(speed=speed)
time.sleep(0.1)
self.api.set_torque(torque=torque)
time.sleep(0.1)
self.api.finger_move(pose=pose)
time.sleep(0.1)
if self.move_on_startup:
self.api.finger_move(pose=pose)
time.sleep(0.1)
def hand_control_cb(self, msg):
# The hardware can be slower than the camera. Always replace a
# pending command with the newest sample and never replay an already
# applied sample; this prevents latency from accumulating in software.
# The hardware can be slower than the publisher, so a pending target is
# always replaced by the newest sample. By default the newest target
# is also resent at the publisher cadence, matching the legacy driver.
position = list(msg.position)
if position:
self.last_hand_post_cmd = (
position
if command_changed(self.applied_hand_post_cmd, position)
else None
)
# Treat every valid sample as an active teleoperation heartbeat,
# even if integer quantisation made it identical to the previous
# target. This keeps all synchronous CAN diagnostics off the bus
# for the entire control session, matching the legacy execution
# path that had no state subscriber.
self.last_position_command_time = time.monotonic()
if position_command_should_queue(
self.applied_hand_post_cmd,
position,
self.repeat_position_commands,
):
self.last_hand_post_cmd = position
else:
self.last_hand_post_cmd = None
velocity = list(msg.velocity)
if velocity:
@@ -286,6 +360,9 @@ class LinkerHand(Node):
self.api.finger_move(pose=pose)
self.applied_hand_post_cmd = pose
self.last_hand_post_cmd = None
cached_state = self.api.get_state_cached()
if cached_state is not None:
self.last_hand_state = cached_state
if self.last_hand_vel_cmd is not None:
vel = list(self.last_hand_vel_cmd)
@@ -317,9 +394,17 @@ class LinkerHand(Node):
self.last_hand_vel_cmd = None
def _poll_state_if_due(self):
if self.hand_state_pub.get_subscription_count() < 1:
if self.hand_state_pub.get_subscription_count() < 1 and self.timed_state_pub.get_subscription_count() < 1:
return
now = time.monotonic()
if state_reads_deferred(
self.last_position_command_time,
now,
self.command_quiet_period,
self.defer_state_reads_while_commanding,
):
# A repeated heartbeat retains its original measurement time.
return
if not state_poll_due(
self.last_state_poll_time, now, self.state_poll_period
):
@@ -328,6 +413,7 @@ class LinkerHand(Node):
# another read on the following timer callback.
self.last_state_poll_time = now
self.last_hand_state = self.api.get_state()
self.last_hand_state_stamp_ns = self.get_clock().now().nanoseconds
time.sleep(0.003)
if state_poll_due(
self.last_velocity_poll_time, now, self.velocity_poll_period
@@ -342,6 +428,12 @@ class LinkerHand(Node):
# Position commands have priority over synchronous state reads.
self._apply_pending_commands()
self._poll_state_if_due()
diagnostics_deferred = state_reads_deferred(
self.last_position_command_time,
time.monotonic(),
self.command_quiet_period,
self.defer_state_reads_while_commanding,
)
if self.cmd_lock == False:
time.sleep(0.003)
if self.run_count == 3 and self.is_touch == True and self.touch_type == 1 and self.modbus == "None" and self.touch_pub.get_subscription_count() > 0:
@@ -360,7 +452,11 @@ class LinkerHand(Node):
if self.run_count == 7:
self.matrix_dic["little_matrix"] = self.api.get_little_matrix_touch(sleep_time=self.sleep_time).tolist()
time.sleep(0.005)
if self.run_count == 8 and self.hand_info_pub.get_subscription_count() > 0:
if (
self.run_count == 8
and self.hand_info_pub.get_subscription_count() > 0
and not diagnostics_deferred
):
"""手部信息"""
self.last_hand_info = {
"version": self.embedded_version, # Dexterous hand version number
@@ -375,8 +471,9 @@ class LinkerHand(Node):
"finger_order": self.api.get_finger_order() # Finger motor order
}
if self.run_count == 9:
if self.run_count == 9 and self.auto_clear_faults:
self.api.clear_faults() # 自动清除错误编码
if self.run_count == 9:
self.run_count = 0
self.run_count += 1
time.sleep(0.003)
@@ -384,9 +481,7 @@ class LinkerHand(Node):
def pub_state(self):
while True:
if self.hand_state_pub.get_subscription_count() > 0:
msg = self.joint_state_msg(self.last_hand_state, self.last_hand_vel)
self.hand_state_pub.publish(msg)
self._publish_feedback()
if self.is_touch == True and self.touch_type == 1 and self.modbus == "None" and self.touch_pub.get_subscription_count() > 0:
msg = Float32MultiArray()
msg.data = [float(val) for sublist in self.force for val in sublist]
@@ -404,6 +499,26 @@ class LinkerHand(Node):
self.hand_info_pub.publish(msg)
time.sleep(self.hz)
def _publish_feedback(self):
"""Publish one atomic measured state; never give cached values a new time."""
if self.api.supports_feedback_timestamps:
snapshot = self.api.get_feedback_snapshot()
if snapshot is None:
return
positions, stamp = snapshot.positions, snapshot.stamp_ns
if (snapshot.channel_stamps_ns != self._last_timed_feedback_stamps
and self.timed_state_pub.get_subscription_count() > 0):
message = String()
message.data = json.dumps(snapshot.as_dict(self.api.get_finger_order()))
self.timed_state_pub.publish(message)
self._last_timed_feedback_stamps = snapshot.channel_stamps_ns
else:
positions, stamp = self.last_hand_state, self.last_hand_state_stamp_ns
if stamp is None:
return
if self.hand_state_pub.get_subscription_count() > 0:
self.hand_state_pub.publish(self.joint_state_msg(positions, self.last_hand_vel, stamp_ns=stamp))
def pub_matrix_mass(self, dic):
"""发布矩阵数据合值 单位g 克 JSON格式"""
msg = String()
@@ -462,10 +577,12 @@ class LinkerHand(Node):
msg.data = json.dumps(self.matrix_dic)
self.matrix_touch_pub.publish(msg)
def joint_state_msg(self, pose,vel=[]):
def joint_state_msg(self, pose, vel=(), *, stamp_ns=None):
joint_state = JointState()
joint_state.header = Header()
joint_state.header.stamp = self.get_clock().now().to_msg()
if stamp_ns is None:
stamp_ns = self.get_clock().now().nanoseconds
joint_state.header.stamp.sec, joint_state.header.stamp.nanosec = divmod(int(stamp_ns), 1_000_000_000)
joint_state.name = self.api.get_finger_order()
joint_state.position = [float(x) for x in pose]
if len(vel) > 1:
@@ -1,414 +0,0 @@
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
'''
编译: colcon build --symlink-install
启动命令:ros2 run linker_hand_ros2_sdk linker_hand_sdk
'''
from re import A
import rclpy,sys # ROS2 Python接口库
import time
import numpy as np
from rclpy.node import Node # ROS2 节点类
from rclpy.clock import Clock
from std_msgs.msg import String, Header, Float32MultiArray
from sensor_msgs.msg import JointState, PointCloud2, PointField
import time, json, threading
from linker_hand_ros2_sdk.LinkerHand.linker_hand_api import LinkerHandApi
from linker_hand_ros2_sdk.LinkerHand.utils.color_msg import ColorMsg
from linker_hand_ros2_sdk.LinkerHand.utils.open_can import OpenCan
class LinkerHand(Node):
def __init__(self, name):
super().__init__(name)
# 声明参数(带默认值)
self.declare_parameter('hand_type', 'left')
self.declare_parameter('hand_joint', 'L6')
self.declare_parameter('is_touch', False)
self.declare_parameter('can', 'can0')
self.declare_parameter('modbus', "None")
# ros时间获取
self.stamp_clock = Clock()
# 获取参数值
self.hand_type = self.get_parameter('hand_type').value
self.hand_joint = self.get_parameter('hand_joint').value
self.is_touch = self.get_parameter('is_touch').value
self.can = self.get_parameter('can').value
self.modbus = self.get_parameter('modbus').value
self.sdk_v = 2
self.sleep_time = 0.005
self.cmd_lock = False
self.last_hand_post_cmd = None # 最新手指位置命令
self.last_hand_vel_cmd = None # 最新手指速度命令
self.last_hand_eff_cmd = None # 最新手指力矩命令
self.last_hand_state = [-1] * 10
self.last_hand_vel = [-1] * 10
self.force = [[-1] * 5] * 4
self.matrix_dic = {
"stamp":{
"sec": 0,
"nanosec": 0,
},
"thumb_matrix":[[-1] * 6 for _ in range(12)],
"index_matrix":[[-1] * 6 for _ in range(12)],
"middle_matrix":[[-1] * 6 for _ in range(12)],
"ring_matrix":[[-1] * 6 for _ in range(12)],
"little_matrix":[[-1] * 6 for _ in range(12)]
}
# 压感矩阵合值,单位g 克
self.matrix_mass_dic = {
"stamp":{
"secs": 0,
"nsecs": 0,
},
"thumb_mass":[-1],
"index_mass":[-1],
"middle_mass":[-1],
"ring_mass":[-1],
"little_mass":[-1]
}
self.last_hand_info = {
"version": [-1], # Dexterous hand version number
"hand_joint": self.hand_joint, # Dexterous hand joint type
"speed": [-1] * 10, # Current speed threshold of the dexterous hand
"current": [-1] * 10, # Current of the dexterous hand
"fault": [-1] * 10, # Current fault of the dexterous hand
"motor_temperature": [-1] * 10, # Current motor temperature of the dexterous hand
"torque": [-1] * 10, # Current torque of the dexterous hand
"is_touch":self.is_touch,
"touch_type": -1,
"finger_order": None # Finger motor order
}
self.version = []
self.touch_type = -1
self.hz = 1.0/60.0
self.hand_setting_sub = self.create_subscription(String,'/cb_hand_setting_cmd', self.hand_setting_cb, 10)
self._init_hand()
time.sleep(1)
self.run_count = 0 # 计数器,用于记录运行次数
self.timer = self.create_timer(0.01, self.run) # 100 Hz
self.thread_pub_state = threading.Thread(target=self.pub_state)
self.thread_pub_state.daemon = True
self.thread_pub_state.start()
def _init_hand(self):
self.api = LinkerHandApi(hand_type=self.hand_type, hand_joint=self.hand_joint,modbus=self.modbus,can=self.can)
time.sleep(0.1)
self.touch_type = self.api.get_touch_type()
self.hand_cmd_sub = self.create_subscription(JointState, f'/cb_{self.hand_type}_hand_control_cmd', self.hand_control_cb,10)
self.hand_state_pub = self.create_publisher(JointState, f'/cb_{self.hand_type}_hand_state',10)
self.hand_info_pub = self.create_publisher(String, f'/cb_{self.hand_type}_hand_info', 10)
if self.is_touch == True:
if self.touch_type > 1:
ColorMsg(msg=f"{self.hand_type} {self.hand_joint} Equipped with matrix pressure sensing", color='green')
self.matrix_touch_pub = self.create_publisher(String, f'/cb_{self.hand_type}_hand_matrix_touch', 10)
self.matrix_touch_pub_pc = self.create_publisher(PointCloud2, f'/cb_{self.hand_type}_hand_matrix_touch_pc', 10)
self.matrix_touch_mass_pub = self.create_publisher(String, f'/cb_{self.hand_type}_hand_matrix_touch_mass', 10)
elif self.touch_type != -1:
ColorMsg(msg=f"{self.hand_type} {self.hand_joint} Equipped with pressure sensor", color="green")
self.touch_pub = self.create_publisher(Float32MultiArray, f'/cb_{self.hand_type}_hand_force', 10)
else:
ColorMsg(msg=f"{self.hand_type} {self.hand_joint} Not equipped with any pressure sensors", color="red")
self.is_touch = False
self.embedded_version = self.api.get_embedded_version()
pose = None
torque = [200, 200, 200, 200, 200]
speed = [200, 250, 250, 250, 250]
if self.hand_joint.upper() == "O6" or self.hand_joint.upper() == "L6" or self.hand_joint.upper() == "L6P":
pose = [200, 255, 255, 255, 255, 180]
torque = [250, 250, 250, 250, 250, 250]
# O6 最大速度阈值
speed = [200, 250, 250, 250, 250, 250]
elif self.hand_joint == "L7":
# The data length of L7 is 7, reinitialize here
pose = [255, 200, 255, 255, 255, 255, 180]
torque = [250, 250, 250, 250, 250, 250, 250]
speed = [120, 250, 250, 250, 250, 250, 250]
elif self.hand_joint == "L10":
torque = [255] * 10
pose = [255, 200, 255, 255, 255, 255, 180, 180, 180, 41]
speed = [200, 250, 250, 250, 250, 250, 250, 250, 250, 250]
elif self.hand_joint == "L20":
pose = [255,255,255,255,255,255,10,100,180,240,245,255,255,255,255,255,255,255,255,255]
elif self.hand_joint == "L21":
pose = [75, 255, 255, 255, 255, 176, 97, 81, 114, 147, 202, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255]
elif self.hand_joint == "L25":
pose = [75, 255, 255, 255, 255, 176, 97, 81, 114, 147, 202, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255]
if pose is not None:
for i in range(1):
self.api.set_speed(speed=speed)
time.sleep(0.1)
self.api.set_torque(torque=torque)
time.sleep(0.1)
self.api.finger_move(pose=pose)
time.sleep(0.1)
def list_check(self,pose):
if isinstance(pose, list) == False:
return False
if len(self.last_hand_post_cmd) != len(pose):
return False
return any(abs(self.last_hand_post_cmd - pose) >= 3 for self.last_hand_post_cmd, pose in zip(self.last_hand_post_cmd, pose))
def hand_control_cb(self, msg):
if self.last_hand_post_cmd == None or self.list_check(msg.position) == True:
self.last_hand_post_cmd = msg.position
if self.last_hand_vel_cmd == None or self.list_check(msg.velocity) == True:
self.last_hand_vel_cmd = msg.velocity
if self.last_hand_eff_cmd == None or self.list_check(msg.effort) == True:
self.last_hand_eff_cmd = msg.effort
def run(self):
if self.sdk_v == 1:
self.sleep_time = 0.009
if self.hand_state_pub.get_subscription_count() > 0:
# 优先获取手指状态并且发布
self.last_hand_state = self.api.get_state()
time.sleep(0.003)
self.last_hand_vel = self.api.get_joint_speed()
time.sleep(0.002)
if self.cmd_lock == False:
if self.last_hand_post_cmd != None:
self.api.finger_move(pose=self.last_hand_post_cmd)
self.last_hand_post_cmd = None
if self.last_hand_vel_cmd != None:
vel = list(self.last_hand_vel_cmd)
if all(x == 0 for x in vel):
pass
else:
if (str(self.hand_joint).upper() == "O6" or str(self.hand_joint).upper() == "L6" or str(self.hand_joint).upper() == "L6P") and len(vel) == 6:
speed = vel
self.api.set_joint_speed(speed=speed)
elif self.hand_joint == "L7" and len(vel) == 7:
speed = vel
self.api.set_joint_speed(speed=speed)
elif self.hand_joint == "L10" and len(vel) == 10:
speed = [vel[0],vel[2],vel[3],vel[4],vel[5]]
self.api.set_joint_speed(speed=speed)
elif self.hand_joint == "L20" and len(vel) == 20:
speed = [vel[10],vel[1],vel[2],vel[3],vel[4]]
self.api.set_joint_speed(speed=speed)
elif self.hand_joint == "L21" and len(vel) == 25:
speed = vel
self.api.set_joint_speed(speed=speed)
elif self.hand_joint == "L25" and len(vel) == 25:
speed = vel
self.api.set_joint_speed(speed=speed)
self.last_hand_vel_cmd = None
time.sleep(0.003)
if self.run_count == 3 and self.is_touch == True and self.touch_type == 1 and self.touch_pub.get_subscription_count() > 0:
"""单点式压力传感器"""
self.force = self.api.get_force()
if self.is_touch == True and self.touch_type > 1 and (self.matrix_touch_pub.get_subscription_count() > 0 or self.matrix_touch_mass_pub.get_subscription_count() > 0 or self.matrix_touch_pub_pc.get_subscription_count() > 0):
"""矩阵式压力传感器"""
if self.run_count == 3:
self.matrix_dic["thumb_matrix"] = self.api.get_thumb_matrix_touch(sleep_time=self.sleep_time).tolist()
if self.run_count == 4:
self.matrix_dic["index_matrix"] = self.api.get_index_matrix_touch(sleep_time=self.sleep_time).tolist()
if self.run_count == 5:
self.matrix_dic["middle_matrix"] = self.api.get_middle_matrix_touch(sleep_time=self.sleep_time).tolist()
if self.run_count == 6:
self.matrix_dic["ring_matrix"] = self.api.get_ring_matrix_touch(sleep_time=self.sleep_time).tolist()
if self.run_count == 7:
self.matrix_dic["little_matrix"] = self.api.get_little_matrix_touch(sleep_time=self.sleep_time).tolist()
time.sleep(0.005)
if self.run_count == 8 and self.hand_info_pub.get_subscription_count() > 0:
"""手部信息"""
self.last_hand_info = {
"version": self.embedded_version, # Dexterous hand version number
"hand_joint": self.hand_joint, # Dexterous hand joint type
"speed": self.api.get_speed(), # Current speed threshold of the dexterous hand
"current": self.api.get_current(), # Current of the dexterous hand
"fault": self.api.get_fault(), # Current fault of the dexterous hand
"motor_temperature": self.api.get_temperature(), # Current motor temperature of the dexterous hand
"torque": self.api.get_torque(), # Current torque of the dexterous hand
"is_touch":self.is_touch,
"touch_type": self.touch_type,
"finger_order": self.api.get_finger_order() # Finger motor order
}
if self.run_count == 9:
self.run_count = 0
self.run_count += 1
time.sleep(0.003)
def pub_state(self):
while True:
if self.hand_state_pub.get_subscription_count() > 0:
msg = self.joint_state_msg(self.last_hand_state, self.last_hand_vel)
self.hand_state_pub.publish(msg)
if self.is_touch == True and self.touch_type == 1 and self.touch_pub.get_subscription_count() > 0:
msg = Float32MultiArray()
msg.data = [float(val) for sublist in self.force for val in sublist]
self.touch_pub.publish(msg)
if self.is_touch == True and self.touch_type > 1 and (self.matrix_touch_pub.get_subscription_count() > 0 or self.matrix_touch_mass_pub.get_subscription_count() > 0 or self.matrix_touch_pub_pc.get_subscription_count() > 0):
# 发布矩阵压感数据JSON格式
self.pub_matrix_dic()
# 发布矩阵压感和值JSON格式
self.pub_matrix_mass(dic=self.matrix_dic)
# 发布矩阵压感点云格式
self.pub_matrix_point_cloud()
if self.hand_info_pub.get_subscription_count() > 0:
msg = String()
msg.data = json.dumps(self.last_hand_info)
self.hand_info_pub.publish(msg)
time.sleep(self.hz)
def pub_matrix_mass(self, dic):
"""发布矩阵数据合值 单位g 克 JSON格式"""
msg = String()
# 获取当前的 ROS 时间
current_time = self.stamp_clock.now()
# 提取 secs 和 nsecs
t_secs = current_time.to_msg().sec
t_nsecs = current_time.to_msg().nanosec
self.matrix_mass_dic["stamp"]["secs"] = t_secs
self.matrix_mass_dic["stamp"]["nsecs"] = t_nsecs
self.matrix_mass_dic["unit"] = "g"
self.matrix_mass_dic["thumb_mass"] = sum(sum(row) for row in dic["thumb_matrix"])
self.matrix_mass_dic["index_mass"] = sum(sum(row) for row in dic["index_matrix"])
self.matrix_mass_dic["middle_mass"] = sum(sum(row) for row in dic["middle_matrix"])
self.matrix_mass_dic["ring_mass"] = sum(sum(row) for row in dic["ring_matrix"])
self.matrix_mass_dic["little_mass"] = sum(sum(row) for row in dic["little_matrix"])
msg.data = json.dumps(self.matrix_mass_dic)
self.matrix_touch_mass_pub.publish(msg)
def pub_matrix_point_cloud(self):
"""发布矩阵数据点云格式"""
tmp_dic = self.matrix_dic.copy()
del tmp_dic['stamp'] # 去掉时间戳字段
all_matrices = list(tmp_dic.values()) # 5 帧,每帧 6×12=72 个数
# 摊平到一维:360 个 float
flat_list = [v for frame in all_matrices for v in frame] # 360
flat = np.concatenate([np.asarray(np.clip(c, 0, 255), dtype=np.uint8) for c in flat_list])
fields = [PointField(
name='val',
offset=0,
datatype=PointField.UINT8,
count=1
)]
pc = PointCloud2()
pc.header.stamp = self.stamp_clock.now().to_msg()
pc.header.frame_id = ''
pc.height = 1
pc.width = flat.size # 360
pc.fields = fields
pc.is_bigendian = False
pc.point_step = 1 # 1 个 float32
pc.row_step = pc.point_step * pc.width
pc.data = flat.tobytes() # 1440 字节
self.matrix_touch_pub_pc.publish(pc)
def pub_matrix_dic(self):
"""发布矩阵数据JSON格式"""
msg = String()
# 获取当前的 ROS 时间
current_time = self.stamp_clock.now()
# 提取 secs 和 nsecs
t_secs = current_time.to_msg().sec
t_nsecs = current_time.to_msg().nanosec
self.matrix_dic["stamp"]["secs"] = t_secs
self.matrix_dic["stamp"]["nsecs"] = t_nsecs
msg.data = json.dumps(self.matrix_dic)
self.matrix_touch_pub.publish(msg)
def joint_state_msg(self, pose,vel=[]):
joint_state = JointState()
joint_state.header = Header()
joint_state.header.stamp = self.get_clock().now().to_msg()
joint_state.name = self.api.get_finger_order()
joint_state.position = [float(x) for x in pose]
if len(vel) > 1:
joint_state.velocity = [float(x) for x in vel]
else:
joint_state.velocity = [0.0] * len(pose)
joint_state.effort = [0.0] * len(pose)
return joint_state
def hand_setting_cb(self,msg):
'''控制命令回调'''
data = json.loads(msg.data)
print(f"Received setting command: {data['setting_cmd']}",flush=True)
try:
if data["params"]["hand_type"] == "left":
hand = self.api
hand_left = True
elif data["params"]["hand_type"] == "right":
hand = self.api
hand_right = True
else:
print("Please specify the hand part to be set",flush=True)
return
self.cmd_lock = True
# Set maximum torque
if data["setting_cmd"] == "set_max_torque_limits": # Set maximum torque
torque = list(data["params"]["torque"])
hand.set_torque(torque=torque)
if data["setting_cmd"] == "set_speed": # Set speed
if isinstance(data["params"]["speed"], list) == True:
speed = data["params"]["speed"]
hand.set_speed(speed=speed)
else:
ColorMsg(msg=f"Speed parameter error, speed must be a list", color="red")
if data["setting_cmd"] == "clear_faults": # Clear faults
if hand_left == True and self.hand_joint == "L10" :
ColorMsg(msg=f"L10 left hand cannot clear faults")
elif hand_right == True and self.hand_joint == "L10" :
ColorMsg(msg=f"L10 right hand cannot clear faults")
else:
hand.clear_faults()
if data["setting_cmd"] == "get_faults": # Get faults
f = hand.get_fault()
ColorMsg(msg=f"Get faults: {f}")
if data["setting_cmd"] == "electric_current": # Get current
ColorMsg(msg=f"Get current: {hand.get_current()}")
if data["setting_cmd"] == "set_electric_current": # Set current
if isinstance(data["params"]["current"], list) == True:
hand.set_current(data["params"]["current"])
if data["setting_cmd"] == "show_fun_table": # Get faults
f = hand.show_fun_table()
except:
print("命令参数错误")
self.cmd_lock = False
finally:
self.cmd_lock = False
def close_can(self):
self.api.open_can.close_can(can=self.can)
sys.exit(0)
def main(args=None):
try:
rclpy.init(args=args)
node = LinkerHand("linker_hand_sdk")
embedded_version = node.embedded_version
if len(embedded_version) == 3 or node.hand_joint.upper() == "O6" or node.hand_joint.upper() == "L6" or node.hand_joint.upper() == "G20":
ColorMsg(msg=f"New Matrix Touch For SDK V2", color="green")
node.sdk_v = 2
elif len(embedded_version) == 6 and node.hand_joint == "L10":
ColorMsg(msg=f"New Matrix Touch For SDK V2", color="green")
node.sdk_v = 2
elif len(embedded_version) > 4 and ((embedded_version[0]==10 and embedded_version[4]>35) or (embedded_version[0]==7 and embedded_version[4]>50) or (embedded_version[0] == 6)):
ColorMsg(msg=f"New Matrix Touch For SDK V2", color="green")
node.sdk_v = 2
else:
ColorMsg(msg=f"SDK V1", color="green")
node.sdk_v = 1
rclpy.spin(node) # 主循环,监听 ROS 回调
except KeyboardInterrupt:
print("收到 Ctrl+C,准备退出...")
finally:
# node.close_can() # 关闭 CAN 或其他硬件资源
# node.destroy_node() # 销毁 ROS 节点
# rclpy.shutdown() # 关闭 ROS
print("程序已退出。")
@@ -0,0 +1,73 @@
"""A CAN receipt, rather than a heartbeat, defines position sample identity."""
import json
from types import SimpleNamespace
from linker_hand_ros2_sdk.LinkerHand.utils.feedback import ReceivedPositionFrames
from linker_hand_ros2_sdk.linker_hand import LinkerHand
def test_cached_unchanged_and_out_of_order_frames_keep_measurement_identity():
frames = ReceivedPositionFrames({1: 2}, [(1, 0), (1, 1)])
assert frames.snapshot() is None
assert not frames.accept(1, [4], 1.)
assert not frames.accept(1, [4, 5], 0.)
assert frames.accept(1, [4, 5], 1.)
first = frames.snapshot()
assert first == frames.snapshot()
assert not frames.accept(1, [8, 9], .9)
assert not frames.accept(1, [8, 9], 1.)
assert frames.snapshot() == first
assert frames.accept(1, [4, 5], 2.) # A new stationary observation is real.
assert frames.snapshot().positions == first.positions
assert frames.snapshot().stamp_ns == 2_000_000_000
assert first.stamp_ns == 1_000_000_000
def test_split_finger_snapshot_does_not_retimestamp_held_channels():
frames = ReceivedPositionFrames({1: 2, 2: 2}, [(2, 1), (1, 0), None])
frames.accept(1, [10, 11], 1.)
assert frames.snapshot() is None
frames.accept(2, [20, 21], 1.02)
first = frames.snapshot()
assert first.positions == (21, 10, 0)
assert first.channel_stamps_ns == (1_020_000_000, 1_000_000_000, 1_000_000_000)
frames.accept(2, [20, 22], 1.04)
assert frames.snapshot().channel_stamps_ns == (1_040_000_000, 1_000_000_000, 1_000_000_000)
def test_sdk_republication_cannot_refresh_cached_state_timestamps():
frames = ReceivedPositionFrames({1: 2}, [(1, 0), (1, 1)])
joint_messages, timed_messages = [], []
publisher = lambda messages: SimpleNamespace(get_subscription_count=lambda: 1, publish=messages.append)
host = SimpleNamespace(api=SimpleNamespace(supports_feedback_timestamps=True,
get_feedback_snapshot=frames.snapshot, get_finger_order=lambda: ['a', 'b']),
_last_timed_feedback_stamps=None, last_hand_vel=[], hand_state_pub=publisher(joint_messages),
timed_state_pub=publisher(timed_messages))
host.joint_state_msg = lambda *args, **kwargs: LinkerHand.joint_state_msg(host, *args, **kwargs)
LinkerHand._publish_feedback(host)
assert not joint_messages and not timed_messages
frames.accept(1, [10, 20], 100.)
LinkerHand._publish_feedback(host)
LinkerHand._publish_feedback(host)
assert len(joint_messages) == 2 and len(timed_messages) == 1
assert all(message.header.stamp.sec == 100 for message in joint_messages)
assert json.loads(timed_messages[0].data)['channel_stamps_ns'] == [100_000_000_000]*2
frames.accept(1, [10, 20], 101.)
LinkerHand._publish_feedback(host)
assert len(timed_messages) == 2 and joint_messages[-1].header.stamp.sec == 101
def test_g20_timed_layout_matches_the_public_sdk_order():
from linker_hand_ros2_sdk.LinkerHand.core.can.linker_hand_g20_can import LinkerHandG20Can
hand = LinkerHandG20Can.__new__(LinkerHandG20Can)
layout = hand.joint_state_to_cmd_state([[(frame, index) for index in range(6)] for frame in range(0x41, 0x46)])
frames = ReceivedPositionFrames({frame: 6 for frame in range(0x41, 0x46)}, [item or None for item in layout])
raw = []
for finger, frame in enumerate(range(0x41, 0x46)):
raw.append([finger*10+i for i in range(6)])
frames.accept(frame, raw[-1], 100.+finger*.01)
snapshot = frames.snapshot()
assert snapshot.positions == tuple(hand.joint_state_to_cmd_state(raw))
assert snapshot.channel_stamps_ns[0] == 100_000_000_000
assert snapshot.channel_stamps_ns[4] == 100_040_000_000
@@ -0,0 +1,75 @@
import ast
from collections import deque
from pathlib import Path
import sys
import threading
import time
from types import SimpleNamespace
PACKAGE = Path(__file__).resolve().parents[1] / "linker_hand_ros2_sdk/LinkerHand/core"
EXPECTED = [
"thumb_cmc_pitch",
"thumb_cmc_roll",
"index_mcp_pitch",
"middle_mcp_pitch",
"ring_mcp_pitch",
"pinky_mcp_pitch",
]
def _finger_order(path: Path, class_name: str) -> list[str]:
module = ast.parse(path.read_text(encoding="utf-8"))
selected = next(
item
for item in module.body
if isinstance(item, ast.ClassDef) and item.name == class_name
)
method = next(
item
for item in selected.body
if isinstance(item, ast.FunctionDef) and item.name == "get_finger_order"
)
returned = next(item for item in method.body if isinstance(item, ast.Return))
return ast.literal_eval(returned.value)
def test_l6_can_and_rs485_publish_the_same_physical_channel_order() -> None:
assert _finger_order(PACKAGE / "can/linker_hand_l6_can.py", "LinkerHandL6Can") == EXPECTED
assert _finger_order(
PACKAGE / "rs485/linker_hand_l6_rs485.py", "LinkerHandL6RS485"
) == EXPECTED
def test_l6_can_position_echo_does_not_replace_measured_feedback() -> None:
linker_hand_root = PACKAGE.parent
sys.path.insert(0, str(linker_hand_root))
try:
from core.can.linker_hand_l6_can import LinkerHandL6Can
from utils.feedback import ReceivedPositionFrames
finally:
sys.path.remove(str(linker_hand_root))
hand = LinkerHandL6Can.__new__(LinkerHandL6Can)
hand.can_id = 0x27
hand.x01 = [10, 20, 30, 40, 50, 60]
hand.position_feedback = ReceivedPositionFrames({0x01: 6}, [(0x01, i) for i in range(6)])
hand._position_echo_lock = threading.Lock()
command = (255, 2, 253, 253, 253, 253)
hand._pending_position_echoes = deque(
[(time.monotonic(), command)], maxlen=32
)
hand._position_echo_timeout_seconds = 0.02
hand.process_response(
SimpleNamespace(arbitration_id=0x27, data=bytes((0x01, *command)), timestamp=100.)
)
assert hand.x01 == [10, 20, 30, 40, 50, 60]
assert not hand._pending_position_echoes
measured = (250, 3, 252, 252, 252, 252)
hand.process_response(
SimpleNamespace(arbitration_id=0x27, data=bytes((0x01, *measured)), timestamp=101.)
)
assert hand.x01 == list(measured)
assert hand.get_feedback_snapshot().channel_stamps_ns == (101_000_000_000,) * 6
@@ -4,7 +4,9 @@ from linker_hand_ros2_sdk.linker_hand import (
COMMAND_QOS,
LinkerHand,
command_changed,
position_command_should_queue,
state_poll_due,
state_reads_deferred,
)
@@ -30,7 +32,25 @@ def test_identical_commands_are_not_reapplied():
assert not command_changed([60, 60], [])
def test_legacy_heartbeat_resends_unchanged_position_target():
assert position_command_should_queue([60, 60], [60, 60], repeat=True)
assert not position_command_should_queue(
[60, 60], [60, 60], repeat=False
)
assert position_command_should_queue(
[60, 60], [60, 61], repeat=False
)
assert not position_command_should_queue([60, 60], [], repeat=True)
def test_state_polling_is_throttled_without_missing_deadline():
assert state_poll_due(None, 10.0, 0.1)
assert not state_poll_due(10.0, 10.09, 0.1)
assert state_poll_due(10.0, 10.1, 0.1)
def test_state_reads_yield_to_recent_motion_then_resume():
assert state_reads_deferred(10.0, 10.1, 0.2)
assert not state_reads_deferred(10.0, 10.2, 0.2)
assert not state_reads_deferred(None, 10.1, 0.2)
assert not state_reads_deferred(10.0, 10.1, 0.2, enabled=False)
Submodule src/linkerhand-l30-sdk added at 0103fc55f9
@@ -0,0 +1,169 @@
# 统一标定需求代码审查(2026-09-10)
## 结论与审查范围
审查当前工作区(包含未提交的统一重构、精简 JSON v2 和本会话启动参数修复),
不是只审查 Git HEAD,也没有把历史 PASS 当作当前实机验收。
通用采集、拟合、URDF 修正、最终文件验收和发布架构已经存在,但仍有迁移残留和精度验证缺口。
本轮修正可由代码及隔离测试确认的问题,没有改变原始 CAD、机械限位、SDK、贴 Tag 配置或相机外参。
没有启动 SDK、相机或发送机械手运动指令。
## 已修复的问题
| 问题 | 影响 | 修改 |
| --- | --- | --- |
| `vendor_sdk_config:=` 等空 launch 实参 | O6/L6/G20 可能在节点启动前退出 | runner 省略空参数,使用 ROS CLI 的实际解析器覆盖四型号 |
| 在线 IO 使用固定 front/side/top 外参加载器 | 新 Profile 即使静态校验通过,在线仍可能不能使用其他机位 | 直接用通用外参加载器,视图和参考机位来自 Profile |
| launch 固定三机位、型号文件名回退和过期参数 | 新型号需修改启动代码;部分开关没有实际作用 | 按 Profile 创建机位;要求受保护 Profile/原始 URDF/Tag 路径;移除旧诊断、速度、scope 和输出目录覆盖参数 |
| MVS 无图像时只显示“等待设备” | 操作者无法区分 SDK、相机或检测链路故障 | 统一显示每机位有效内参/检测消息等待项、SDK 条件和内参不匹配原因 |
| READY 后设备断流仍保留可开始状态 | 启动请求可能使用已失效的就绪条件 | 开始前可撤销 READY;Start 再次检查完整设备条件 |
| 实时 CameraInfo 只比较 P 的部分字段 | K、D、R 改变但 P 未变时,整流图可能与原外参不一致 | 实时校验宽高及 K/D/R/P 指纹;开始后不匹配按原有坐标变化故障停止继续采集 |
| 每个稳态采样点清空整方向统计 | 显示有效帧和覆盖率归零,误导进度判断 | 分离方向数据与当前稳态点数据,重扫/换方向才重置方向统计 |
| 标题只显示序列号 | 序列号不含型号时无法确认当前型号 | 所有型号统一显示 model、side、serial_number |
| 正式 finalizer 仍允许旧格式绕过指令拟合 | 新型号配置旧版本可能没有所需的 SDK 指令→rad 产物 | 正式 Profile 与 finalizer 均只接受统一输出版本 2,取消旧输出分派 |
| 完成回调读取精简 JSON 已删除的 `quality` | 文件可能已经发布,但节点在结束时抛 KeyError | 删除没有消费者的 `final_quality` 赋值,增加无 quality 字段的完成回调回归 |
| 视觉暂停与控制定时器之间的状态竞争 | 定时器等待锁期间已暂停,取得锁后仍可能继续动作 | 在控制锁内再次检查终止状态 |
开始前等待相机消息的 2 秒窗口不用于扫描中的实时暂停;空检测数组算视觉链路存活。
短时 Tag 丢失、低检测率等仍按方向结束的数据质量策略处理,同速重扫一次后仍不足才暂停。
清理的在线残留包括无人读取的回调计数、命令频率缓存/方法、型号质量标签、旧 tracker reset hook
以及每次反馈回调不必要构造的备用 Adapter。没有删除 SDK 协议接口的抽象方法。
五个过渡期写出器已移出生产包并保留原文件:
`calibration_output/code_review_retired_serializers_umhxlepj/`
包括 `select.py``generic_v1.py``legacy_v4.py``legacy_v6.py``native_v7.py`
历史读取器和历史诊断工具仍在;此归档不包含或改写用户的历史标定产物。
## 视频复核后的补充修复:共用姿态分支筛选
视频复核发现旧配置把 `pnp_reprojection_tie_px` 和姿态拒绝阈值均设为 `1.5 px`
导致通过质量门槛的 IPPE 候选都参加时间连续性比较,旧镜像分支可能压过明确更好的图像拟合。
用现有候选姿态复现时,旧分支 `0.24 px` 会压过另一分支 `0.05 px`
公共模块统一默认近似同误差阈值为 `0.03 px`ROS 与离线采集继承该值。
五份旧 YAML 的重复覆盖已移除,四产品的 `calibration_config_sha256` 已同步更新。
新增校验拒绝非有限、负值或不小于图像拒绝阈值的分支容差,防止旧的 `1.5/1.5` 配置再次生效。
保持实际近似同误差时的连续性选择、原有倾角/图像质量限制和方向结束重扫策略。
因此前文“配置保持不变”仅描述首轮结构审查;这次补充修复修改了上述算法配置及其哈希。
覆盖五份配置和四型号默认采集的九个复现用例,在修复前全部失败,修复后全部通过。
第一组姿态、配置、采集与加载实际参数文件的四型号隔离 ROS host 检查共 51 项通过。
第二组公共采集/拟合与发布、O12 姿态解析、来源保护、四型号配置及启动检查共 141 项通过,
耗时 114.53 秒。两组为本次修改的定向回归,没有宣称本次又执行过全包测试。
本次 `colcon build --packages-select linkerhand_calibration --symlink-install` 通过;
安装环境读取四产品配置、校验受保护哈希及公共默认值均通过,`git diff --check` 通过。
这不代表所有平面双解已消除,也不替代下述相机时序和实机重复性验证。
## 仍不能由本轮软件审查保证的事项
### P1:真实采集时间与传输积压
`hikrobot_camera.py::_publish_frame` 使用主机取到帧时的 ROS 时间戳,未将设备曝光时间映射到统一时钟。
SDK 帧结构有设备时间戳和帧号,但设备时钟单位、偏移、漂移与主机时钟同步还没有在当前硬件验证。
`runtime/ros/io.py::_state_callback` 对无时间戳反馈也采用接收时间。
因此,USB/驱动缓存可能把旧图像与新反馈配对;50 ms 的消息时间戳检查和“只取最新帧”不能排除该问题。
本会话已经实测到三台相机共享 480 Mbps Hub 上行,且正面/上方取帧超时。
后续应在恢复 USB 3.x 链路后,核验设备时间戳、帧号、曝光与反馈时序,并用独立运动数据检查配对误差。
不能猜测设备时钟单位、直接给时间戳减一个常数,或放宽质量阈值作为修复。
### P1:重复性和实机/仿真验收尚未完成
前三轮训练、第四轮独立验证、冻结 Tag 安装和双文件重读校验已接入。
但同手多次独立重采、重新摆放和重新贴 Tag、不同保持姿态/多关节组合动作的实际验收仍缺证据。
L6/O6 部分关节由小指参数迁移,不属于所有关节独立实测。
发布报告仍应如实保留 `arbitrary_multiaxis_validated: false`
外部仿真要加载本次修正 URDF,用同一 SDK 指令话题及配套 JSON 驱动;用实机反馈直接驱动仿真,
只能显示反馈对应姿态,不能证明指令映射正确。桥本身是消息转换器,不是独立精度测量工具。
### P2:数学模块和旧配置仍有维护负担
`core/fitting/spatial.py` 约 3,656 行,包含轴线拟合、基座位姿、零位优化、可观测性和验证,
其中 `solve_urdf_zero_offsets` 从约第 1,921 行开始,嵌套函数及共享局部状态较多。
它是当前生产算法,不是可直接删除的废代码。后续拆分应按这些数学职责进行,
用固定输入的参数、残差、失败原因和产物回归验证等价性,不应复制成各型号独立算法。
当前约 58 个兼容模块仍用于旧导入路径、历史回放/迁移工具和旧回归;不能仅因目录名称或静态引用少就删除。
几行的公开兼容导出不是第二套拟合算法。旧型号诊断、旧 URDF plan 和部分旧 YAML 参数仍需明确退休范围后继续收缩。
受保护的历史配置在本轮保持不变;旧 YAML 中一些已不被在线节点读取的参数,不应当作当前算法的生效开关。
原始 G20/O12 的 mimic/限位冲突和 CAD 保留策略仍需机械资料确认;当前通用 URDF 修正
主要处理零位、限位和线性 mimic,不估计连杆长度、轴位置、mesh 或惯量。
## 按需求核对
| 需求 | 当前判断 |
| --- | --- |
| 多型号共用拟合和 URDF 修正 | 已实现统一生产链;SDK 新协议仍需 Adapter 和 ROS 启动绑定 |
| 新型号只关心 SDK、贴 Tag、避让 | 基本成立,但必须声明关节绑定、可观测性、迁移/保留策略和可修改字段,不能从未知安装的单轴 Tag 自动猜出 CAD 零位 |
| 统一 SDK 指令→rad JSON 与修正 URDF | 正式仅生成统一 v2 指令表;u8 为 256 项,rad 为显式节点;被动表由标准 mimic 推导并验收,详细证据保留在报告 |
| 开始前可移动、开始后固定 | 每次 Start 丢弃预览参考,正式锁定;可见固定基准有漂移监测;遮挡对象是否移动不能实时保证 |
| 正确、可重复 | 有独立验证和来源保护,但仍受采集时序与未完成实机验收限制 |
| 少暂停且原因明确 | 统一策略保留;开始前明确缺失设备,不用短时视觉丢失实时打断扫描 |
| 统一进度且明确型号 | 已修正型号、设备等待原因及方向统计显示 |
| 同源指令实机/仿真验证 | 有统一转换桥;实际动态/组合动作对比尚未完成 |
## 本轮验证
- 审查中完整标定包回归:**544 通过、1 跳过,383.25 秒**。跳过项要求显式提供
`O12_REPLAY_RAW` / `O12_REPLAY_REFERENCE`,没有用合成数据替代历史实测。
- 随后补充的启动清理、单一正式输出及归档改动:启动/runner/正式 Profile 拟合/精简产物定向组
**74 通过,108.78 秒**,包含四型号与虚拟型号的实际公共拟合和发布路径。
- 最后完成回调、暂停竞争、ROS host、等待诊断、实时内参保护:**25 通过,0.98 秒**。
与上一组有重叠,不相加冒充一次全量结果;没有宣称最后所有变更又跑过一次全量。
- `colcon build --packages-select linkerhand_calibration --symlink-install` 通过。
安装入口读取四产品的受保护配置、ROS launch 实参解析均通过;167 个生产/兼容 Python 文件 AST 通过;
`git diff --check` 通过。
- 新 launch 测试实际构造四型号及改名机位的 launch actions,但不执行这些动作。
隔离 ROS 节点/标准加载器测试不启动 SDK 或相机,不代表实机验收。
## 数学核心与 ROS 组合重构的验收记录
本轮以包含公共 `0.03 px` 姿态分支修复的工作区为基线,只拆分职责、明确数据和并发边界。
没有删除基线中仍存在的历史入口,没有修改配置、拟合公式、优化初值/顺序、残差权重或验收阈值。
- `core/fitting/spatial.py` 保留 45 个原定义符号的显式导出;实际计算在 `spatial_solver/`
`solve.py` 的入口顺序为输入准备、训练求解、统计、独立验证及结果装配。
数据通过 `ZeroProblem``TrainingProblem``TrainingFit` 和各验收结果类型传递。
训练接口只接收训练观测,训练几何中的掌部姿态按训练轮筛选。
- `core/geometry/pnp.py` 保留 14 个原定义符号;实现拆为 IPPE、单 Tag 跟踪、刚性组和轨迹选择。
跟踪默认值集中在 `tag_pose/parameters.py`,ROS 参数只负责加载和 deg/rad 转换。
- `UnifiedCalibrationNode` 直接继承 ROS `Node`,组合消息 IO 与 `CalibrationCoordinator`
协调器不加载 ROS 库,复用 `SessionExecution`、采集、运动、固定基准和断点组件。
运行参数、观测输入和快照使用明确类型;阶段只有 `CalibrationSession.phase` 一个来源。
- `FinalizationController` 组合原 worker 与发布器,协调器接收阶段事件并授权提交。
状态锁先于 worker 锁;PnP 在状态锁外计算,提交时检查会话版本、采集/运动对象及稳态边界。
暂停、中止与提交互斥;失败不触发自动恢复运动,成功只提交一次。
SDK 绑定显式接收反馈新鲜度、时钟、健康订阅和发布接口。
固定输入记录在本次工作区快照 `/tmp/calibration_structure_baseline_30qf3ozo/`
持久化的对比报告及压缩前后输出位于 `calibration_output/structure_refactor_20260910/`
其中 `comparison.json` 记录输入摘要、差异、产物摘要和拒绝原因,`protected_inputs.json` 保存输入哈希。
这些记录是合成数据的结构等价证据,不是实机精度验收。
| 固定输入 | 比较浮点值数量 | 最大绝对差异 | JSON/拟合结果/阶段事件及 URDF 文本 |
| --- | ---: | ---: | --- |
| G20 | 46,046 | 0 | 一致 |
| L6 | 15,304 | 0 | 一致 |
| O6 | 15,304 | 0 | 一致 |
| O12 | 14,494 | 0 | 一致 |
| 虚拟型号(任务重排) | 15,304 | 0 | 一致 |
对五组输入分别移除第四轮或注入第四轮 Tag 姿态滑移,10 个拒绝原因和阶段事件与基线一致。
167 个配置、Profile、原始 URDF/资源文件的哈希不变。
另一个虚拟型号同时重排 SDK 通道、机位名称和扫描任务,经过真实公共采集、finalizer、
最终 JSON/URDF 验收和发布,再检查同一 SDK 指令经过模拟传输、查表和标准 mimic 图的一致性。
验证结果:
- 整个标定包:581 passed1 skipped337.43 s。
- SDK 绑定时钟补充后,协调器、四型号 ROS 构造和安全策略回归:27 passed。
- 跳过项是 `test_o12_recorded_replay.py`:未提供 `O12_REPLAY_RAW` 与独立参考模型;不计作实机通过。
- 改动模块通过 Pyflakes`git diff --check` 通过。
- `colcon build --packages-select linkerhand_calibration --symlink-install` 成功。
- 安装后的 12 个 console entry 可导入;两个产品 CLI 的 `--help` 正常。
两个 ROS 节点入口均通过四型号受保护 Profile 的转交检查;安装环境下四型号 ROS 构造/Start 回归 4 passed。
本轮不处理相机采集时刻同步,也不替代实机多次标定和同话题实机/仿真运动对照。
@@ -0,0 +1,937 @@
# 四轮采集与图像拟合的修改前评估
日期:2026-09-16。目标:稳定完成采集,从可信标定 JSON 重建修正 URDF。
## 当前结论(2026-09-16
当前 O6 按重新实测的 16.5 mm Tag 完成新采集,完整数据经正式离线流程验收通过,
已生成合格 JSON、修正 URDF 和发布清单。本次在线收尾发生系统内存耗尽,不能把离线
成功表述为在线全过程无中断。已修正收尾读取原始日志时的整文件副本开销,独立验证见后文。
没有增加 O6 专用容错分支或降低物理精度门限,也不据此宣称其他型号实机均已通过。
| 环节 | 当前证据与结论 |
| --- | --- |
| 四轮连续采集与必要稳态点 | 新会话 `20260916_105528` 完成 36/36,1 个小指方向自动重扫后通过;采集约 6 分 47 秒。随后收尾时发生 OOM,反馈超时暂停;完整数据已用于正式离线验收。此前两次 16 mm 会话也完成采集,但空间验收失败。 |
| 反馈时间 | 使用真实 CAN 接收时间;相同缓存不得伪装成新测量。逐通道时间保留至图像插值。 |
| 相机时间 | 三台实际设备的曝光事件与帧计数器吻合;设备锁存换算曝光中点,替换约晚 16.5 ms 的取帧时间。 |
| 调度 | ROS 控制/反馈顺序处理,图像独立有界队列;实机对照消除了共享锁造成的反馈排队。 |
| 映射输入 | 当前报告的分段线性反馈曲线保留曝光时刻的小数输入;实际字节命令与旧整数表保持量化。 |
| 图像证据 | 同一运动保存全部机位的原始角点,不依赖任务主机位或 PnP;最新会话保存 16,822 个图像记录。没有增加运动或停点。 |
| 图像验收 | 独立三维测量对原始角点仍须 ≤1.5 px;最终文件按原角度/位置门限检查,额外视角只提供物理容差内的图像相容性证据。已知真值反例和反向拒绝测试通过。 |
| 平行转轴零位 | 已修正参考点沿轴移动导致零位变化的公共公式;四型号源 URDF 的已知真值及任意轴上参考点检查通过。数学规则不再由型号开关选择。 |
| JSON → URDF | `20260916_105528/offline_165mm` 已生成通过验收的文件与 manifest;另外调用重建 CLI,仅从保存 JSON 和源 URDF 得到的文件与发布 URDF 逐字节相同。 |
| 整链空间精度 | 新 16.5 mm 会话已通过原门限。主观测运动/稳态独立验收中,各角色最差角度 P95 为 1.873°,位置 P95 为 2.122 mm;最终文件空间与图像检查通过。精度适用范围是观测关节、扫描方向与固定的其他关节姿态;未独立观测的关节仍按配置迁移。 |
| 尺度输入核验 | 用户实测更正为 **16.5×16.5 mm** Tag、**27×27 mm** 棋盘,与独立图像检查基本吻合。当前 O6 的 Profile、检测器和标定节点尺寸已同步并记录新指纹;未修改相机参数,未将旧 16 mm 数据冒充新尺寸采集。 |
先前的分机位空间不一致已在修正实测尺寸后的新数据上通过验收,未靠增加拟合参数
或放宽门限达成。当前稳定性问题集中在已确认的收尾内存峰值;保留完整采集数据直接
验证收尾修复,不为重复计算重新运动机械手。其余型号仍需对应硬件实测,O6 当前 Tag
布局中的其他三指曲线迁移不等于各关节独立标定。
上述两次未验收候选的三个主动零位差均小于 0.03°,说明当前分段求解可重复,
不能据此证明其绝对空间精度。正式发布仍必须通过整链与最终文件验收。
以下按发生顺序保留评估依据;早期 CAN 未连接等现场状态已由上述新实采结果更新。
## 已核实的问题
1. 当前生产采集要求准备运动先确定唯一 PnP/铰链姿态;不唯一时整场暂停,后续三轮训练数据尚未获得。
2. 准备阶段的自由铰链几何与最终保留骨长、轴向的 URDF 使用不同参数空间。
3. 后续姿态已受准备模型约束,不能把其重复性直接解释为原始图像上的整链正确性。
4. JSON 保存后读取并重建 URDF 已有独立字节一致性验证,应保留这条文件链。
## 修改前的离线实验
没有改动生产算法、现场硬件、原始采样或原始 URDF。以 `20260915_150015` 保存的原始
角点进行诊断:前三轮抽取 480 个 Tag 观测,第四轮 160 个观测仅用于冻结后的检查。
直接按源 URDF 的运动链拟合一个共同基座、三个允许修改的零位和每个 Tag 的固定安装;
另一个实验允许带固定基准的双向曲线修正。两个实验均使用记录时的相机参数。
- 固定曲线:39 个参数,32 次迭代收敛,约 0.60 秒。
- 联合曲线:119 个参数,23 次迭代收敛,约 1.90 秒,曲线最大修正约 1.38°。
- 带 0.05 px 噪声的已知真值实验,三个初值均收敛,三个零位最大误差约 0.023°。
- 实采第四轮仍有不合格残差;小指末节最大约 4.15 px。上述实验不是实机通过证据。
- 使用后来核验的相机参数进行的私有反事实诊断也没有消除全部残差,不能将旧数据改挂新参数发布。
脚本、参数、雅可比和报告位于
`calibration_output/O6_RIGHT_001/20260915_150015/review/architecture_assessment/`
原型只证明统一参数空间可以计算、在所测合成条件下可恢复;没有证明全型号实机精度。
## 采用的边界
- 采集保存原始角点、相机身份、SDK 指令/反馈、方向、轮次与基准图像;固定基准漂移、
SDK 故障、时间同步和必要采样覆盖检查仍然有效。
- 稳定基准图像先冻结,三维解释后求解;未求解的图像不得标成已测准的角度或零位。
- 训练只使用前三轮;第四轮与独立稳态验证不参与几何、安装、曲线或候选的选择。
- 图像拟合使用与输出 URDF 一致的拓扑和可修改参数;不自动改变骨长、转轴、相机参数,
不新增 CAD 零位假设。显式复制关节继续保留来源声明。
- 像素残差小不是角度/位置精度合格的充分条件。原物理门限、参数可确定性和最终文件读取验证需保留。
- 模型尚未通过回归与实测验收前,不凭局部收敛结果替换正式发布逻辑,不降低门限以生成 PASS。
- 采集完成后的拟合失败应保留完整数据并支持离线重算,区别于运动中暂停。
## 第二轮评估:不采用未经验证的整体替换
对同一会话进一步测试直接从原始角点初始化、共同拟合 33 节点双向单调曲线的版本。
共 359 个参数、6160 个训练残差,150 次迭代、43.39 秒后仍未收敛;拇指末节
第四轮最大误差约 45.18 px,小指末节约 7.77 px。这个结果说明,前一个实验依赖
既有拟合初值的局部收敛,不足以证明新流程具备独立初始化和实机鲁棒性。
因此该原型**未接入生产**,相关采集/基准/拟合草稿已移出程序源码,连同源代码快照保存到
上述评估目录的 `unadopted_prototype/``prototype_source_snapshot.tar.gz`
`monotonic_urdf_image_replay.json` 保留失败数据。下一步不能通过增加迭代次数、减小残差权重
或放宽门限将其直接上线。
经过四型号 FK 对照测试的投影基础 `core/fitting/urdf_image.py` 和对应测试也归档到
`unadopted_prototype/`,便于复查。生产源码不保留未接入的备用算法。
## 本次实际接入的流程修改
1. 四轮及稳态采集完成、数据同步落盘后,显式关闭图像采集,撤销尚未返回的图像回调。
拟合使用已经保存的图像与相机参数;此后相机消息/Tag 位置变化不再改变这份输入或触发
采集中暂停。运动采集期间的固定基准漂移检查保持生效。
2. 反馈接收、SDK 故障保护和用户中止在后台拟合期间持续生效。
3. 数值拟合、文件验收和后台阶段协议错误统一结束为 `FAILED`,保存失败阶段、原始数据路径、
完成/计划采集单元数和异常堆栈;不再写成采集中 `PAUSED`
4. 状态输出始终提供原始数据路径,拟合失败提供诊断路径,提示先离线复算。
失败结果不会填写已通过的 JSON/URDF 路径,也不会被迟到观测或清理用中止覆盖。
5. 修复反馈曲线整链回放中丢失非活动关节到达方向的问题。比如下游关节正在弯曲,
上游关节保持不动时,上游仍应使用其到达方向对应的回差曲线,不能自动改用双向均值。
反馈回放和稳态指令回放现在使用同一个方向解析函数;记录了不完整或非法方向时明确拒绝。
独立两轴合成数据证明:旧处理会使固定 Tag 看起来有超过 4° 的安装误差,修复后
训练和第四轮最终 URDF 回放恢复真值。当前旧 O6 数据的对应影响小于 0.03°,不是
该次实采大误差的主要原因。数值见 `held_joint_direction_effect.json`
6. 用户确认 O6 顶部 ID7 与正面 ID2 固定在同一末节后,明确设置其安装连杆为
`rh_thumb_distal`,不再从侧摆任务名称推断为侧摆根部连杆。Profile 哈希同步更新;
旧会话的配置身份保持原样。独立合成场景中的 O6 安装真值同步修正,其他型号不变。
单独侧摆且弯曲角固定时,错误连杆可以被固定安装矩阵吸收,因此不能声称该错误
一定解释先前全部残差;必须在正确绑定及完整关节状态下继续验证。
这组流程修改解决采集完成后继续受实时视觉影响和失败状态误导的问题;准备段的几何
不一致另按下面的评估处理。保持原文件生成与发布验收链路。
## 第三轮评估:使用输出 URDF 已保留的几何约束
源 URDF 中相邻平行转轴的平行关系及垂直轴距,在任意关节零位修正下均不变。
旧准备模型却允许这些量自由变化,求出的几何无法由最终保留骨长和轴向的 URDF 表达。
因此只对真实 Tag 安装连杆、观测拓扑和源 URDF 均满足条件的相邻平行轴,消去两个
自由轴倾角和一个自由轴距参数。相机、Tag 安装与逐帧角度仍由图像估计;不新增
CAD 零位约定、不使用 SDK 角度作为拟合先验、不引入额外惩罚权重。
适用性由共同编译器判断,没有 O6 专用分支;不满足条件时保留原图像模型。
冻结模型记录新的策略版本、约束及源 URDF 哈希,最终处理前从源文件重新推导验证。
旧模型的序列化字节结构保持不变,避免破坏旧证据哈希。内存记录与磁盘记录均保留会话头。
断点策略升级为 `confirmed_branch_v6_source_geometry`;旧自由几何会话仍可离线读取,
不能作为新约束流程的断点混入新采集。
`20260915_160040` 保存的小指准备段 104 帧原始角点:
- 原自由模型返回 `image_motion_families_not_distinguishable`
- 使用源平行关系与 36.9995936 mm 轴距后,四个候选均在原 150 次迭代上限内收敛。
- 最优两个候选收敛到相同完整 Tag 姿态,其他候选仍用原统计检验排除。
- 最优候选留出帧最大误差约 0.696 px,低于原 1.5 px 门限;正式求解器约 4.69 秒返回成功。
- 原像素、角度/轴心不确定性、候选等价性及显著性门限均未放宽。
独立合成图像验证同向及反向平行轴的真值恢复;四型号源几何验证约束在独立零位变化下
仍成立。这证明该准备段可以使用与输出一致的模型解算,**尚不等于新四轮整手实机通过**。
结果保存在 `20260915_160040/review/cad_geometry_resolution.json` 及相邻诊断文件。
随后经实际 `BranchInitialization` 入口回放该会话三个准备段的原始图像,拇指侧摆
98 帧、拇指弯曲 88 帧、小指 104 帧均通过,耗时约 0.45 / 3.78 / 4.94 秒。
该回放使用用户已确认的当前 Tag 绑定,仅保存诊断;没有改写旧会话或发布新标定。
`all_preparations_source_geometry.json` 和同目录可复现脚本。
## 验证顺序
1. 受约束图像模型的已知真值、多初值、旧证据兼容和错误几何反例。
2. 四型号共同采集逻辑与数据身份;反馈/固定基准保护及中止持续生效。
3. 前三轮训练身份与第四轮隔离;完整性、重复图像、篡改与旧会话身份检查。
4. JSON 落盘后独立重建 URDF、标准 ROS 加载及最终文件验收。
5. O6 当前相机配置下重新实采;只有通过实际验收才更新发布结果。
## 本轮验证结果及实机阻塞
- 最终几何、真实图像入口、采集边界、来源身份、JSON 独立重建回归:87 项通过(39.81 秒)。
- 后台进程、诊断采集、四型号产物与 launch 相关集成检查:80 项通过(157.08 秒)。
- 新旧断点策略、恢复与原始图像凭据检查:92 项通过(4.85 秒)。
各组有交叉,不将数量相加作为独立测试总数。
- 已知真值覆盖两/三关节链和正/反向平行轴;故意把 37 mm 轴距改为 80 mm 时,
求解器因原重投影门限不合格而拒绝,不能仅凭源几何声明强行通过。
- `colcon build --packages-select linkerhand_calibration --symlink-install` 通过。
- 用户确认现场为 O6 后,清理了仍订阅同一控制话题的 O30 残留 SDK/GUI。
实际执行新的 `calibrate_hand --no-resume` 命令时,启动检查报告 `CAN interface does not exist: can0`
系统仅有 lo/eno1/wlo1;采集未开始、未发送电机指令、没有发布本轮新标定。
需恢复 O6 原通信适配器后继续四轮实测;当前不能宣称整手实机精度或全部型号实机通过。
启动报告保存在上述 review 目录的 `o6_cad_geometry_startup.json` 和日志中。
## 第四轮评估:有限恢复、精简停点与最终原始角点验收
本次优化以保留四轮连续扫描、独立验证集和原精度门限为前提。没有采用直接用运动中
角度代替静态指令映射的办法,因为反馈滞后和方向回差会使两种角度不同。
### 准备段恢复
`runtime/zero_recovery.py` 为每组尚未冻结零位的关节提供一次局部恢复额度。
图像数量不足或候选暂时不确定时,复用该组原归零路径补采;几何已确认而零位图像
不足时,仅保持当前姿态重新打开图像窗口。恢复撤销旧观测回调,保留真实日志;
不替换已冻结几何/零位、不改变保持条件,不将设备故障当成视觉缺样处理。
两种恢复共享一次额度,持续失败仍暂停,不无限重做整场四轮运动。
### 训练停点精简
`core/fitting/command_sampling.py` 只读取前三轮连续数据,在原静态训练网格上求最少
保留节点,使每关节、每方向在原网格节点的插值变化不超过 0.25°。端点、baseline、
中间支撑点和显式非线性补充节点必须保留;训练不足或单调修正过大时保留完整网格。
该准则决定采样位置,不证明实际静态误差已经达标。每个保留点仍需采集新稳态图像;
原独立稳态验证点全部保留,只有这些验证与最终产物验收通过才可发布。
计划记录训练来源及自身哈希,采集、恢复、离线回放和拟合共同消费同一声明,禁止
使用第四轮或稳态验证点反过来选择训练节点。断点策略升级为
`confirmed_branch_v7_final_image_sampling`,旧数据不会混入新流程冒充新证据。
另一项优化只省去重复准备等待:上一段已经完成到位检查、下一目标的完整指令向量
相同、一秒内反馈未变化时复用该结果;真实位移、反馈变化、记录过期或归零准备段
仍执行原运动检查。原运动速度、避让和采样稳定性门限保持不变。
对 O6 `20260915_150015` 的已保存数据离线比较,稳态训练点由每方向 9/9/10 个变为
7/6/10 个;独立验证仍为 10/10/11 个,双向总停点由 118 次降至 108 次。
小指因训练曲线噪声保留原网格,不以强行减点换取精度风险。
原方案和精简方案均通过原独立稳态角度验证,全部关节最大误差仍小于 1.69°。
这是旧数据的训练点选择对照,不能当作新四轮实机或最终空间精度通过。
原始文件哈希保持不变,脚本及结果为该会话的
`review/assess_static_sampling.py``review/static_sampling_assessment.json`
### 最终文件到原始图像
`runtime/artifacts/image_evidence.py` 从原始记录绑定图像身份、四角、Tag 尺寸以及
受哈希保护的相机外参和 CameraInfo。`core/urdf/image_acceptance.py` 读取最终文件,
使用冻结的安装和基座,把 JSON+URDF 的 FK 直接投影到第四轮及独立稳态验证四角。
这里不调用 PnP、优化器或重新配准;同一最终产物还须通过原来的 3D 和角度门限。
每帧四角重投影 RMS 沿用 1.5 px 门限,超限保存 `final_image_diagnostics.json`
原始证据缺失、相机身份变化、训练/验证重叠及重复图像都会被拒绝。
新软件的自动测试与离线数据检查不代表四型号已经逐一实测通过。
O6 仍需恢复 `can0` 后运行新的完整四轮,并通过上述最终文件验收。
本次新增链路的验证结果:
- 四型号精简采样策略及实际 JSON→URDF→角点验收共 11 项通过(96.41 秒),
包含静态验证异常仍被拒绝、第四轮不影响训练点选择等反例。
- 局部恢复、真实定时器流程和重复等待边界共 24 项通过(21.62 秒)。
持续缺少第十张零位图像在一次补采后仍停止,不能少采冒充合格。
- 最终角点、真实相机来源、混合实测/复制产物及后台进程共 23 项通过(25.13 秒)。
- `colcon build --packages-select linkerhand_calibration --symlink-install` 通过(1.81 秒)。
- 构建后再次执行 O6 正式 `calibrate_hand --no-resume`,仍在配置检查阶段报
`CAN interface does not exist: can0`,没有进入运动采集;日志保存为
`20260915_160040/review/o6_optimized_flow_startup.log`
这些组存在覆盖交叉,不相加作为独立总数;四型号整链使用独立构造的合成观测,
真实来源绑定另用 O6 已保存角点检查,两者均不能冒充新实机结果。
全包回归中同步修正两处旧测试约定:零位不足的用例须观察到唯一一次恢复结束,
而非要求第一次超时立即暂停;历史侧面准备图像须调用带源 URDF 约束的当前生产入口,
不能继续要求旧自由几何模型必然可解。针对这两处的 7 项复测通过(11.49 秒)。
同一份侧面历史数据 121 帧使用当前入口通过,训练/验证仍为互不重叠的 61/60 帧,
候选统计门限与绝对像素门限保持不变;没有为了通过测试修改原始数据或生产门限。
全包首次运行共 1497 项:1494 项通过、上述 2 项旧测试断言失败,另有 1 项因未提供
O12 本地实采与独立参考文件而跳过(1175.15 秒)。两处失败修正后的 7 项复测通过,
没有剩余已知失败。完整日志、专项回归和构建日志保存在
`20260915_160040/review/common_flow_validation/`;不将跳过的 O12 实采检查视为通过。
## 第五轮评估:训练选择与准备验证的边界
2026-09-15 19:32 使用新相机外参进行正式 O6 采集,拇指两任务完整采完,
小指准备阶段两次均报 `image_motion_families_not_distinguishable`。原始数据与修改前
评估保存在 `20260915_193254/review/solver_assessment/`;先完成诊断实验,再修改生产代码。
整体链路保留:准备几何 → 冻结零位 → 四轮连续扫描 → 必要稳态映射 → JSON → 回读生成
URDF → 独立图像验收。本次问题位于准备模型的选择和验证,不涉及末端非线性存储格式。
独立稠密 Jacobian 求解证实增加迭代无效;四个初值有三个收敛到同一解,剩余解与其
相差约 19°、9 mm,不能把真实空间差异当作浮点误差。
审查发现原实现按验证误差选择优胜模型,再用同一验证集证明它较好;拟合使用平方误差,
统计比较却使用 RMS 差;逐个立即返回还保留了最保守的固定多重比较惩罚。
现统一为训练选择、验证否决,候选所用的训练资格不受验证结果或验证异常影响;
验证失败不得改选。公共 `image_model_selection.py` 按同一平方损失比较,完成全部候选
比较后使用 Holm 校正,原始与校正后的概率均写入诊断。
1.5 px、0.03 px、1°/1 mm 等价界、原角度和空间验收门限均保留,名义族错误水平仍为 0.01。
同帧等价检查按角色名对齐,并拒绝缺失或重复角色;不再依靠 `zip` 恰好有相同顺序。
姿态策略升级为 `confirmed_branch_v8_training_model_selection`,旧会话仅供诊断,不能
自动续接为新策略采集。新增策略适用于公共流程,不增加 O6 专用回退或停点。
修改前的候选策略诊断中,两次失败数据均可区分(概率约 0.00373、0.00576),拇指保持通过,
真正歧义的合成数据保持拒绝;这些仅是修复依据,不代表整手新实机验收完成。
统计假设和条件性边界参考 [SciPy Wilcoxon 文档](https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.wilcoxon.html)
多重比较方法参考 [Holm 方法文档](https://www.statsmodels.org/stable/generated/statsmodels.stats.multitest.multipletests.html)。
新策略实机 `20260915_195619` 完整采集 36/36 个方向单元,0 次暂停、0 次局部补采;
小指准备一次通过。146 项准备求解与产物链专项检查通过。最终空间验收仍失败,
候选 JSON/URDF 未发布;这与采集是否完整是两个不同结论。
## 第六轮评估:先修复测量时间,再判断空间求解
该会话的原始角点、生产重放和参数对照保存在 `20260915_195619/review/spatial_assessment/`
轴线投影开关没有改善整体误差;共同 URDF 的 39 参数图像求解虽然收敛,第四轮仍有
4.124 px 误差,因此没有采用这一替换。相同反馈和相同方向的第四轮图像还能构成
与任何确定性查找表不相容的反例:两帧像素距离 3.753 px,最大误差必有一帧不低于
1.877 px,超过现有 1.5 px 门限。该反例不依赖拟合算法,也不证明硬件损坏。
SDK 代码确认了一个时间来源错误:发布线程给低频轮询缓存反复填上当前时间。
相机与发布消息的时间差很小,并不能证明相机与真实反馈同步。修复位于共同数据边界:
- CAN 接收端原子保存位置和接收时间,O6/L6 单帧、G20 按各指分帧保留。
- SDK 的标准状态保留真实接收时间;标定专用接口携带每通道时间,仅发布新增测量快照。
- 标定按每通道实际时间插值。重复、乱序、过期或同一接收身份对应不同值的记录不能刷新
稳态样本或反馈新鲜度;原始反馈快照落盘,便于复核。
- 共用采集策略升级为 `unified_engine_v6_measured_feedback_time`,四型号 Profile 同步声明。
源 URDF、相机和精度门限保持原样;旧采集不能改挂新时间策略发布。
时间契约、缓存、分帧、回归与稳态检查 150 项通过;四型号配置、ROS 接入、断点与产物链
集成检查 69 项通过;SDK 与标定两包构建通过。不运动的实机接口核验中,486 条标定反馈
全部匹配独立旁路收到的原始 CAN 帧,0 条无来源、0 条重复身份。真实测量频率约 25 Hz,
同一时段标准接口发布 1147 条消息,说明之前约 58 Hz 的发布频率不能作为测量频率。
该核验启动 SDK,SDK 仍执行其初始化速度/力矩设置;没有发送位置运动命令。
完整实机产物仍须用新的测量时间重新采集后验收。
## 第七轮评估:控制消息顺序处理,图像任务独立执行
真实时间采集 `20260915_202203` 暴露了图像缺样:一段约 0.30 秒的有效图像空白
内仍有 15 条真实 CAN 测量,不应扩大同步窗口来掩盖。最初加入每视角有界图像工作线程,
但 ROS 端仍保留多线程执行器;`20260915_203104` 恢复导入期间仍因反馈超过一秒暂停。
这项中间实现没有完成目标,不能算作修复成功。
进一步评估保存在 `20260915_203104/review/dispatch_profiling/`:真实 64 条记录编码
约 6 ms,不支持直接减小导入条数。实机锁计时和线程栈显示,多个 ROS 回调争用同一把
会话锁,图像队列还在持有自己的锁时反向查询会话。仅在诊断入口替换为单线程 ROS
执行器后,反馈进入回调的延迟由初段中位约 113 ms、P95 247 ms,降到约 9 ms、17 ms
同一精度、采样和时间策略下已成功复用 8 个合格单元。
据此整理公共 ROS 接入,而不更换拟合器或增加型号特例:
- 控制、反馈、相机消息解析和状态更新由一个 ROS 执行器顺序处理,删除冗余回调组。
- 每个视角只保留一项执行中的图像任务和一项待处理任务,过载时保留最新待处理图像。
- 只有 coordinator 接受的新测量才能推进图像队列的时间水位;多帧协议使用最旧通道时间。
队列持锁时不调用 coordinator,开始新会话时清空旧水位和未执行图像。
- 原 50 ms 同步界限继续有效;图像估计仍在锁外,提交结果仍检查会话和运动版本。
- 图像线程异常传回控制入口,不将线程停止当成正常无图;检查点落盘规则保持原样。
调度、ROS 接入、真实测量时间、coordinator 和断点恢复共 51 项针对性检查通过;
图像元数据并发、过期结果提交和隔离计算共 28 项检查通过;标定包构建通过。
单线程对照运行及后续正式运行的终态以各会话记录为准;这里的延迟改善不能代替
四轮完整性、独立角点和 JSON+URDF 的最终空间验收。
## 第八轮评估:相机曝光时间与取帧时间分离
单线程对照会话 `20260915_203750` 完成 36/36 单元(复用 8 个已验证单元)、0 次暂停,
生成候选 JSON 和由该 JSON 重建的 URDF,但顶部 ID7 的训练安装位置 P95 为 18.285 mm
超过 3 mm,产物未发布。其余四个实测 Tag 角色通过该项空间检查。共同 URDF 的 39 参数
原始角点拟合虽然收敛,第四轮仍达 4.779 px,因而没有替换生产拟合器。
随后检查图像时间来源,发现相机驱动在 `MV_CC_GetImageBuffer` 返回后使用当前 ROS
时间,不能表示曝光时刻。实测三台相机的设备时钟、40 次独立时钟锁存,以及曝光开始/
结束事件,确认:
- 三台取帧时间分别比设备帧时间晚约 16.44 / 16.52 / 16.49 ms;不应靠拟合器吸收该延迟。
- 帧计数时间与独立曝光开始事件相差约 150 ns;固定曝光时可使用曝光中点时间配对。
- 当前固件的 `DeviceTimestampIncrement` 返回 100,000,000,实测约 10 ns/tick。
设备 XML 的单位说明与此不一致,因此程序启动必须用独立经过时间核验计数单位。
`camera_timing.py` 单独负责设备/主机时钟对应和曝光时间换算。每 0.5 秒重新锁存,
选择往返区间最短的有效测量;偶发缓慢事务丢弃后有界重试,不能刷新原两秒有效期。
错误单位、时钟跳变、计数器重置、重复帧和过期对应均不能伪装成新图像。
相机驱动在图像及 CameraInfo 中使用同一曝光中点时间,保留每次锁存和每帧换算记录
`camera_timing_<view>.jsonl`。固定外参和内参不因软件时间修正而改变。
共用采集策略升级为 `unified_engine_v7_camera_device_time`,四型号 Profile 及其受保护
哈希同步更新;旧会话保留诊断用途,不能直接续接到新时间规则。改动没有增加停点、
调整非线性查找表语义或放宽最终精度门限。
时钟、真实三相机记录、ROS 接入、配置、启动图和断点/队列共 57 项检查通过;构建通过。
随后实际驱动对三台相机各连续发布 242 帧、执行 16 次校时,无时序或时钟错误。
O6 运动期间另旁路监听 CAN:370 条位置命令的响应均不是原命令回显,仍可作为真实状态
观测;没有把 L6 的回显过滤规则未经核实套到 O6。
完整新策略实采会话为 `20260915_205704`,不复用断点、不使用调度诊断包装。
共完成 36/36 个单元,0 次人工暂停;拇指弯曲第四轮增加方向和小指第三轮增加方向
各自动重扫一次。候选 JSON/URDF 已生成,Tag 安装刚性训练检查失败,未发布 manifest。
正式失败时位置 P95:小指末节 3.108 mm、拇指末节 6.476 mm、拇指弯曲 3.875 mm、
顶部拇指 10.086 mm。会话中 `review/spatial_assessment/live_run_summary.json` 保留统计。
## 第九轮评估:反馈插值契约与空间模型边界
反馈拟合使用曝光时刻的连续插值值,报告声明 `piecewise_linear`;原 `JointMapping`
`SerializedJointMapping` 却对所有 `_u8` 输入再次取整,使训练和读取语义不同。
统一规则为:现代反馈曲线连续插值,实际 `command_u8` 量化,历史整数表保持原兼容行为。
序列化读取在量化前检查输入支持域,不能将越界小数取整后当成合法测量。
此修改不改变非线性曲线本身、不重拟合零位、不调整采集版本或原始记录。
先用相同 39 参数整链诊断做对照:保留原 URDF 骨长、轴向和已拟合曲线,前三轮拟合
共同基座、允许的零位及固定安装;独立第四轮不参与优化。取消反馈取整后,小指末节
最大像素误差由 3.390 降至 2.449 px,但仍超过原 1.5 px,因此该诊断模型未接入生产。
另加有界双向曲线修正的 119 参数实验,第四轮小指末节最大仍为 3.136 px,且出现
非单调局部曲线;增加参数并没有提供足以替换现有生产求解器的证据。
正式公共回放也验证:反馈取整修复不能消除空间不一致。位置 P95 为小指末节 3.080 mm、
拇指末节 6.489 mm、拇指弯曲 3.877 mm、顶部拇指 10.088 mm;保持失败且未发布。
原始会话与在线候选文件未覆盖;新回放位于 `20260915_205704/review/spatial_assessment/`
下的 `continuous_production_replay/`。同目录保留诊断脚本、参数、Jacobian 和测试日志。
补充整链可行性对照仍只使用前三轮、相同 39 个参数:将训练目标换成按原 2°/3 mm
尺度归一化的三维位姿残差,局部收敛后拇指弯曲零位触及原 20° 边界;第四轮顶部
位置 P95 仍为 5.148 mm,原始图像最大误差增至 28.718 px。此结果没有证明全局无解,
但否定了将这一局部三维优化结果直接用于替换现有求解器的做法。
原始保存位姿经受保护的相机变换投回同帧角点,各角色最大误差均小于 0.64 px;
没有发现这条原始公共坐标变换的方向或图像身份错配。检查见
`continuous_spatial.json``common_pose_projection_check.json`
反馈输入、双映射、被动非线性与读取兼容检查 57 项通过;补充真实时间插值、四型号
合成产物链、JSON 重建、独立角点和标准 URDF 检查 49 项通过,两组有交叉。
SDK 和标定包最终构建通过,`git diff --check` 通过。
合成验证不等于其他型号实机通过;本轮 O6 的最终空间精度问题仍未解决。
## 第十轮评估:保留跨运动的原始图像证据
此前原始角点随任务主机位的姿态采集写入,缺少顶部 ID7 在弯曲中、正面 ID2 在侧摆中
的完整原始图像。这两块 Tag 已确认属于同一末节;只保存各自任务下的图像不足以独立
核查共享运动链。公共采集器现独立保存所有机位的有效原始角点,保留原检测质量、
窗口与会话身份,不依赖 PnP 是否成功。主采样门限、运动计划和稳态点数量均不变。
同机位图像时间去重只在成功提交时推进;迟到或窗口已撤销的图像不能推进水位。
采集策略升级为 `unified_engine_v8_all_view_images`,四型号 Profile 及产品哈希同步更新。
旧会话不修改哈希、不补造新策略记录。新增采集与公共并发检查 43 项通过,补充采集、
四型号配置与断点身份检查 58 项通过,两组有交叉;标定包构建通过。
新会话 `20260915_213337` 完成 36/36,0 次人工暂停;小指第二轮反向扫描自动重试一次。
三机位共保存 16,822 个原始图像记录。正式算法仍在 Tag 安装空间一致性处失败,
未发布 manifest。主动零位与上一会话差均小于 0.03°,不能以重复性替代正确性。
保持当前源 URDF、相机参数和 Tag 尺寸,前三轮拟合共享基座、零位和每块 Tag 的固定安装,
第四轮检查的新全视角对照仍未通过。39 参数固定曲线模型的最大像素误差约 3.52 px;
359 参数双向单调曲线模型最大约 4.32 px。后者按 SDK 输入均匀选择训练图像,
不再按停留时间重复加权;非负曲线增量和参考输入零约束仍不足以消除残差。
静止姿态也出现约 1%~3% 的投影边长偏差,因此另做一次受限的统一骨长尺度诊断。
其训练最优尺度约 1.0327,第四轮最大误差仍约 4.37 px;没有采用该尺度,没有修改
源 URDF 或相机文件。原始公共位姿可重投回其源角点,暂未发现转换方向错误。
上述原型和数值结果均仅保存在 `20260915_213337/review/whole_chain/`,未接入生产求解器。
仅用稳态共同视野核查 ID2/ID7 刚体关系时,训练 6 个姿态、留出 7 个姿态全为侧摆,
不能独立确定完整外参。连续扫描保存了两种动作的共同视野,可在原曝光时间上对齐后
进行独立于 URDF 和曲线的外参一致性诊断;诊断结果不能自动覆盖已记录的相机参数。
## 第十一轮评估:尺度一致性与全机位最终图像验收
对 ID2/ID7 同末节约束做独立检查,不使用 URDF 尺寸、轴向或 SDK 查找表。
连续图像按曝光时间对齐,线性/局部三次及留一插值的一致性误差不超过 0.1 px;
训练前三轮 96 个姿态,第四轮 32 个姿态仅检查冻结的参数。每帧刚体位姿作为局部未知量,
两个 Tag 之间始终只有一个固定安装变换。多初值只按训练平方误差选择。
| 诊断假设 | 第四轮 front/top 最大误差 | 结论 |
| --- | --- | --- |
| 记录的外参与两个 16 mm Tag | 1.28 / 1.69 px | 与原始图像仍有不一致。 |
| 自由估计 front/top 外参,Tag 均 16 mm | 0.44 / 0.48 px | 外参改变量约 2.43°/23.72 mm;条件三倍标准差约 1.96°/22.24 mm,不能作为精确外参替换。 |
| 记录外参,单独估计 ID7 尺度 | 0.84 / 0.49 px | 固定 ID2 为 16 mm 时 ID7 约 16.52 mm;不是两块 Tag 实测尺寸差的独立结论。 |
| 记录外参,同时估计两个 Tag 尺度 | 0.42 / 0.52 px | 有效边长 ID2 约 16.38 mm、ID7 约 16.52 mm;依赖现有相机模型和刚体/正方形假设,尚不能替代物理尺寸。 |
自由外参不能解释此前保存的棋盘格图像:其最大误差约 12.93 px,原外参约 0.56 px。
因此不能因为手指图像误差变小就认定相机移动或自动重写外参。只读采集当前静止画面后,
正面背景匹配显示约 3 px 变化,侧面约 1 px 内;顶部旧图被棋盘格遮满,没有足够共同背景,
不能判断顶部机位是否变化。该检查没有启动 SDK 或驱动手指。
另将整链中的所有相机基线或所有 Tag 尺寸分别只增加一个公共尺度参数:前三轮最优值
约为 0.97226 和 1.03073。两者都明显降低拇指正面误差,但第四轮小指末节最大仍约
4.29/4.27 px、顶部最大仍约 2.66/2.44 px;没有采用任何尺度修正。
条件协方差未包含相机、打印尺寸和模型的系统误差,不能用其很小的数值宣称物理尺寸测准。
已请求分别精测 ID2/ID7 黑框宽高及棋盘格跨五格长度,现有整数尺寸记录不足以区分尺度来源。
结果在 `review/whole_chain/rigid_*_assessment.json``board_camera_comparison.json`
`static_background_comparison.json``all_view_*_scale.json`
程序结构上,只有任务主机位参与最终图像验收会漏掉同一 Tag 在其他动作下的不一致。
现将相机身份核验和原始图像索引集中在 `ImageEvidence`,反馈和稳态指令各自使用独立窗口。
前三轮、未通过尝试和无同步反馈的原始图像不能成为最终验收测量;主观测必须在新原始记录
中找到完全一致的角点、SDK 值和方向。其他动作中的原始角点也直接进入冻结 JSON/URDF 投影,
不伪造 PnP 位姿,不参与训练,不增加运动或停点。历史主图像读取接口保留原证据语义。
实际 O6 输入核对:反馈原有 1,818 个主观测全部一致,增加 3,270 个其他观测;
稳态指令原有 312 个主观测全部一致,增加 677 个观测。此结果仅证明证据绑定正确,
不代表最终空间精度通过。专项和现有产物流程 36 项通过,补充最终检查、四型号合成产物、
JSON 重建及非有限输入反例 34 项通过;两组有交叉,不相加为独立测试总数。
标定包构建通过;实际数据证据核对和测试日志保存在
`review/whole_chain/all_view_final_source_verification.json``final_image_validation/`
### 剩余运动残差的解释边界
固定上述未启用的候选相机尺度与整链几何,仅逐帧调整活动关节角度进行定位。
小指两关节最大改变量约 0.63°/0.49°,侧摆约 0.38°;图像最大残差分别可降至
0.43/0.94 px 和 0.47 px。最大原始残差集中在 SDK 约 244 的运动区间;
同一 Tag 静止、其他手指运动时的误差显著更小。
这项实验在每个被检查图像上估计局部角度,明确**不是独立第四轮验收**,也不证明
其条件几何正确。它提示剩余像素误差中包含运动映射误差,不能据此继续修改骨长或零位。
前三轮相同输入位置的角度改变量有约 0.1° 的轮间差,增加曲线节点也未必能消除。
后续需分别评估反馈分辨率、允许的运动映射误差与图像测量误差;不能直接把 PnP 的
像素测量门限等同于整条运动链的角度/位置精度要求。本轮没有修改任何验收门限。
完整逐帧结果见 `review/whole_chain/local_angle_residuals.json`
## 第十二轮评估:分离图像测量误差与运动精度
先用完全独立的已知真值验证规则:单关节运动存在恒定 0.5°、0.873 mm 偏差,
满足原角度和位置要求。仅将相机焦距从 1000 改为 3500 px,最终 FK 的最大投影误差
就从 1.10 变成 3.84 px;原规则因此对相同物理精度作出不同结论。测量位姿对原始角点
的误差均近于零。这证明 PnP 的 1.5 px 图像测量门限被误用于含运动映射误差的总投影量。
公共验收现在分为以下职责:
1. 发布器原有的独立三维反馈、指令验收仍先执行,角度 MAE≤1°、P95≤2°、最大≤3°,
位置 P95≤3 mm。非线性被动关节仍读取完整 JSON,不以 mimic 近似替代。
2. `image_acceptance.py` 将独立位姿与原始角点、SDK 输入、方向和图像身份直接绑定。
测量位姿重投影 RMS≤1.5 px;最终 FK 的总投影误差保留为单独统计,不据此改变物理精度要求。
3. 额外视角无独立三维测量时,`image_consistency.py` 查找一个能解释原始角点的局部刚性位姿,
直接核验其像素残差及相对最终 FK 的物理位移。数值见证仅说明图像与容差相容,
**不是真实位姿测量,不证明最优,不替代独立三维精度**。求解器返回成功也不能单独使其通过。
不回写或重新估计共享的相机、安装、基座、零位和曲线;按运动任务分别统计,防止静止帧稀释误差。
4. manifest 明确记录 `independent_measurement_and_physical_consistency_v2`
独立精度与额外机位相容性分开报告。缺失独立位姿不能由数值见证补齐。
解析投影 Jacobian 已用独立有限差分检查。已知真值、不同焦距、物理超差、角点/相机篡改、
证据身份和静止图像稀释等 42 项专项检查通过;四型号合成产物、非线性、JSON 重建及原物理
验收共 36 项通过;最后的规则版本/任务身份补充检查 15 项通过。测试集合有交叉,不相加。
标定包构建通过,`git diff --check` 通过。没有增加实机轮次或停点。
最新真实 O6 的 1,818 个第四轮及 312 个稳态主观测均核验了真实保存位姿与原始角点,
最大误差分别为 0.613/0.616 px。此检查没有拟合任何参数,属于测量证据核验,不是产物 PASS。
复算原训练安装检查仍得到拇指位置 P95 为 5.371/3.591/11.352 mm,仍失败,没有 manifest。
尺度/相机输入尚待现场精测区分,未采用之前任何诊断尺度,也未直接替换整链求解器。
已知真值脚本使用保存的旧规则源码复现反例,当前规则由生产测试验证;全部记录位于
`20260915_213337/review/whole_chain/image_error_budget/`
## 第十三轮评估:轴线参考点不能决定零位
一条转轴可用轴上任意一点表示;沿轴移动该点不会改变物理几何。当前 O6 的旧零位公式
先把两轴参考点的差直接投到图像平面,没有消除沿轴任意坐标。相机斜看转轴时,这个
任意坐标被误解释成弯曲零位。旧代码已为 O12 增加可选开关,但 O6/L6/G20 仍使用原公式。
在真实 O6 的相同轴线输入上,仅将被动轴参考点沿轴移动 ±20 mm:拇指弯曲零位从
9.8736° 变成 +1.0529° 或 −16.1650°,三个结果竟都通过原轴线角度检查。整个实验没有
改变相机、Tag 尺寸、运动图像或真实轴线,足以证明程序中存在几何表示相关的错误。
公共 `ObservationGeometry.phase_error` 现始终先取两轴的垂直间距向量,再投影到相机平面。
移除了 `project_axis_gauge_before_image` 型号开关及其编译参数;O12 删除已无必要的 YAML
声明并同步产品哈希,其数学行为保持原先已启用的正确规则。O6 的采集配置与受保护输入
不变,可以用原四轮数据离线复算,无须增加运动。
独立测试使用四型号的原始 URDF、自行生成已知 0.07 rad 零位,改变相机视角、共同坐标系,
并任意移动预测和观测的轴上参考点;所有平行轴对均恢复相同真值。源码专项、四型号完整
合成产物链、配置与 O12 安装/机位变化共 24 项检查通过;源代码的轴线/求解顺序 9 项通过。
构建后另核对实际加载的安装包与源码字节一致,安装包的轴线、图像与物理门限检查共
51 项通过,避免用旧安装文件检验新修改。各测试集合有交叉,不相加。
正式回放保留原始输入身份检查,输出到 `review/whole_chain/axis_gauge_production_replay/`
拇指弯曲零位约 −0.5384°、小指约 −1.8044°,不再依赖轴上参考点;侧摆仍约 14.5179°。
JSON 已先保存并被读取生成候选 URDF;空间安装检查仍失败,位置 P95 如页首所列,
未生成 manifest。这项数学修正没有消除后续空间不一致,也不能替代尚待核实的尺度输入。
对最近两次已保存的轴线输入使用同一修正公式,侧摆、拇指弯曲、小指零位差分别为
0.02955°、0.01044°、0.02942°。这是条件零位的重复性对照,不是绝对精度通过;
没有改写旧 v7 会话的采集策略或将其伪装成 v8 发布。结果见 `corrected_zero_repeatability.json`
旧公式及配置已备份到 `axis_gauge_before/``assess_axis_line_gauge.py` 从备份源码复现
历史反例;不将旧公式或型号条件保留在生产求解器中。数值对照见 `axis_line_gauge_assessment.json`
源码、安装包测试及构建日志均在同一 `review/whole_chain/` 目录。
## 第十四轮评估:独立双目尺度检查
先使用已冻结的 ID2/ID7 刚体关系,在第四轮 32 个共同观测姿态上仅估计逐帧刚体位姿,
不使用 URDF 或 SDK 曲线。保持原相机与两个 16 mm Tag 时,front/top 图像最大误差仍为
1.280/1.693 px;与原主机位三维测量的位置 P95 分别相差 26.84/19.09 mm。
此前“调整外参”或“调整 Tag 尺寸”的条件模型均能把最大图像误差降至约 0.5 px,
但得到不同的三维位置,且部分与原主测量相差约 20~33 mm。因此不能将其中任意一个
条件重建当成独立真值,亦不能用小图像残差证明最终空间精度。共享参数均未在第四轮更新。
对照见 `stereo_primary_measurement_comparison.json`
只读保存的静止画面还发现:侧面相机可以同时看见正面 ID1、ID2。用两机位对同一个 Tag
的四个对应角点直接三角测量,不预设 Tag 尺寸、刚体安装、URDF 或 LUT。两帧曝光相差
6.55 ms,保存时没有 SDK 控制进程、机械手静止。尺度仅来自记录的 front/side 外参基线。
- ID1 两组对边平均长度为 **16.496 / 16.503 mm**;两机位重投影 RMS 为 **0.536 / 0.443 px**
这项独立对照进一步提示当前尺度与“16 mm”的记录不一致。
- ID2 两组对边约 **16.593 / 16.713 mm**,但重投影 RMS 为 **1.583 / 1.412 px**
最大单角误差约 2.28 px,不能将该尺寸估计作为已通过质量检查的测量。
此检查仍依赖棋盘格确定的相机基线尺度,不能独立区分打印 Tag 偏大与棋盘格尺度偏差。
没有把 ID1 的尺寸复制给其他 Tag,没有更改原相机或尺寸配置,没有增加整手运动。
原始 PNG、相机元数据、逐角三维坐标及误差均可复查,见 `current_camera_images/`
`same_tag_stereo_assessment.json` 与相应同名诊断脚本。
下一项必要现场证据仍是已请求的精确黑框尺寸与棋盘格跨五格长度。现有图像能发现
尺度不一致,但不足以无假设地决定应修改哪一项物理输入。尺寸证据到位后,应先核实输入、
评估与原始数据的兼容性,再复算整链;不能把某个局部拟合最优值直接写入生产配置。
## 第十五轮评估:固定已确认尺寸,排除图像坐标处理与局部拟合误导
2026-09-16 用户再次明确:全部 Tag 黑色码区为 16×16 mm,棋盘格为 27×27 mm。
这解除上一轮的尺寸信息等待;本轮按用户确认值作为固定物理输入,不把双目条件估计
当作新的 Tag 尺寸。以下诊断记录在 `20260915_213337/review/whole_chain/confirmed_dimensions_audit/`
### 1. 原始图像与去畸变图像的数值核验
仅启动相机,机械手不运动;保存同一曝光时间戳的原始和去畸变 PNG。按 CameraInfo 的
K、D、R、P 独立执行 OpenCV remapfront/side/top 三机位与 image_proc 输出的最大、平均
灰度差均为 **0**。程序以 P 解释 image_rect、棋盘物点间距为 0.027 m;没有发现单位错误、
缩放裁剪、重复去畸变或原始 K 与去畸变图像混用。该证据只核验软件坐标契约,不证明
K/D 本身就是准确物理内参。结果见 `raw_rect_consistency.json``raw_rect_capture/`
新静止图像中 ID1 的两组对边平均为 **16.5057/16.5078 mm**,两机位重投影 RMS 为
**0.830/0.687 px**。ID2 的重投影 RMS 为 1.836/1.635 px,继续不作为合格尺度测量。
因此上一轮发现的矛盾可再次观测,并非一次检测结果;见 `fresh_same_tag_stereo.json`
### 2. 棋盘低像素误差不能单独决定新相机参数
使用已有原始角点,保持 27 mm 间距,对正面/侧面及正面/顶部训练对分别联合估计内参。
两组的留出误差均下降,但同一 front 的主点横坐标分别约 866/1078 px,明显不同。
又将单机位训练集分为互不重叠子集:side 的部分子集可给出明显不同焦距/主点,同时
在同一留出图像上仍保持约 0.45~0.48 px 的 RMS。不能凭这个像素指标选择物理参数。
结果见 `joint_intrinsics.json``intrinsic_stability.json`;未修改相机配置。
进一步以 OpenCV object-releasing 仅在训练图像中估计一个固定棋盘角点形状,保持
第一行两端 7×27 mm 间距。三个机位独立估计的形状均存在亚毫米偏差;其中 front、side
估出的形状相近。冻结形状后,未参与拟合的单机位图像 RMS 分别从
**0.276/0.465/0.317 px** 降为 **0.133/0.168/0.138 px**
这提示理想平面棋盘假设可能影响内参,但仍是条件模型证据,不等于已测准棋盘变形。
为避免“各相机自由拟合都能变好”的误判,另只采用 front 训练图像估出的同一形状,
重新估计 side/top 内参,再用 19:25 棋盘原角点训练新的条件外参(每四组留一组验证)。
side 的留出双目 RMS 最大约 0.395/0.383 px;随后检查从未参加相机拟合的新静止 Tag,
ID1 两组对边仍为 **16.5038/16.5058 mm**。棋盘误差改善没有解决目标尺度矛盾,故明确
拒绝启用这组候选参数。见 `board_shape.json``shape_camera_transfer.json`
### 当前下一步
已请求保持手和相机不动,将棋盘置于手旁,使 front/side 同时看到完整棋盘及 ID1,
采集同一时段的静止画面。这可直接核查已确认的两种尺度,减少先标棋盘、后拍手之间
场景变化的影响。另询问棋盘载体是否刚性平整,作为几何假设的现场证据。
本轮不增加整手运动、不启用候选相机参数、不改变 Tag 尺寸或精度门限;原数据和受保护
配置保留。整手发布条件仍未达成,无新的合格 manifest。
### 当日现场反馈与贴平后检查
用户补充“棋盘没有完全弯曲变形,tag贴纸有弯曲不平整”,随后确认“已贴平固定”。
因此不能把之前的平面 Tag 位姿当作已证实准确的物理测量,也不能把棋盘条件拟合
直接解释成全部误差的原因。正式复标需使用处理后的新图像,而非沿用旧 Tag 安装结果。
10:09 的贴平后静止对照,front/side 曝光差为 1.763 ms。ID2 的双目重投影 RMS 为
1.349/1.201 px,较处理前改善;ID1 为 2.254/1.872 px,仍不满足图像质量门限。
两者的条件尺度仍约 16.5 mm,不能据此宣布测量链修复完成。数据保存在
`confirmed_dimensions_audit/flattened_tags_20260916_100921/`
用户表示棋盘已摆好后进行了实际拍摄,并持续开启无机械手控制的双目预览;至 10:10
两路画面实际均未出现棋盘。已提供正面/侧面画面帮助摆放,没有将空缺的棋盘观测标成
成功采集。新的同帧棋盘/Tag 核验脚本已用已知真值验证:正确相机下恢复 16 mm 各边,
误差小于 1e−5 mm;该合成检查不是现场相机验证。下一步仍需让棋盘实际进入两个机位,
目前未重新启动整手运动或输出新的合格产物。诊断汇总与受保护输入哈希见
`confirmed_dimensions_audit/audit_summary.json`
## 第十六轮评估:分时核验棋盘与 Tag,取消不必要的同框限制
固定相机外参时,棋盘负责估计相机之间的刚体变换;Tag 的位置和姿态在检查时独立
估计。因此棋盘与机械手不必同时出现在画面中,手在两次拍摄间也不必保持原位置。
此前把同框设为必要条件,给现场带来了不必要的摆放限制,现已取消。只需相机及镜头
设置在两段采集中保持固定。这一修改属于诊断采集流程,不改变整手正式运动流程。
10:21~10:22 保存四组正面/侧面完整棋盘原图:
`board_placement_preview_20260916_102131``102213``102215``102217`
原始双目曝光差 10.79~10.99 ms;每组以已确认的 27 mm 棋盘估计条件相机变换,
棋盘重投影 RMS 为 front 0.2810.364 px、side 0.4380.571 px。
四次相对首帧的变化最大为 0.0272°、0.458 mm。姿态相近的四组只说明条件重复性,
不能充当完整多姿态外参标定,也没有写入活动相机配置。
临时预览的自动保存器还曾设置 20 ms 配对条件,而一次启动后的相机帧相位差约
24.66 ms,造成没有进入检测。这是诊断脚本的问题,已采用原相机核验的 50 ms 配对
范围,并保留真实曝光时间;正式图像与空间精度门限不变。另一个临时触发条件要求
全部棋盘角点连续 2 秒变化均小于 0.4 px,未触发自动保存。上述四组是另行直接保存的
原始画面,经离线图像核验后使用;不能把预览中的 `saved_count=0` 改记成自动触发成功。
后续共面检查直接保存可见原图,在原图上评估几何质量,不先用这种触发条件丢弃证据。
用户移开棋盘、放回 O6 后,10:24 保存新的 Tag 图像,曝光差 11.75 ms,位于
`joint_board_tag_20260916_102403/`。分别使用上述四组棋盘估计的相机变换,得到:
- ID1 两组对边平均约 **16.48716.500 mm**;自由角点双目重投影 RMS 为
front 1.0401.107 px、side 0.8780.935 px。
- 强制 ID1 为 16 mm 平面方形时,front RMS 仍为 **1.6291.677 px**,超过 1.5 px。
- ID2 两组对边约 **16.49916.585 mm**;自由角点的双目 RMS 约 0.410.51 px。
新旧手位姿分别估计,不参与棋盘相机参数的拟合。使用已知真值的另一个位置/姿态
Tag 验证过这一分时检查方法,两种来源均恢复 16 mm,误差小于 1e−5 mm。
结果记录在各棋盘目录的 `board_tag_consistency_joint_board_tag_20260916_102403.json`
这仍不足以单独判定应改 Tag 尺寸、棋盘尺寸还是相机模型;没有因此运行新一轮整手运动。
### 共面长度比例的独立检查
用户提供同批备用 Tag,并把 ID1 固定在棋盘同一块板上。10:27 的三个原始正面图像
位于 `coplanar_capture_20260916_102710/frame_00``frame_02`。图中左侧的备用 ID1
与棋盘共面;右侧手掌 ID0 不共面,不将其按棋盘平面换算的数值解释为尺寸。
使用棋盘格交错的 20 个角点拟合平面单应变换,另 20 个角点检查投影误差,不使用
相机外参、URDF、SDK 或关节曲线。原始像素直接计算的 ID1 两组对边约
**16.47416.478 / 16.59516.602 mm**,留出棋盘角点 RMS 约 0.1860.191 px。
用当前内参去畸变后约 **16.68016.685 / 16.59316.600 mm**;旧内参、未采用的
棋盘形状内参仅作预先规定的敏感性对照,结果也记录在同一报告中,不按 Tag 尺寸选模型。
该 ID1 接近原图左边缘,畸变模型之间仍有约 0.2 mm 的差异,所以未据此改配置。
随后请用户将备用 ID1 移到棋盘上方中间,再核验边缘畸变的影响。
### 中央位置复核与实测尺寸的矛盾
用户将备用 ID1 贴到棋盘中间上方后,10:43 保存三个新的原始正面图像,位于
`coplanar_capture_20260916_104349/frame_00``frame_02`。图中 ID1 位于同一底板上方,
纸张棋盘完整可见;手掌 ID0 仍不与棋盘共面,不解释其尺寸换算值。
沿用预先规定的棋盘交错训练/留出划分,三个图像的结果为:
- 原始像素直接计算的 ID1 两组对边为 **16.51116.513 / 16.59416.602 mm**
留出棋盘角点 RMS 为 **0.1880.197 px**
- 当前内参去畸变后为 **16.52116.523 / 16.63616.645 mm**,相较原始像素仅变化约
0.01 / 0.04 mm,不能解释相对 16 mm 的约 0.5 mm 差异。
- 另用传统棋盘角点检测加亚像素细化,并分别使用全部角点、靠近 Tag 的三行或两行
拟合平面变换;不同检测方法和网格子集的边长变化小于 0.1 mm,比例差异仍存在。
这是同一图像上的方法敏感性检查,不是新增独立实物样本,也不是新的合格标定。
结果见各帧的 `coplanar_scale_assessment.json` 及汇总目录中的
`central_coplanar_sensitivity.json`。上述长度全部以 27 mm 棋盘间距、平面棋盘及
Tag 与棋盘共面为条件;图像本身不能无条件证明 Tag 的绝对毫米尺寸或两者物理共面。
这项检查未使用外参、URDF 或关节拟合,不能通过增加关节拟合参数解决其比例矛盾。
用户随后确认 16 mm 和 27 mm 均为打印后实物测量值。保留这两个生产输入,不能将
图像条件换算直接盖过实测值并断定打印错误。已请求测量工具、Tag 黑框宽高及棋盘连续
五格的未取整总长度,以区分测量精度与几何假设的问题。此次仅进行静止图像与离线
检查,没有新增整手运动,没有修改相机、尺寸或验收门限,没有生成合格发布 manifest。
## 第十七轮评估:依据实物复测修正 Tag 尺寸
用户再次量取后明确:“tag 是 16.5×16.5 mm,棋盘还是 27×27 mm”。这提供了独立的
实物输入修正依据,与先前双目约 16.49 mm、中央共面约 16.51×16.60 mm 基本一致。
不是从 URDF 拟合残差反推并选择一个易于通过的尺寸。
原先按 16 mm 计算 16.5 mm 的方形 Tag,会在同一单目姿态解下将平移距离缩小到
16/16.5,约低估 3.03%;跨相机变换中的基线平移来自 27 mm 棋盘,并不会随之缩放。
这构成明确的输入尺度不一致,不能通过增加运动曲线参数来合理消除。
修改范围仅为当前 O6 的以下配置及指纹:
- `profiles/o6_right_8.yaml`8 个 Tag 均显式声明 `size_m: 0.0165`
- `o6_right_8_tags.yaml`:检测器默认尺寸和每个 ID 的尺寸均为 0.0165 m。
- `o6_three_camera_calibration.yaml`:标定节点尺寸及全部 ID 覆盖值同步为 0.0165 m。
- `o6_right_product.yaml`:更新以上三个受保护文件的 SHA-256。
程序已有 Profile、检测器和标定节点的尺寸一致性检查,本次无需增加型号特判或
修改拟合算法。其他型号/打印批次没有新的实测依据,不自动替换其配置。棋盘尺寸未变,
没有因此修改相机内外参、源 URDF、旧采集文件或验收门限。非线性关系仍由完整 JSON
和修正 URDF 配合表达。改前配置与改后指纹见
`20260915_213337/review/whole_chain/tag_size_correction_20260916/`
28 项配置加载、实际 launch 参数和 PnP 参数测试通过;已重新构建安装,核实四个
配置文件与源码逐字节一致,`--validate-only` 通过。用户移开棋盘及备用 ID1 并确认
O6 可运动后,启动新会话 `20260916_105528`,显式 `--no-resume` 完整采集。
最终是否合格须以该会话的独立验收及发布文件为准,不能仅凭尺寸改正确就声明成功。
### 新会话结果与收尾内存问题
36 个采集单元全部通过,小指第三轮 increasing 方向第一次存在反馈分箱空白 42,
按原最大空白 16 的门限自动重扫一次后通过。完整采集约 6 分 47 秒;保存 15,167 个
原始图像记录、6,828 个关节样本及 15,629 个 SDK 反馈样本。11:02:20 完成最后采集单元,
随后开始安全回位与收尾。11:05:28 系统内核明确记录 OOM、4 GiB swap 耗尽,终止了
一个 VS Code 进程;三相机同时出现长时间时钟读取中断,SDK 反馈超时,终端随后发生
BrokenPipeError。不能把该现象归因于机械手硬件,也不能将此会话记为在线无暂停完成。
原始日志完整保留,正式 `--offline-raw` 流程验证了配置指纹、全部采集单元、运动来源、
独立第四轮与稳态验证数据,生成 `offline_165mm/` 下的 JSON、URDF、报告和
`release_manifest.json`,退出码 0。JSON/URDF 的 SHA-256 与清单相同;已从保存 JSON
再独立调用重建 CLI,得到逐字节相同的 URDF。原来的角度 MAE/P95/max 1°/2°/3°、
位置 P95 3 mm 及对应图像门限没有修改。
同时发现 `read_journal_prefix` 先整文件读取、再 `splitlines()`,在解码对象之外还保留
两份原始文本。这份日志为 318,772,511 字节,额外副本会显著增加相机仍运行时的收尾
峰值。已改为按冻结字节边界逐行解析,保留全部记录、顺序和不可越过边界的校验;
不通过丢数据或修改安全超时规避问题。新增 16 MiB 日志的临时缓冲内存上限回归检查,
并验证短文件与非对象记录拒绝;独立进程、取消、真实 finalizer 等共 9 项测试通过。
已重新构建安装,进一步用同一实采数据检查独立收尾进程与已验收文件的一致性。
修改后的已安装独立进程以 `require_motion_evidence=True` 处理同一份完整原始数据,
**55.21 秒**完成,子进程峰值 RSS 为 **1,344,164 KiB(约 1.28 GiB**
JSON、URDF 两个文件的 SHA-256 均与正式离线验收版本相同。该检查没有启动相机或 SDK,
也没有重新运动机械手,不能单凭它宣称修复后在线全过程已经完成第二次实机验证。
报告为 `20260916_105528/isolated_finalization_verification.json`
通过已验收文件的字节/哈希与 JSON→URDF 对应关系复核后,使用现有发布器将相同文件
发布到产品根目录的 `20260916_105528_verified_165mm/`,并更新 `latest_partial_passed`
原始中断会话、离线验收目录与独立进程检查目录全部保留;没有把原始 PAUSED 状态改为
COMPLETE。正式目录中的 `release_origin.json` 记录已验收来源;完整结论和实际采集范围
`20260916_105528/verification_summary.json`。当前布局直接测量拇指和小指共 5 个关节,
食指、中指、无名指共 6 个关节的运动曲线按原配置从小指迁移,不宣称这些关节独立实测。
## 第十八轮评估:被动关节由 JSON 提供角度,消除线性导出冲突
用户指出拇指 IP 在修正 URDF 中明显弯得更小。复核确认,原 URDF 的线性倍率为 1.86,
旧修正产物的倍率为 1.16556243135602,而实测 JSON 的 IP/pitch 比例随行程约从 1.71
变化到 1.94。父子下限分别为 -0.00172483145889436、-0.0020103987491093 rad
旧“共同零点+整段不越界”约束把这两个小负端点的比值变成全行程倍率上限。
这不是非线性 JSON 的拟合失败,而是将独立实测曲线再次约束为一条直线的导出设计问题。
旧完整 JSON+自定义 FK 验收虽通过,标准 ROS mimic 仍会覆盖被动角,因此显示与验收不一致。
评估后采用统一的 JSON 角度来源规则,不修补 O6 的某个倍率:
- 新 v3 修正 URDF 移除原被动关节的 `<mimic>`,全部关节角由 JSON 明确给出。
父子链、转轴、连杆、零位与限位保持原有受检语义;原被动关节没有新增实机驱动通道。
- `urdf_correction` 升为版本 2,保存 `joint_angle_source: calibration_json`
`passive_joint_sources` 和明确的删除操作。只有完整覆盖运动关节、来源与原 URDF 一致、
被动关节与来源关节共享 SDK 通道时,才能授权移除;其他数值编辑仍使用 Profile 原授权。
导出版本与采集配置分开,未修改 Profile 指纹或原始记录来绕过保护。
- JSON 先保存,重新读取后重建 URDF;普通 FK、角度/位置验证及原始图像验收使用该最终文件。
标准 `robot_state_publisher` 可直接接收全部关节状态,不需要忽略 mimic 的定制算法。
- 线性拟合仅保留报告诊断,标记 `diagnostic_only_not_exported`,不再按限位限制拟合倍率,
不参与导出和运行角度。旧 correction v1 的精确重建和旧读取方式继续兼容。
专项回归覆盖微小负限位、非线性行程、普通 FK、JSON 重建字节一致性、错通道、缺失曲线、
非法来源、未经授权删除与拓扑改动拒绝。真实 ROS TF 测试在独立命名空间验证三个弯曲点,
未连接 SDK;末节角度不再被覆盖。O6/G20/L6/O12 的合成完整收尾、独立图像验收、迁移来源、
双向映射及旧格式检查通过;这不是其他三型号的新增实机精度证明。
已构建安装,并用 `20260916_105528/raw_samples.jsonl` 的完整 16.5 mm Tag 实采数据
执行正式离线回放与发布,退出码 0。新目录为
`calibration_output/O6_RIGHT_001/20260916_123347_json_driven_165mm/`
`latest_partial_passed` 已更新。原来的角度、位置和原始图像门限全部通过,没有放宽。
对新旧发布文件逐项比较:11 个关节的全部指令表、反馈映射、零位和限位不变;
URDF 的结构差异仅为移除 5 个线性 mimic。重新读取新 JSON 重建得到逐字节相同的 URDF。
相同输入的拇指结果如下(单位:度):
| SDK 指令 | CMC pitch | 旧 URDF 线性 IP | 新 JSONURDF IP |
| --- | ---: | ---: | ---: |
| 192 | 9.3781 | 10.9307 | 16.0523 |
| 128 | 18.8914 | 22.0191 | 33.0907 |
| 64 | 27.2887 | 31.8067 | 49.6742 |
| 0 | 34.2391 | 39.9078 | 66.3515 |
复核脚本与结果保存在原会话的 `verify_json_driven_export.py`
`json_driven_export_verification.json`,正式回放日志为 `json_driven_replay.log`
旧发布三文件的哈希未变。本次没有重新运动机械手,沿用原 5 个独立观测关节、6 个迁移关节
及 CAD 零位假设的精度边界。使用新产物时须通过 JSON 同时驱动全部关节;单独拖动 URDF
父关节滑条不再自动联动末节,外部仿真也须移除与 JSON 冲突的旧线性更新逻辑。
## 第十九轮评估:按用户要求保留标准 mimic 近似联动
用户明确使用 `https://viewer.robotsfan.com/` 直接预览 URDF,并进一步说明:
非线性关节也必须保留 `<mimic>` 自动联动,单独 URDF 尽量接近实机;
需要实测非线性结果时再由标定 JSON 查找表控制。上一轮删除 mimic 的正式导出方向
不符合这一使用要求。已核对网站当前构建及官方 URDFAdapter:滑条直接设置 URDF 关节,
没有本项目标定 JSON 的读取逻辑,不能期望它自动获得 JSON 中的非线性曲线。
### 方案评估与实现
标准 mimic 使用 `q_child = a*q_parent+b`。将 `b` 强制为 0,再约束父子全行程,
使拇指的两个微小负端点决定全局倍率上限,这是旧版 1.16556 倍率失真的原因。
仅恢复 `<mimic>`、手填倍率或扩大限位都不能解决模型与约束之间的矛盾。
现在使用两端子关节角度作为拟合变量:
- 父关节输出范围 `[p0,p1]` 保持不变,拟合 `c0=q_child(p0)``c1=q_child(p1)`
- 两个变量分别受子关节原输出限位约束;区间内的线性插值自然不会越界。
- 由 `a=(c1-c0)/(p1-p0)``b=c0-a*p0` 导出标准 mimic 倍率和偏移。
- 沿用前三轮成对视觉样本与稳健损失拟合,第四轮只验证;不改变 SDK→角度表、
坐标零位、Tag 安装、相机参数或训练支持的限位。
- `b` 是线性近似在零点的误差,不是新测出的机械零位。报告同时保存
`baseline_error_rad``physical_zero_correction: false` 和实际残差。
生产 `urdf_correction` 使用版本 3,并声明
`mimic_policy: range_bounded_affine_approximation``joint_angle_source: calibration_json`
仍先保存 JSON,再读取生成 URDF。所有原 mimic 来源与 XML 拓扑保留;
数值修改继续使用已有 Profile 授权,没有更改采集指纹。版本 1、2 产物保留重建兼容,
不原地覆盖。完整 JSON 回放显式使用非线性被动角,URDF 单独回放使用 mimic;
两者分别报告精度,线性近似不能冒充 JSON 的实测精度。
专项检查覆盖微小负限位不压低全程倍率、正负运动方向、全行程限位、留出样本禁止参与拟合、
重建字节一致性及缺失/降级导出规则拒绝。真实标准 ROS TF 检查只给父关节角度,
验证被动关节按导出 mimic 联动;历史无 mimic 版本的独立关节测试也继续通过。
### O6 实采数据复验与发布结果
相关 121 项检查通过,覆盖四型号合成产物、迁移来源、双向表、历史格式及标准 ROS 联动。
构建安装后,以同一份完整 `20260916_105528/raw_samples.jsonl` 正式离线回放,退出码 0
生成并发布 `20260916_124718_mimic_json_165mm/`,更新 `latest_partial_passed`
JSON+URDF 的角度、位置及独立原始图像验收通过,原精度门限不变。
新旧三版的全部 11 关节指令/反馈曲线、几何零位、限位逐项一致;5 组 mimic 来源与
原始 CAD 相同。拇指 IP 倍率为 **1.86604745134131**,偏移为
**0.00120821859875384 rad0.06923°)**。小指及三个迁移末节的倍率为
**0.870364582419037**,偏移为 **0.0348981257618717 rad1.99952°)**
这些偏移是近似关系在零点的残差,完整 JSON 的 baseline 仍为 0。
| SDK 指令 | CMC pitch | 旧 URDF IP | 新 mimic IP | JSON IP |
| --- | ---: | ---: | ---: | ---: |
| 255 | 0.0000° | 0.0000° | 0.0692° | 0.0000° |
| 192 | 9.3781° | 10.9307° | 17.5691° | 16.0523° |
| 128 | 18.8914° | 22.0191° | 35.3214° | 33.0907° |
| 64 | 27.2887° | 31.8067° | 50.9912° | 49.6742° |
| 0 | 34.2391° | 39.9078° | 63.9611° | 66.3515° |
相对 JSON 的 256 项指令网格,拇指线性近似最大差约 **2.3905°**,小指及对应迁移末节
最大差约 **4.1965°**。这是两个表示之间的差异,不是新增的实机独立精度测量;
独立留出结果分别保存在 manifest 的 `full_json_urdf_holdout``urdf_mimic_approximation`
因此网站中的联动预览仍是近似,需要实测非线性结果时必须使用 JSON。
另行读取新 JSON 重建得到逐字节相同的 URDF。所有旧发布文件哈希不变,
没有重新运动机械手,也没有改写旧采集结果。复核脚本、结果及正式回放日志分别为
原始会话下的 `verify_affine_mimic_export.py``affine_mimic_export_verification.json`
`affine_mimic_replay.log`。本次仍只独立观测拇指、小指 5 个关节,其他 6 个关节沿用原迁移关系。
## 第二十轮评估:全型号保留原始 mimic,实测非线性由 JSON 提供
用户确认统一采用原始 mimic 作为基础联动,暂时取消正式标定中的 mimic 参数优化。
该规则替代第十九轮新产物的拟合策略;历史产物保持原样及原有读取语义。
### 全型号依据与边界
检查 L6、O6、O12、G20 的左右手原始 URDF,以及当前四套右手产品配置。
L6/O6 部分曲线从小指迁移,O12 包含两级 mimic 链和无名指曲线迁移,G20 配置覆盖全部实测关节。
左右手系数不能互相套用:O6 拇指左右分别为 2.29/1.86G20 为 1.02/1.03。
原始文件也不等于限位自洽:按原始右手父关节全行程,L6 拇指约超出子限位 3.6795°,
O12 无名指中节约 0.7162°,G20 四个 DIP 约 0.4297°。这些是文件内冲突,不是硬件越限结论。
O6 的已声明实测范围策略不能自动授予其他型号放宽 CAD 硬限位的权利。
### 统一实现
- 新 correction schema v4`mimic_policy=preserve_source_mimic``joint_angle_source=calibration_json`
- measured 正式拟合及收尾不再调用 mimic 参数优化,原 mimic 来源、倍率、偏移及省略属性全部保留。
- `source_mimic.py` 统一计算负倍率、非零偏移及串联关系的可达范围,不含型号分支。
- `mimic_envelopes` 单独记录基础范围、联动可达范围和最小导出包络;不会改写实测范围证据。
- 包络扩展只适用于已有实测范围策略;默认 CAD 硬限位和独立安全限位继续生效,冲突明确报告关节及范围。
- 零位仍修正 origin。原始 mimic 只作为修正坐标中的预览近似,精确 FK 使用 JSON 提供的各关节角。
- 新 JSON 禁止请求 mimic 参数修改或删除;最终验收对照受保护原文件核对 mimic 元素和可信范围证据。
- correction v1/v2/v3 继续兼容重建与读取,既有原始记录和配置指纹不变。
### 验证与 O6 发布
两组相关检查分别 **68 项、72 项通过,共 140 项**。覆盖八份左右手源文件、负倍率、零倍率、
非零偏移、多级链、独立安全界限、非法 JSON 修正、历史版本重建及标准 ROS 的父关节联动。
四型号合成完整流水线同时验证:禁止调用 mimic 优化函数,仍能完成 JSON→URDF→独立角点验收→发布→读取。
这些合成检查不代表其他型号已经完成实机标定。
构建后以完整 `20260916_105528/raw_samples.jsonl` 正式离线回放,退出码 0,发布:
`20260916_134154_source_mimic_165mm/``latest_partial_passed` 指向该目录。
5 个 mimic 元素与原始 O6 URDF 一致,拇指 1.86、其余四个末节 0.89offset 全为 0。
所有 11 关节指令/反馈查找表、几何零位、实测范围证据与前三版逐项一致,旧产物哈希不变。
相对上一版 URDF,除 mimic 参数外仅 IP 下限改变:
`0.0020103987491093 → 0.00320818651354352 rad`,约增加 **0.068628°** 的联动兼容范围。
SDK 指令 0 时,拇指 pitch 为 34.2391°,原始 mimic 预览 IP 为 63.6848°,JSON IP 为 66.3515°。
完整 JSONURDF 的拇指 IP 独立角度 P95 为 0.43466°、位置 P95 为 2.11692 mm
小指 DIP 分别为 1.87341°、1.94512 mm,原精度门限未改。
独立图像验收的门限、样本数量和通过结论保持不变;重复数值优化的末位浮点值及诊断项可能变化。
读取保存的 JSON 重建得到逐字节相同 URDF;本次没有重新运动机械手。
复核材料位于原始会话目录:`verify_source_mimic_export.py`
`source_mimic_export_verification.json``source_mimic_replay.log`
当前 8 Tag 布局仍只独立测量 5 个关节,其余 6 个按既有声明迁移;在线采集曾有 OOM 的边界保持原记录。
@@ -0,0 +1,230 @@
# 按关节零位与实测/复制 JSON v3 实施记录
更新:2026-09-15。范围:`linkerhand_calibration`
## 当前收敛结果
- G20、L6、O6、O12 右手共用四轮连续扫描,随后执行独立的稳态指令映射训练与验证。原精度门限保留。
- 正式入口在启动硬件前检查运动覆盖与绝对零位来源。O12 的 `thumb_mcp``pinky_pip``ring_pip`
按用户确认沿用原始 URDF 零位,记录为 `assumed_source_cad_zero`;运动继续实测或按声明复制。
- 收尾先保存 JSON,再由 `runtime/artifacts/urdf_from_json.py` 读取它重建 URDF。
发布器检查修正元数据能否逐字节重建最终文件;可独立使用 `rebuild_calibrated_urdf` 命令。
- 四型号的合成原始观测均已通过实际拟合、文件生成、验收、发布及读取器加载。
合成夹具使用明确的几何真值与范围,这不代表四种实机均完成验证。
- O6 会话 `20260915_134211` 实采 36/36 单元一次通过,无重扫、无采集中断,采集耗时 412.35 秒。
同一原始日志已生成候选 JSON 与修正 URDF,独立 JSON 重建得到相同字节;整链几何验收仍失败,未发布。
最大 Tag 位置 P95 为 15.32 mm,小指末节姿态 P95 为 7.76°,不能称为整手精度合格。
- 随后用棋盘格重新测得 O6 三相机外参:两组各 15 个姿态,最大联合重投影 RMS 为 0.928 px
最大重复平移差 0.341 mm、旋转差 0.043°,通过原门限。新配置为
`config/o6_three_camera_extrinsics_20260915.yaml`;旧数据保留原外参身份,不混用。
- 新外参会话 `20260915_144514` 完成拇指 24/36 单元后,小指准备求解有两个候选耗尽 150 次迭代。
同一批 120 帧离线复现确认是内部 LSMR 线性求解精度不足;调整 `atol/btol``1e-10` 后,
四候选均在原 150 次预算内收敛。外层收敛条件、分支统计和物理精度门限未变。
`test_recorded_pinky_alternatives_converge_within_original_budget` 保存该实采回归;相关 34 项测试通过。
- 恢复会话 `20260915_145437` 暴露旧日志在运动回调中校验导致的反馈超时报错:离线复现读取约
1.41 秒、完整凭据/单元校验约 13.88 秒。`PreparedResume` 将静态校验移到回调启动前;
当前固定基准与逐关节安装仍在现场独立验证。后续 `20260915_145748` 暴露一次性导入大日志
仍会阻塞;改为每个控制周期最多导入 64 条记录,全部持久化后才确认复用。
恢复相关 73 项测试通过,包含导入期间持续处理反馈、接收中止及落盘前不得跳过单元;原 1 秒反馈保护未变。
- 图像模型零位恢复还需同步导入该零位的 `pnp_candidate_frame`,不能只复制 `joint_zero_sample`
已修复该来源闭合问题;真实中断恢复会话离线核对发现恰好漏掉 10 帧零位图像,按新导入逻辑保留后
原凭据检查通过。原始日志未修改。相关 74 项测试通过,包含连续两次恢复的零位图像回放。
- 新外参完整会话 `20260915_150015` 已完成 36/36 单元,采集中无暂停,两次自动重扫均通过。
生成的候选 JSON 可独立重建出字节相同的修正 URDF,并通过标准 ROS URDF 加载。
整链训练阶段的刚性 Tag 安装检查仍失败:拇指 IP 位置 P95 为 9.73 mm,小指 DIP 姿态 P95 为 9.40°。
这不是第四轮整链精度通过的结果,未生成发布 manifest,原始 URDF 未覆盖。
用户已确认 ID4/ID5 安装在平整硬片上,Tag 黑框为 16×16 mm,棋盘格单格为 27 mm。
尚不能据这些残差判定硬件故障。
- 随后完成三相机原始棋盘格内参核验:正面 34、侧面 25、顶部 24 个不同姿态;
每第 4 张预先留作独立验证。验证重投影 RMS 从旧内参的 0.880/0.955/0.484 px
降至 0.276/0.465/0.317 px。新内参保存在 `config/o6_camera_intrinsics_20260915/`
配套外参使用正面/侧面 25 组同步观测和补采的正面/顶部 15 组同步观测,均无剔除;
最大 RMS 0.575 px、旋转重复差 0.024°、平移重复差 0.269 mm,通过原门限。
O6 产品配置已指向 `config/o6_three_camera_extrinsics_intrinsics_20260915.yaml`
并通过无运动的配置及受保护输入校验。旧相机文件与旧整手记录保留原身份。
原图、角点、训练/验证划分、拟合报告及导出哈希保存在
`calibration_output/O6_RIGHT_001/camera_intrinsics_audit_20260915/`
相机核验通过后仍须新采整手数据,不能将旧整手会话改挂新相机参数发布。
- 新相机完整重采 `20260915_154743` 的拇指 24/36 单元全部完成,无重扫;小指准备的四候选
均在 26–36 次迭代内收敛,但候选区分证据不足,停在 `image_motion_families_not_distinguishable`
随后发现统一 launch 将四型号受保护 Tag 配置中的 `detector.decimate=1.0` 强制覆盖为 `1.5`
已移除这层未声明在产品配置内的覆盖,检测设置统一来自受保护 YAML;不修改候选区分或物理精度门限。
新回归实际展开 ROS 组件参数,修复前四型号均复现 `1.5 != 1.0`;修复后启动及 runner 的 55 项测试通过。
构建和 `git diff --check` 通过;旧降采样记录不混入新结果。
- 原分辨率完整重采 `20260915_160040` 已从三台运行节点读回 `detector.decimate=1.0`
拇指 24/36 单元一次通过,无重扫;小指准备的 104 帧、四候选均在 21–35 次迭代内收敛,
仍因 `image_motion_families_not_distinguishable` 暂停,未生成本次完整候选 JSON/URDF。
离线回放中两个未分开的 ID4 姿态旋转差 P95 为 18.64°,不能当作同一解;
ID4 法线与相机光轴夹角在两解下分别约 6–8°、10–11°,运动中一直接近正视。
已请求保持相机、手掌和 ID3 不动,将 ID4 硬片垫斜后重新采集;这不是硬件故障结论。
`review/all_view_detections/` 保留本次后 200.7 秒的三机位检测与 SDK 消息。
其中侧摆期间同时看到 ID1/ID2 的匹配样本仅 9 帧、反馈范围 206–255,
不足以判断整个侧摆行程是否引起被动关节联动。所有精度门限保持原值。
- 上述修复后的四型号文件链、JSON 重建、恢复凭据和图像求解定向回归共 55 项通过(76.55 秒),
构建通过,`git diff --check` 通过;142 个原始 URDF/mesh 文件仍与 HEAD 相同。
本次生成测试见 `test_all_model_artifact_pipeline.py``test_json_urdf_rebuild.py`
O6 最新采集及暂停证据在工作区 `calibration_output/O6_RIGHT_001/20260915_160040/review/`
最近一次完成采集并生成候选文件的证据在 `calibration_output/O6_RIGHT_001/20260915_150015/review/`
旧外参会话证据保留在 `calibration_output/O6_RIGHT_001/20260915_134211/review/`
下方分批测试数字为各次改造的历史记录。
2026-09-15 最后一次全包回归覆盖 1424 项:1408 项直接通过、15 项旧夹具兼容问题、1 项缺少实测记录跳过。
这些失败项已修复,按失败清单重跑 15/15 通过,没有遗留失败;这是全量运行加失败项复测的结果。
O12 使用当前 CAD 假设的完整生成/验收测试另行通过。构建与四型号安装入口检查通过,
142 个原始 URDF/mesh 文件与 HEAD 内容一致。
完整日志和复测日志保存在上述 O6 `review/` 目录的 `linkerhand_final_full_tests.log`
`linkerhand_final_rechecks.log``software_validation.json` 汇总了验证范围。
## 实现与坐标约定
`q_CAD = q_output + delta`。每个关节自己的实机 baseline 对应 `q_output=0`
不把某次局部参考的完整手姿态解释为全手零位。
生产流程仍为统一 runner、ROS 接线、协调器、任务执行器、公共拟合与发布器。
新增型号通过 Profile、SDK Adapter、Tag 观测关系与避让动作接入,没有新增型号专用算法。
| 职责 | 实现 |
| --- | --- |
| 显式关节零位配置、已知几何依据 | `core/domain/profile.py``profiles/loader.py``profiles/validator.py` |
| 不可变零位证据及持久化校验 | `core/domain/reference.py` |
| 避让后按声明姿态采集零位 | `runtime/execution.py``runtime/motion_execution.py``runtime/capture.py` |
| 会话/运动版本核对及冻结提交 | `runtime/coordinator.py` |
| 原零位核验后逐任务恢复数据 | `runtime/joint_resume.py` |
| 冻结参考下的相对运动与指令映射 | `core/fitting/motion_fit.py``command_mapping.py``anchored_mapping.py` |
| 明确声明的运动复制、零位迁移及近似来源 | `core/fitting/motion_transfer.py` |
| 独立线性 mimic 近似 | `core/fitting/mimic_approximation.py` |
| 空间零偏及授权 URDF 修正 | `core/fitting/session.py``spatial_solver/``core/urdf/result_plan.py` |
| v3 序列化、读取和最终文件验收 | `runtime/artifacts/``core/urdf/measured_acceptance.py``tag_acceptance.py` |
Start 清空预览后只锁固定参考。运动 Tag 轮到相应关节时才成为必需观测。
Profile 必须声明完整零位指令和非零的最终到达动作;扫描起点不等于 baseline 时先访问零位。
参考保存完整指令、反馈、到达方向、父子相对/公共位姿、图像身份、触发任务和会话/运动版本。
其他手指的非零避让姿态保持在证据中。无法分离同通道多个关节影响的观测在启动检查时拒绝。
同一关节只冻结一次。重复任务、方向、轮次、重扫及恢复不替换参考。
恢复先验证整场指纹,再在各关节声明姿态验证旧安装,保留原参考身份,通过后才导入已完成单元。
原生 rad 输入的第一轮行程统计也随已验证单元恢复,避免跳过第一轮后丢失覆盖依据。
迟到的预览、暂停前观测或旧运动结果不能进入当前参考。
零位稳定性判断使用各机位最近的足量独立帧,短暂异常帧会随新帧退出判断窗口;
原始记录保留,稳定性门限不变。
JSON 关节来自自己的视觉样本,或 Profile 明确声明的对应实测来源;同一 SDK 通道允许绑定多个关节。
接收关节保留自己的 SDK 通道,复制来源的指令/反馈节点及角度,不再拟合。
接收关节没有独立零位帧和第四轮指标,报告标记 `independently_measured=false` 并引用来源证据。
运动曲线只用前三轮,第四轮分别验证两方向;独立稳态分区用于指令映射训练与验证,约束 baseline=0。
O6 侧摆按已确认策略输出正反向指令曲线;其他关节沿用单值指令曲线。两方向共用同一参考。
验证数据不能用于重新拟合曲线、零位、Tag 安装或基座。
序列化只导出冻结结果;u8 必须实测支持 0–255 才能生成 256 项,rad 只在 `input_values` 范围内插值。
修正 URDF 保留各型号、本侧的原始 mimic,单独使用时提供近似联动;完整非线性角度由配套 JSON 提供。
`urdf_correction` v4 明确 `preserve_source_mimic`,正式标定不再拟合倍率或偏移,
不复制其他手指的 mimic 参数,也不因修正 origin 而重写原 mimic offset。准确 FK 显式使用 JSON 的各关节角。
`mimic_envelopes` 与实测 `motion_range_evidence` 分开:只在明确声明的实测范围策略下,
取实测范围与原始联动可达范围的最小包络;禁止突破独立安全界限及默认 CAD 硬界限。
多级 mimic 使用父关节真正的联动可达范围递推,不将其更宽的 JSON 运动范围继续放大。
JSON 重建与最终文件验收均核对原 mimic 元素、最小包络、通道一致性及非线性表。
旧 correction v1、移除 mimic 的 v2、拟合近似的 v3 保持原有重建/读取语义。
既无运动观测也无显式复制声明,或缺少绝对零位来源时,保留相对运动诊断并拒绝完整发布。
旧 mimic 及旧迁移字段不能自动授权 v3 复制,不用默认零偏补全未知几何。
接受原 CAD 末节零位的前提单独记录在 `cad_zero_assumptions`,不作为独立实测证据;
仅明确列出的关节取 δ=0,保持其自身 `origin`;O12 的三个额外关节假设见上方当前结果。
## 验收与消费
发布器从最终文件回读并分别检验:
1. 最终 JSON 对第四轮原始视觉角度;逐实测关节、逐方向检查;复制表另核验来源与一致性。
2. 最终 JSON+修正 URDF 的完整 FK 对第四轮 Tag 位姿;显式使用所有被动角。
3. 相同主动映射配合标准 mimic 的独立近似损失,单独报告。
前两项精度门限仍为角度 MAE ≤1°、P95 ≤2°、最大 ≤3°,Tag 位置 P95 ≤3 mm。
三文件哈希、保护输入、观测覆盖和通道绑定一并检查;拟合失败不自动运动,取消后不能提交。
混合产物的 manifest acceptance 为 `measured_and_transferred_json_urdf_v3`
明确记录观测关节、复制关节和 CAD 零位假设。未观测接收关节不具有独立精度结论。
公共读取器和 `calibrated_joint_state_bridge` 支持 v3 的完整关节映射及历史 v1/v2 原语义。
完整 JSON 回放使用 `independent_mimic_angles=True`,防止线性近似覆盖实测被动角。
标准 `robot_state_publisher` 和普通 URDF 网页查看器提供线性联动预览,不据此宣称实测非线性精度。
本轮未修改 MuJoCo、相机时间同步和接触动力学;现有 MuJoCo 脚本尚不能视为支持 v3。
## 当前型号的发布边界
| 型号 | 独立运动观测 | 声明的运动复制 | 绝对几何边界 |
| --- | --- | --- | --- |
| G20 | 21 / 21 个运动关节 | 无 | 五个末节的原 CAD 零位为明确假设 |
| L6 | 5 / 11 | 其余三指 MCP、DIP ← 小指对应关节 | MCP 零偏明确迁移;五个末节 CAD 零位为假设 |
| O6 | 5 / 11 | 其余三指 MCP、DIP ← 小指对应关节 | 同上 |
| O12 | 16 / 19 | 无名指 MCP/PIP/DIP ← 小指对应关节 | pinky_pip、ring_pip、thumb_mcp 及末节沿用已确认的 CAD 零位假设 |
`--validate-only``independent_measurement_coverage` 返回具体关节列表。
配置检查通过表示合同可执行,不能解释为全关节精度通过。
上述运动策略已经覆盖各型号全部运动关节;完整发布仍要求实测来源验收通过、零位有来源且范围合法。
四个产品配置的零位来源均已明确;L6/O12/G20 的原 CAD mimic/限位冲突保留诊断,
收尾保留限位约束内的线性近似;所有 JSON 查表角度和 URDF mimic 全行程都必须处于合法范围内。
现有 Tag 布局没有被补写成未经验证的新布局,新增参考动作也没有标为实机已验证。
## 原按关节零位改造的软件验证记录
修改前在原工作区测试路径上加载备份代码:**582 passed、1 skipped**。
固定基线保存在 `/tmp/calibration_v3_baseline_ck_q_c99/`,包含原包、输入哈希和运行日志。
旧数值回归继续显式使用 v2 语义,未通过重写黄金输出或放宽验收门限消除差异。
最终代码全量回归:**614 passed、1 skipped485.43 秒**。
日志:`/tmp/calibration_v3_final_verified.log`
包含最后补入的零位异常帧恢复用例;未放宽原有空间求解或精度门限。
已完成的额外检查:
- `colcon build --packages-select linkerhand_calibration --symlink-install` 成功。
- 安装后的 12 个 console entry point 均能导入;四个型号的安装入口 `--validate-only` 均以 0 退出。
- 两种合成布局生成的 v3 修正 URDF 另经实际 ROS `robot_state_publisher` 隔离加载通过。
- 对比修改前备份,142 个原始 URDF/mesh 文件与 16 个其他相机、Tag、设备配置文件内容不变。
- 四份 Profile 新增参考动作、输出版本及采集策略,对应 product 仅更新 Profile 哈希。
- `git diff --check` 通过;新实施记录随包安装,安装后的 README 文档链接检查通过。
主要回归场景:
| 场景 | 测试 |
| --- | --- |
| 四型号先锁固定参考、逐关节仅一次归零,食指在其他手指第四轮后开始 | `test_joint_zero_runtime.py` |
| 起点与 baseline 不同、非零保持姿态、错误到达/禁用通道、跨任务复用、rad 断点行程恢复 | `test_joint_zero_runtime.py` |
| 零位只采自身观测、短暂缺 Tag 丢帧、固定 Tag 缓存、姿态候选不重选冻结参考 | `test_observation_capture.py``test_o12_observation_resolution.py` |
| 任意固定 Tag 安装旋转、冻结零位、非线性被动曲线、第四轮滑移、共零点约束 | `test_measured_zero_mapping.py` |
| 最终非线性组合达标但 mimic 超差、文件篡改、通道错误、迁移/几何缺失禁止发布 | `test_measured_finalization.py` |
| 更换通道顺序、机位名、任务顺序的虚拟型号走同一完整发布链 | `test_measured_finalization.py``measured_capture_fixture.py` |
| 旧格式读取、标准 URDF 授权和结构、READY 后断流、迟到观测、后台反馈、中止与提交竞争 | 原有相关回归及扩展的 `test_calibration_coordinator.py` |
## 本轮观测策略扩展
全量回归:**700 passed、1 skipped495.34 秒**,日志 `/tmp/calibration_transfer_full_verified.log`
全量收集后补入的两项反馈范围一致性用例另行通过(`/tmp/calibration_transfer_feedback_domain_verified.log`),
本轮共覆盖 **702 项通过、1 项缺少实测数据而跳过**。未放宽任何空间求解或最终精度门限。
构建成功;安装后 12 个 console entry point 均能导入,四型号 `--validate-only` 均以 0 退出。
对照当前提交检查 142 个原始 URDF/mesh 文件,内容全部不变;四份产品配置仅更新对应 Profile 哈希。
新增回归见 `test_observation_strategy.py``test_motion_transfer.py``test_transfer_artifact_contract.py`
覆盖三种观测策略、虚拟型号改名/重排、接收通道、非线性被动表、来源与假设篡改拒绝、
O12 PIP 不补零、末节自身 origin 和各指 CAD 几何保持、机械范围及混合产物读回。
历史全实测合成夹具显式清空新的产品 CAD 假设,继续使用原非零合成真值;未修改其数值期望。
`test_measured_transfer_finalization.py` 进一步覆盖 O6 及重排后的虚拟 O6:
由 5 个关节的原始合成记录实际拟合,展开为 11 个关节,经过最终文件验收、发布及 manifest 读回;
6 个复制关节保留自己的通道,且没有被制造为独立观测。完整组合通过时仍报告 mimic 近似损失。
合成数据的已知 baseline 几何明确标注 `synthetic_fixture_truth`,不写入产品 Profile。
完整 v3 非线性回放覆盖合成 G20、O6 拓扑及重排后的虚拟型号;四型号执行器测试不等于四型号都有完整实测回放。
原 G20 两种布局的合成回放均得到:最终关节映射最大误差约 0.079°,完整 FK 的 Tag 姿态最大误差约 0.078°、
最差 Tag 位置 P95 约 0.025 mm;标准 mimic 的姿态最大误差约 5.465°,仍被如实报告。
这些数字用于检验软件是否分开评价非线性映射与线性近似,不是相机或实机测量结果。
O12 实测回放因未提供 `O12_REPLAY_RAW``O12_REPLAY_REFERENCE` 跳过。
没有运行实机动作,没有整只手精度、USB 延迟或多关节接触工况的实测结论。
复现命令:
```bash
source /opt/ros/jazzy/setup.bash
source install/setup.bash
PYTHONPATH=src/linkerhand_calibration:$PYTHONPATH python3 -m pytest -q src/linkerhand_calibration/test
colcon build --packages-select linkerhand_calibration --symlink-install
ros2 run linkerhand_calibration calibrate_hand --config src/linkerhand_calibration/config/o6_right_product.yaml --validate-only
```
+740
View File
@@ -0,0 +1,740 @@
# LinkerHand 多型号统一标定
G20、L6、O6、O12 使用同一个产品启动器、在线状态机、采集器、拟合器、标准 URDF
验收和发布器。型号差异来自 `config/profiles/*.yaml` 和 SDK Adapter,不再调用型号节点。
本包按观测关节冻结实机 baseline 零位,输出实测曲线及明确声明的复制曲线组成的 JSON v3,
以及保留线性 mimic 联动的修正 URDF;报告区分独立实测、迁移和 CAD 零位假设。
保留 CAD 连杆尺寸、轴位置、mesh、惯量。完整运动由 JSON 的全部关节角度配合修正 URDF 计算。
软件仿真通过不等于实机精度通过;当前原始 G20/L6/O12 存在需要确认的 mimic/限位冲突。
本轮实现与验证见 [v3 实施记录](CALIBRATION_V3_IMPLEMENTATION.md)。
## 启动
```bash
cd /home/lxp/projects/linkerhand_retarget_ros2
source /opt/ros/jazzy/setup.bash
colcon build --packages-select linkerhand_calibration --symlink-install
source install/setup.bash
# 无运动、无 SDK 启动的配置检查
ros2 run linkerhand_calibration calibrate_hand --config \
src/linkerhand_calibration/config/o12_right_product.yaml --validate-only
# 正式启动:自动启动 SDK、相机、检测及标定,READY 后自动开始运动
ros2 run linkerhand_calibration calibrate_hand --config \
src/linkerhand_calibration/config/o12_right_product.yaml
```
其他型号替换为 `g20_right_product.yaml``l6_right_product.yaml``o6_right_product.yaml`
AprilTag 检测参数直接读取产品指定的受保护 Tag YAML,统一启动文件不再覆盖其中的采样分辨率。
`calibrate_g20_right` 是同一 runner 的旧命令别名。不要同时运行 GUI、单独 SDK 或其他控制器。
`--commands-disabled` 只预览,不发送运动;`--no-resume` 强制新采集。Ctrl+C 中止,不自动快速张手。
重新开始时先安全回基准,仍必须确保现场没有障碍物。
### JSON 到修正 URDF
统一收尾顺序为:拟合与独立角度验证 → 保存标定 JSON → 重新读取 JSON 修正原始 URDF → 文件回读与整链验证 → 发布。
JSON 的 `urdf_correction` 保存原始 URDF SHA256 和必要的关节字段修正,重建不依赖内存拟合对象。
原始 URDF 不被覆盖;关节拓扑、字段授权、完整 mimic 范围和 JSON 查表范围在文件安装前检查。
新 v3 使用 `urdf_correction.schema_version: 4``joint_angle_source: calibration_json`
并记录 `mimic_policy: preserve_source_mimic`:所有型号保留本侧原始 `<mimic>` 元素,
不改来源、倍率、偏移或省略的默认属性;正式标定不再拟合 mimic 参数。JSON 保留各关节
非线性曲线及方向分支,精度验收使用 JSON 提供的全部关节角。父子链、轴和连杆不变,
被动关节与驱动关节仍绑定同一 SDK 通道,所有数值编辑仍按 Profile 授权。
`mimic_envelopes` 单独记录基础范围、原始联动可达范围和导出范围;在明确的实测范围策略下,
导出范围取二者的最小包络,实测范围证据与查找表不变。未声明该策略的 CAD 限位,以及
另行声明的硬件安全限位,仍是硬边界;冲突时报告具体关节和所需范围,不压低倍率或放宽容差。
关节 origin 按标定零位修正后,原始 mimic 作为输出坐标中的近似预览,不作为准确物理约束。
旧 correction v1、移除 mimic 的 v2、拟合线性近似的 v3 产物继续兼容读取与重建。
导出规则单独版本化,不改变采集配置指纹或旧原始记录,已发布产物不能原地覆盖。
已生成的 JSON 可单独重建对应 URDF,无需相机或 SDK:
```bash
ros2 run linkerhand_calibration rebuild_calibrated_urdf \
--config src/linkerhand_calibration/config/o6_right_product.yaml \
--calibration-json /path/to/calibration.json \
--output /path/to/new_corrected.urdf
```
输出路径必须尚不存在。没有 `urdf_correction` 的历史 JSON 不会被猜测或自动补全。
同一 JSON 和源 URDF 可逐字节重建相同修正文件;发布器会核对此项一致性,读取器要求相应验收记录。
历史无修正元数据的已发布文件继续沿用原有读取规则。
完整文件验收失败时保留候选 JSON/URDF 和诊断,便于复核;只有 `release_manifest.json`
和成功发布指针才代表通过,运行时读取器拒绝把未发布候选当作合格标定。
| Profile | SDK 位置通道 | 扫描任务 | Tag | 输出格式 |
| --- | ---: | ---: | ---: | --- |
| G20/right/g20_right_19/v1 | 20 | 16 | 19 | unified v3256 项指令查表 |
| L6/right/l6_right_8/v1 | 6 | 3 | 8 | unified v3256 项指令查表 |
| O6/right/o6_right_8/v1 | 6 | 3 | 8 | unified v3256 项指令查表 |
| O12/right/o12_right_16/v1 | 12 | 11 | 16 | unified v3rad 输入节点查表 |
任务数、主动关节数和实测关节数不必相等。一条运动可观测多个主动/被动关节。
观测策略由 `measurement.transferred_motion_sources` 声明,目标关节指向直接实测的来源关节:
| 型号 | 独立采集 | 复制的运动映射 |
| --- | --- | --- |
| O6/L6 | 拇指、小指 | 食指/中指/无名指的 MCP、DIP 分别复制小指对应关节 |
| O12 | 拇指、小指、中指、食指 | 无名指 MCP/PIP/DIP 分别复制小指对应关节 |
| G20 / 全关节模式 | 每个关节 | 不声明复制关系 |
复制保留接收关节自己的 SDK 通道、URDF 名称与 CAD 几何,不要求接收关节另贴 Tag 或采集零位帧。
复制参数不等于接收关节独立实测;接收关节仍须有明确的绝对零位来源及合法机械范围。
`zero.cad_zero_assumptions` 显式记录“实机 baseline 与原 CAD 末节零位一致”的使用前提,
相应 DIP(及拇指末节 IP)δ=0、自身 `origin` 不改。这是假设,不是视觉测得的绝对 CAD 朝向。
按用户确认,O12 的 `pinky_pip``ring_pip` 和主动 `thumb_mcp` 也采用原始 URDF 零位作为 baseline 假设。
运动曲线仍按实际观测/明确复制关系生成,报告标记 `assumed_source_cad_zero`,不称为独立实测零偏。
正式入口在驱动硬件前检查运动覆盖及零位来源,缺失依据时直接指出缺项,不采完才发现无法发布。
启动检查分别列出 `unmeasured_joints`(未独立观测)、`unresolved_motion_joints`(也未声明复制)
`missing_baseline_geometry`。旧迁移字段单独存在不会自动授权新的 v3 曲线复制。
本次沿用现有 Tag 和避让布局声明延后零位动作,没有把新增动作或缺失 Tag 布局标为实机验证通过。
每任务的运动模型采集四个往返,共八趟单程:前三轮训练,第四轮独立双向验证。
四个产品配置统一使用 `command_capture_mode: separate`:这八趟均为完整连续扫描,随后另做一组稳态映射训练往返和一组独立验证往返。
主扫描不逐点停顿;稳态点直接执行移动、稳定反馈、新图像采样,不重复等待两次到位。
稳态训练点由前三轮连续数据生成 `command_sampling_plan`:在原节点网格上寻找最少的
保留节点,各关节、各方向的预测曲线插值变化均不超过 0.25°;端点、基准、中间支撑点及
显式增加的非线性节点保留。缺少完整训练数据或曲线噪声过大时使用原网格。
这份计划只决定停在哪里,每个保留点的静态角度仍由新图像独立测量,不用连续运动角度
冒充静态角度。独立验证网格保持原样,不参与节点选择。计划及训练来源哈希随原始记录保存,
拟合、离线回放和断点恢复均重新核验;缺点、篡改或最终误差超限仍禁止发布。
上一段已经完成到位/稳定检查且下一准备目标完全相同时,可复用一秒内、反馈未变化的
完整保持姿态,省去重复准备等待;有实际位移、反馈变化或记录过期时执行原准备流程。
原训练网格每方向通常为 9 点,并保留行程内的 baseline;O6 小指额外包含 239,共 10 点。
新流程在这份网格上按上述规则精简训练停点。
验证使用交错点加端点(O6 小指包含 247),两类数据不混用。
日志中的运动轮编号为 0–3,独立指令映射训练和验证编号为 4、5;界面分别显示“稳态指令映射训练/独立验证”。
其他采用 `interleaved` 的配置仍在四轮各方向中穿插稳态点,并使用三轮稳态训练、第四轮稳态验证。
行进图像用于旋转/几何拟合,停稳后的新图像用于 SDK 指令映射;同一张图像不会复制成两类样本。
每点至少 3 个稳定同步图像,轨迹结束后最多等待 2 秒,不能用精确指令/反馈相等判断稳定。
关节零位冻结另需至少 10 张独立有效图像。`acquisition.joint_zero_timeout_seconds`
单独限定首张合格零位图像后的采集窗口,四个产品配置均为 5 秒;几何求解后的跟踪恢复仍最多等待 2 秒。
反馈不稳定会清空当前零位候选,但不会重置总采集期限;图像数量、1° 稳定性及分支证据要求不变。
稳定窗口使用每条新收到的 SDK 反馈,按消息时间戳去重,不再只取控制定时器看到的最新一条。
定时器延迟不能制造“独立反馈不足”;期间真实运动仍会使窗口失效。
候选清空时的 `steady_capture_reset` 保存反馈窗口和丢弃样本数量,便于区分抖动、断流和调度问题。
零位样本的 `capture_timing` 记录接收、历史快照和提交时间,用于检查处理延迟。
图像数量未齐时先返回缺样结果,不在控制定时器中反复校验全部姿态。
四轮及稳态采集完成、数据落盘后关闭图像采集;拟合只使用已保存的相机和参考证据。
之后的新图像和相机消息不再改变拟合输入,也不触发采集暂停;SDK 反馈、故障保护及操作者中止持续生效。
拟合或文件验收失败结束为 `FAILED`,保存 `failure_diagnostic.json` 中的失败阶段、原始数据路径和异常堆栈。
这类失败应先离线复算,不能仅因状态失败就重复四轮运动。运动采集期间的固定参考检查保持生效。
O6 的顶部 ID7 与正面 ID2 实际固定在同一末节,Profile 将 ID7 明确绑定到 `rh_thumb_distal`
`thumb_yaw` 是它承担的侧摆观测角色,不是其安装连杆名称。整链回放包含该 Tag 上游的侧摆、弯曲和末节角度,
保持关节沿用记录的到达方向。新绑定改变了 Profile 哈希,旧会话不会被自动复用为新配置的通过结果。
固定 Tag 的 ID 解码成功不代表边框完整。独立的视觉边框检查进程按完全相同的图像时间戳配对整流图像与检测结果,
检查当前检测四边的黑白边界支持;部分遮挡产生的畸变四边形会被记为 `detection_quality_rejected`
不能进入 PnP、关节样本或固定基准移动计数。配对缓存有界,不用邻近时间的图像替代。
检查使用当前检测四边形,清晰 Tag 发生真实位移仍进入原有 5 px/连续 10 帧保护。
控制进程只订阅 `detections_checked` 小体积观测;三路大图像的传输、配对和边框处理不占用 SDK 控制节点的回调队列。
ROS 宿主将准备几何求解和逐帧图像角度求解交给独立计算进程,避免与 SDK 反馈争用 Python 执行资源。
准备请求仍限时 120 秒;逐帧请求最多三个在途,超时或工作进程不可用的图像不生成有效样本。
工作进程只接收不可变图像/模型输入,没有 ROS 或 SDK 控制接口。返回后仍检查当前图像、跟踪版本和会话版本,
不能将迟到结果提交到下一运动段。`motion_solver_diagnostic.jsonl` 记录请求、计算耗时、线程 CPU 时间及提交结果。
当前任务需要的固定 Tag 无有效观测时,仍按原采样规则等待或暂停;其他机位的遮挡不误报为移动。
稳态点缺失与连续扫描不足统一在方向结束处理,最多同速重扫一次。
首次关节零位准备、避让及收尾属于独立的必要动作,不计入四个正式往返;不会每轮重新锁定零位。
分段加减速和停点仍有耗时,八趟行程不等于耗时恰好降为原来的三分之一。
新产品会话记录 `capture_schedule_version=unified_schedule_v3_separate_mapping`;旧 `interleaved` 配置仍记录 v2。
采集模式或保护配置不同的旧断点不与新流程拼接,
历史原始数据仍保留原语义供对应配置离线读取和分析。训练/验证身份与精度阈值保持不变。
字节拟合不再改写原始 command/feedback,也不将接近端点的读数伪装成 0/255。
## O6 右手拇指短程诊断
当前 O6 实物在 2026-09-16 复测的 Tag 黑色码区边长为 **16.5 mm**,棋盘格边长仍为
**27 mm**。O6 Profile、检测器配置和标定节点配置已同步采用 `0.0165 m`,产品配置记录
对应文件指纹。该数值属于这套实物标记,不是所有型号或所有打印批次的统一尺寸。
更换标记时应实测黑框,并同步维护上述配置;旧 16 mm 会话保留原尺寸记录,不作为
新尺寸的恢复断点。Tag 尺寸修正不改变 JSON 中非线性运动曲线与修正 URDF 的配合方式。
本套 O6 的 2026-09-16 实采数据已生成通过验收的产物,目录为
`calibration_output/O6_RIGHT_001/20260916_134154_source_mimic_165mm/`
`latest_partial_passed` 已指向该目录。JSON 和修正 URDF 应配套使用;当前 8 Tag 布局
直接测量拇指、小指,其余三指沿用配置中的曲线迁移关系。采集完成后的在线收尾曾发生
系统 OOM,最终产物由完整原始数据的正式离线流程验收;日志读取内存优化另经独立进程
验证,不能将此记录描述为在线全过程无中断。详细证据见 `CALIBRATION_PIPELINE_REVIEW.md`
此目录用同一份原始数据重新验收,5 组 mimic 的完整元素与原始 URDF 一致:
拇指倍率 **1.86**、其他四个末节 **0.89**,偏移均为 **0**
11 个关节的指令/反馈查找表及几何零位保持不变;仅拇指 IP 的 URDF 下限从
0.115187° 调整到 −0.183816°,覆盖原始联动所需范围,实测范围证据仍为原值。
拇指 SDK 指令 0 时,IP 的 URDF 预览为 **63.68°**JSON 为 **66.35°**
这是线性预览与实测非线性的区别。完整 JSON+URDF 和独立原始图像继续满足原精度门限。
`20260916_123347_json_driven_165mm/`(移除 mimic)及
`20260916_124718_mimic_json_165mm/`(拟合倍率和偏移)为历史版本,已被本次发布替代。
旧目录均保留;始终使用同一 manifest 指定的 JSON、报告和 URDF。
定位 Tag 姿态跳变时,可沿用相同产品配置添加诊断选项:
```bash
ros2 run linkerhand_calibration calibrate_hand \
--config src/linkerhand_calibration/config/o6_right_product.yaml \
--diagnostic-thumb
```
诊断选择原 Profile 中的 `thumb_yaw_top``thumb_pitch_ip_front`,各采一次往返,
观测拇指 CMC yaw、CMC pitch、IP 三个关节。保留原速度、完整扫描行程、准备/避让、
固定基准锁定、按关节尝试零位确认和安全收尾;其余手指不执行扫描任务,但开始时仍会安全回到 baseline。
本次诊断只启动正面和顶部机位:开始时一次锁定正面 ID0、顶部 ID6;运动观测使用顶部 ID6/ID7、正面 ID0/ID1/ID2。
侧面不参与本次诊断,ID3 缺失或双解、侧面相机未连接均不会阻塞;因此无需为拇指诊断调整 ID3 或改动相机外参。
范围由本次全部任务的主/辅观测依赖统一确定,启动、ROS 订阅、设备检查、基准锁定、指纹和进度使用同一范围。
它不会随当前任务切换而重新锁定;正面和顶部的配置变化、固定基准漂移及所需 Tag 数据不足仍按原策略处理。
正式全手标定继续使用 Profile 声明的全部机位和固定基准,外参文件及保护输入的完整校验保持不变。
“短程”缩短的是任务和轮次数,不截断需要复现异常的运动行程。
本模式跳过稳态映射采样与三轮训练/第四轮验收,收尾后显示 `DIAGNOSTIC_COMPLETE`
并正常退出。诊断按当前运动 Tag 的原始图像覆盖推进,持续姿态歧义不阻断原始候选采集;
缺少当前运动 Tag 角点或同步反馈导致方向覆盖不足时,仍最多同速重扫一次,严重故障和中止行为不变。
归零准备证据为 `unresolved` 时,仅诊断模式允许继续收集原始观测:必须先到达 Profile 声明的零位指令,
SDK 反馈稳定,当前运动 Tag 具有足量、带会话和运动版本的图像,才保存独立的 `diagnostic_joint_zero_attempt`
它表示访问并观察了声明姿态,不表示已建立关节零位;不写入 `zero_references`,方向切换及重扫不重新归零。
该组关节后续仅保存原始角点和候选,即使普通跟踪恢复,也不能生成 `joint_sample` 或伪造有效角度。
缺少零位原始图像、SDK 不稳定、固定基准移动及已冻结参考冲突仍按现有策略暂停;正式标定仍须先确认分支和冻结零位。
会话目录保存 `raw_samples.jsonl`(含角点和 IPPE 候选)、实际成功冻结的关节零位与 `diagnostic_capture.json`
`diagnostic_observation_frame` 在零位尝试和扫描阶段单独记录各帧候选,包括未接受的姿态,不能当作关节角度样本。
报告 `unit_quality` 分别记录原始观测覆盖与 `accepted_pose_quality` 可信姿态覆盖;
`joint_zero_status``missing_joint_zero_references` 列出各关节真实零位状态,`motion_initialization_summary` 保留各候选的关节残差。
进度显示原始观测覆盖,并单列可信姿态数量和未确认关节;暂停时保持“已暂停”,不会被零位动作名称覆盖。
可信姿态不足会在退出时明确提示。相邻帧姿态跳变和图像身份用于定位问题,诊断完成不代表标定精度通过。
每方向的 `observation_timing` 分别统计原始/可见/同步图像数、缺命令与缺反馈图像数、最大图像及同步观测间隔;
缺命令与缺反馈可以发生在同一帧,不能相加当作总丢帧数。少于两帧的间隔记为 `null`
方向失败时这些信息同时进入暂停诊断。反馈跨度 100% 仅表示两端都有观测,不代表中间没有缺口。
原始帧的 `timing` 保存回调接收、历史快照及最新命令/反馈时间戳,用于区分消息延迟与状态匹配问题;
接收到快照的时间包含线程调度和锁等待,不能单独解释为写盘耗时。未匹配反馈时同步误差为 `null`,不会显示为 0 ms。
可额外加 `--record-bag` 保存现有录制话题;仅检查入口与配置可加 `--validate-only`,不会启动硬件。
诊断始终完整新采,不复用断点,不拟合或发布标定 JSON/URDF,也不更新发布指针。
原始会话带有诊断标记,正式离线拟合和断点发现会拒绝将其用作正式标定数据。
当前选项只开放给 O6 右手;其他型号仍走原来的正式流程。原始 URDF、Profile 和设备配置无需修改。
诊断任务预设在 `profiles/diagnostics.py` 声明,运行时只选择既有 Profile 任务,复用公共调度与采集组件。
诊断采集完成后,runner 先关闭相机和 SDK 进程,再自动运行公共图像几何分析,生成独立的
`diagnostic_analysis.json` 和可直接阅读的 `diagnostic_analysis.md`。耗时优化不进入实时状态锁或相机回调。
分析按任务、机位、方向和尝试分别保留原始图像身份,比较所有候选族的三维 PnP 残差、深度误差比例、
图像域关节链拟合以及隔帧和前后区间验证。验证时几何参数冻结,只从当前图像反解几何角度,不使用 SDK 角度映射。
图像误差小不等于毫米精度通过,优化收敛也不代表分支唯一;分析结果不能写回跟踪器、零位或正式标定产物。
未采到的方向和旧会话缺少的归零前证据会明确列出,不拼接不同会话,不使用第四轮重新拟合。
图像分析失败或被取消时,已完成的采集及原始文件仍保留,不会改标成 SDK 故障。
已保存的数据可直接离线复算,无需启动相机或重复运动:
```bash
ros2 run linkerhand_calibration analyze_calibration_capture \
--config src/linkerhand_calibration/config/o6_right_product.yaml \
--raw calibration_output/O6_RIGHT_001/20260914_192637/raw_samples.jsonl \
--output calibration_output/O6_RIGHT_001/20260914_192637/image_review
```
这些修改解决诊断采集被未确认零位阻塞及报告不完整的问题。现有真实 IP 数据仍有分段几何不一致,
正式分支门槛、拟合器、URDF 修正和独立精度验收保持原要求;软件回归不能替代实机精度结论。
### O6 右手 yaw 方向差保持诊断
需要区分“到达指令后仍在缓慢稳定”和“保持后仍存在方向差”时,使用独立选项:
```bash
ros2 run linkerhand_calibration calibrate_hand \
--config src/linkerhand_calibration/config/o6_right_product.yaml \
--diagnostic-yaw-hold
```
此模式仅选择 `thumb_yaw_top`,只启动顶部机位,使用固定基准 ID6 和运动 Tag ID7。
每次 Start 都重新采集会话基准并尝试确认 yaw 零位,不复用旧会话;同一会话的两个方向共用首次冻结零位,
不会分别归零。未能确认零位时按诊断规则保留原始观测,并明确标记无法生成可信角度。
在原速度和行程下执行一次往返。每个方向扫描后先回到该方向的起点,再沿同一方向访问三个保持点:
递减方向为 `255 → 192 → 128 → 64`,递增方向为 `0 → 64 → 128 → 192`
每个保持点在指令轨迹完成后至少保持 10 秒,两方向合计六次保持;反馈确认可能使实际等待稍长。
保持过程继续记录图像角点、候选、实际指令和同步反馈,用于比较到达后的角度随时间变化及双向最终差异。
保持阶段的原始观测单独标记,不计入连续扫描的覆盖统计;遮挡、设备故障、中止及安全收尾仍沿用公共策略。
`--diagnostic-yaw-hold``--diagnostic-thumb` 互斥,不能同时使用离线拟合、发布或禁用运动选项。
可加 `--validate-only` 只检查入口与配置,或加 `--record-bag` 留存现有录制话题。
结果保存在 `raw_samples.jsonl``diagnostic_capture.json`,记录保持指令及时间;
不拟合或发布标定 JSON/URDF,不形成正式断点,也不算精度验收。
本诊断用于查清实机稳定时间和方向差,不改变最终单值 `angle_rad` 的验收要求。
## 启动前相机检查
`--validate-only` 检查相机配置文件、序列号、内外参指纹和已有外参质量记录,不采集现场图像。
正式启动还检查实时 `CameraInfo` 的宽高及 K/D/R/P 是否与受保护外参中的记录一致;
参数不匹配、无效或消息过期时不能开始,并显示具体机位和原因。
**配置匹配不等于现场内外参有效。** 当前只有手上的 Tag,各机位使用不同 Tag,
没有声明跨机位共视或已知的 Tag 间几何关系;手掌和贴纸又允许在开始前移动、重贴。
因此光学内参与相对外参的现场有效性记录为 `unverified`,不能用重新锁定掌心 Tag、
较小的 PnP 残差或未变化的配置哈希代替实测,也不能由 Tag 位置变化断定相机被碰。
`READY` 只表示设备与消息满足现有开始条件。显示上述能力提示后沿用现有启动行为,
不要求各手指 Tag 可见,不增加人工确认或新的暂停条件;`READY` 不表示物理相机参数已验证通过。
会话目录中的 `camera_preflight.json``raw_samples.jsonl` 保存检查结果与依据,
开始前仅在结果变化时记录,接受 Start 前再检查并冻结一次;后续状态中的检查结果指这次启动检查。
单凭这些记录不能判定是否需要重新标定。`expected_cameras` 是配置期望身份,不是现场实测结果。
镜头、焦距或成像设置变化后应重新核验内参;相机之间相对位置变化后应重新核验外参。
未来同一个物理 Tag 被多个机位共同看到时,可以检查相对外参一致性,但当前尚未实现该能力。
本次检查不修改 Profile、SDK 或原始 URDF。
## 基准、避让与恢复
- 开始前可以调整手掌、相机和 Tag;预览数据不会成为正式参考。
- 开始后清空预览缓存,先安全回基准,再为每机位固定 Tag 收集至少 10 个有效帧。
缺少基准时保持并显示缺少的 ID,不因为等待时间长暂停。
- 锁定后禁止人工移动手掌、相机、支架、Tag 粘贴位置;程序驱动关节和避让仍是允许的。
- 全手 baseline 时只要求固定参考可见,不要求四指运动 Tag 同时可见。
每关节在 `motion.joint_zero_references` 指定任务前,按声明的到达姿态和避让分组回到自身 baseline,
反馈稳定后收集至少 10 个稳定视觉帧,冻结 `JointZeroReference`。此后方向、轮次、重扫均不重新归零。
例如 G20 食指开始前,已完成四轮的其他手指可以保持非零避让;这些完整指令和反馈写入参考。
task.start 不代表零位,例如侧摆 baseline=127、扫描起点=255,会分别执行两种姿态。
- 相机相对位置改变时须重新确认/标定外参;仅重新锁定掌心 Tag 不能替代外参标定。
- 固定 Tag 被避让遮住时可用已锁定参考继续采集;软件不能保证检测完全不可见对象的移动。
运动 Tag 的刚性和几何一致性仍须通过最终独立验证。
- 采集断点版本为 `unified_engine_v8_all_view_images`;另以 `pose_tracking_policy_version=confirmed_branch_v8_training_model_selection`
记录姿态分支策略。缺少此字段或版本不同的旧断点不能自动或手工恢复到新会话,历史文件保持原样供离线诊断。
首先检查配置哈希与整场固定参考;各关节轮到时,在同一完整保持姿态核对原零位与父子 Tag 位姿
(5 px、2°、5 mm),通过后才导入对应已完成扫描单元,并保留原零位记录。
整场参考不匹配则放弃旧数据;逐关节安装验证失败则暂停,禁止在旧样本上新建零位拼接。
排除原因后用 `--no-resume` 开始新会话。旧会话文件不修改。
旧日志的读取、凭据回放与单元质量检查在实时回调启动前完成,不占用运动/反馈回调。
现场验证通过后,每个控制周期最多导入 64 条历史记录;全部落盘后才跳过对应单元。
导入期间持续处理反馈、固定基准监测和中止,原 1 秒反馈超时保护不变。
连续图像、候选和样本按图像批量追加并刷新到系统缓存,避免持有控制状态锁逐条 `fsync` 阻塞反馈和运动定时器。
固定参考、关节零位、方向完成和暂停记录仍强制同步到磁盘;拟合、报告哈希、发布和关闭前另有同步屏障。
断点只复用通过且持久化的完整方向,意外断电时未完成方向不能拼入旧数据;同步失败不得发布。
此存储策略不改变 50 ms 图像/状态配对门限、方向覆盖门限、拟合公式或精度验收条件。
O6/L6/G20 标定读取 SDK 的 `*_hand_state_timed` 反馈接口;位置与每个通道的 CAN 接收时间
一同保存。G20 五指的分帧反馈分别插值到图像时间,重复发布的缓存不能刷新反馈超时或增加
稳态独立样本。标准 `JointState` 仍可供其他程序使用,其时间戳保留测量时间。
`sdk_feedback_sample` 原始记录支持复查时间配对;旧版本只保存发布时间,不能事后补成真实
CAN 测量时间,也不能与新版断点混用。O12 继续使用独立 HCAN 同步读取接口。
共同坐标约定:`q_CAD = q_output + delta`;每个关节在自身冻结 baseline 时 `q_output=0`
局部零位通过采样时的完整保持姿态和运动链进入同一掌部参考。固定 Tag 不提供未知末端连杆的 CAD 朝向。
`zero.known_baseline_geometry` 可声明有独立依据的 CAD 角度、误差界与来源;不能填入猜测的 0、旧 mimic
或 Tag 安装偏角冒充几何证据。相应 `origin.rpy` 和限位修正仍须在 Profile 中明确授权。
O12 避让:小指和无名指到 Profile 的最大弯曲指令;标定食指前,中指 MCP/PIP 到最大弯曲指令,
中指侧摆为 0 rad。到位采用完成轨迹、正确方向、至少 80% 请求位移和短时稳定,
不要求反馈数值精确等于命令。SDK 指令最大值与实测 URDF 角度不是同一个量。
O12 使用 HCAN device 0/channel 0 的已锁定 vendor Python wheel,不检查 `can0`
连接参数由本包的 `config/o12_sdk.yaml` 保存,并通过产品配置中的 SHA256 校验;
不依赖第三方 SDK 子仓库中未提交的示例 YAML 修改。
只读反馈桥可在第一次运动命令之前取得状态,不会发送零位命令来“激活回读”。
POSITION 和活动故障由 SDK 确认;无法单独回读温度时使用 SDK 错误位过热保护,不伪造温度。
字节 SDK 当前没有独立硬件故障遥测,状态会明确注明这一能力边界。
## 哪些情况暂停
实时仅因人工中止、竞争控制器、活动硬件故障/模式错误、真实失联、反馈超过 1 秒未更新、
物理越限、明显运动要求下连续 2 秒无推进、已锁固定基准连续 10 帧漂移超过 5 px 而暂停。
开始后相机投影参数变化也会拒绝继续使用混合坐标数据。
当前关节持续缺失零位观测会暂停并列出关节、机位和所需 Tag;尚未轮到的手指遮挡不阻塞当前任务。
零位和稳态点的反馈稳定窗口以最新独立反馈为终点,并保留跨越窗口起点的一条反馈。
控制定时器的调度抖动或重复读取同一反馈不会清空稳定观测;真实反馈波动或反馈过期仍会使当前窗口失效。
零位仍需至少 10 帧且姿态波动不超过 1°;拟合验收门限不变。
短时 Tag 丢失、PnP/同步失败、低检测率/低反馈 Hz、正常跟随滞后、固有耦合和非目标小幅运动
不会实时中断轨迹。无效视觉帧直接丢弃。
所有型号的独立 Tag 姿态筛选共用 `core/geometry/pnp.py` 中的近似同误差阈值,默认 `0.03 px`
ROS 和离线采集使用相同默认值,型号 YAML 不重复声明该值。`1.5 px` 是图像质量拒绝阈值,
不能用它判断两种姿态是否同样可信;仅在误差差值不超过近似同误差阈值时使用时间连续性。
分支容差覆盖值必须有限、非负且小于图像质量拒绝阈值。
独立 Tag 首次接受姿态前,需要默认至少 5 帧、跨越至少 0.1 秒的图像证据确认。
近似同误差的候选只有旋转相差不超过 1°、平移相差不超过 1 mm 时才视为同一姿态;这些参数统一定义于
`core/geometry/tag_pose/parameters.py`ROS 通过 `pnp_branch_confirmation_frames`
`pnp_branch_confirmation_minimum_seconds``pnp_branch_equivalence_deg` 加载,离线采集使用相同默认值。
这只是图像分支的一致性确认,不能把多帧低重投影误差视为物理姿态真值。
以下独立图像跟踪规则保留用于预览和历史兼容;新会话的运动 Tag 首次零位还必须通过后述公共运动几何确认。
参考冻结前,明显更优但与历史姿态发生大跳变的新分支也必须经过确认,过渡帧不进入零位或拟合样本。
运动 Tag 参考冻结后,若原分支仍有连续、图像质量合格的候选,但另一候选的重投影误差明显更低,
该帧标为 `pose_branch_ambiguous` 并丢弃,内部保留原分支轨迹等待图像证据恢复;
不能仅凭另一候选连续数帧得分更高,就判定冻结参考损坏。不会切换分支、重建零位或把歧义帧送入拟合。
如果持续不存在符合原分支连续性约束的候选,仍在确认后暂停;长期失联后无法确认分支一致时也会明确拒绝。
正式标定方向结束仍检查可信样本、覆盖与稳态点,最多同速重扫一次;诊断模式另报原始观测与可信姿态覆盖。
需要更换分支或重新安装 Tag 时,必须重新开始新会话,
不能把旧冻结参考与新分支数据拼接。原有 0.03 px 容差、35° 跳变、75° 倾角及 1.5 px 重投影质量阈值保持不变。
真实角点噪声、采集时序和最终运动精度仍须通过独立实测验证。
### 首次零位前的公共运动几何确认
新生产会话复用 Profile 已声明的**最后一段归零准备运动**,在 `zero_approach` 中保存父子 Tag 的原始候选,
`joint_zero` 后先确认分支,再采集原有的稳定零位帧。没有增加扫描轮次、额外试探动作或型号专用求解器。
几何确认与 SDK 反馈稳定分别检查;几何通过不会提前开启零位采集窗口。
计算在状态锁外进行,期间持续接收 SDK 反馈并保持声明姿态;计算本身有 120 秒上限。
几何确认后,先在原有 2 秒观察窗口内等待恢复后的首个有效零位样本,再用
`joint_zero_timeout_seconds` 限定的独立窗口采齐稳定帧(字段默认 2 秒,四型号产品配置均为 5 秒)。
窗口不会无限延长。正式采集中,每组尚未冻结零位的关节最多自动恢复一次:可恢复的图像
准备失败重新执行该组原有的归零准备路径并采集全新图像;几何已经冻结但零位图像不足时,
保持原姿态和模型,只重新开启一次零位图像窗口。原始失败和恢复动作写入 `joint_zero_recovery`
已冻结的零位、部分已经提交的多视角几何、保持姿态变化、设备故障不重新拟合;恢复后仍
不足时明确暂停,保留原有反馈、固定基准保护及用户中止能力。
中止、暂停或运动版本改变后,迟到的计算结果不能提交。
前面的避让段、预览、训练扫描和第四轮不参与首次确认。准备段同时核对其他通道的完整保持指令及反馈;
开始后不能改变相机、固定基准或 Tag 安装关系。
生产入口 `core/geometry/tag_pose/production_image_motion.py` 枚举 IPPE 候选家族,
使用父子转动副 `R = exp(theta × axis) R_ref``t = pivot + R × mount` 拟合原始角点的重投影残差。
稀疏线性子问题使用 `atol=btol=1e-10`,避免近似计算误差导致外层耗尽迭代预算;
外层收敛条件、默认 150 次迭代预算和各项精度门限保持不变,记录每个候选的求解状态及迭代次数。
父 Tag 本身运动时联合求解整条已声明的观测链;已冻结的上游几何不会重新拟合。
角度来自当前图像,不使用 SDK 到角度的先验、旧 mimic 或 CAD 零位假设选择分支。
单帧 PnP 深度噪声较大时,不再将其三维平移估计直接作为关节运动的真值。
当源 URDF 中两个转轴相邻且平行,并且观测 Tag 实际固定在对应连杆上时,
`profiles/observations.py` 提取零位修正后仍不变的平行关系和垂直轴距。
`CadImageHingeBundle` 消去三个多余几何变量,使准备模型遵守最终 URDF 保留的几何;
不修改原骨长/轴向、不增加权重或 CAD 零位假设。不满足条件的观测链使用原模型。
新冻结模型保存源 URDF 哈希和约束,最终处理从源文件重新推导核验;旧模型保持可读。
修改前的原始图像对照实验及适用边界见 [流程评估](CALIBRATION_PIPELINE_REVIEW.md)。
准备段至少需要 24 帧,训练与留出图像身份分离,默认各最多 72 帧。
训练帧取交错序列,未进入训练的图像才可留出;不因训练帧增加而复用验证图像。
局部几何可观测性检查计入逐帧角度的不确定性,轴线与转心的不确定度界分别不超过 3° 和 3 mm。
候选只按训练图像的平方误差选择;留出数据不用于选优,所选候选验证失败不能改选另一候选。
候选家族还须在准备段留出图像上区分:按运动角度分组,比较与拟合一致的平方误差损失,
差值为 `alternative_rms² - (candidate_rms + 0.03 px)²`,保留原像素容差。
同时通过单侧均值和符号秩检验,再对完整比较集采用 Holm 逐步校正,名义族错误水平保持 0.01。
`image_model_selection.py` 统一负责候选选择、同帧角色对齐、损失比较和多重校正;
诊断记录原始/校正后概率、选择策略和实际源 URDF 约束,失败时也保留这些证据。
缺失、未收敛或无法评估的替代家族不能被当作已排除;几何等价的家族才可合并。
这些检查是图像几何的条件性证据,不是独立毫米精度证明,不能替代第四轮 JSON/URDF 验收。
确认成功后冻结轴线、参考旋转、转心、安装关系及其来源图像。
`image_motion_projection.py` 后续每帧只反解当前图像角度,不重新拟合这些几何参数。
保留 1.5 px 重投影、75° 倾角及单帧角度不确定度检查;信息不足或角度仍有歧义时丢弃该帧,
按现有方向覆盖和最多一次同速重扫处理。35° 连续性和失联后的确认仍由公共 tracker 检查。
Tag 中心接近转轴本身不再是必然失败条件,是否可解取决于全部角点提供的几何信息。
准备段暂时缺帧或候选不确定时,每组尚未冻结零位的关节最多沿原归零路径补采一次;
已确认几何但零位图像短缺时,保持几何和姿态,只重新打开一次图像窗口。
两个恢复方式共享一次额度,保存 `joint_zero_recovery`,不延长成无限重试。
多机位已有部分几何冻结、保持姿态改变、设备故障和诊断采集不会触发重新拟合。
恢复后仍无法确认时不冻结零位、不发布;已有零位和旧数据不能被重新解释后拼接。
`motion_branch_observation` 保存准备段图像、角点、候选及 SDK 状态;`motion_branch_initialization` 保存各假设残差、
冻结模型、来源图像和内容哈希。新姿态来源为 `image_constrained_current_image`,保留原始 IPPE 候选,
同时记录 `frozen_image_model``image_model_sha256`。每个运动 Tag 的 `motion_evidence_id` 绑定首次证据,
`JointZeroReference.branch_references` 将它与零位、机位和分支版本关联。
正式读取与收尾校验这些关联,缺少证据、篡改来源或删除表头均不能降级为旧图像确认;
`motion_branch_provenance.json` 只证明证据链完整,不是精度证书。
恢复时共享来源只导入一次,关节零位及历史验证记录在该关节实际验证后导入,支持恢复期间再次中止。
正式回放用同一冻结模型复算同帧角点,核对父链、相机、Tag 尺寸及实际提交姿态,不重新拟合模型。
`confirmed_branch_v1/v2/v3` 及 v4 三维候选约束记录仍按原语义读取,不能自动升级或恢复成新 v5 证据。
`motion_evidence.py` 的旧三维筛选入口保留供历史兼容和诊断。这里 v5 仅指姿态策略,最终标定 JSON 仍为 schema v3。
真实 O6 `20260914_175815` 的旧三维筛选只读回放中,yaw 与 pitch 分别存在唯一通过几何条件的家族;
但 pitch→IP 的四种父子组合全部超差,IP 转心 RMS 约 4.9–11.4 mm,完整拇指链仍为 `unresolved`
这批图像来自旧零位冻结之后的单轮诊断扫描,只能用于反事实回归,不能代替新准备段证据、
覆盖旧零位或证明 JSON+URDF 精度。
可复算摘要见[本次分支修复回放](../../calibration_output/O6_RIGHT_001/20260914_175815/motion_branch_fix_review/README.md)
测试夹具同时保留“yaw/pitch 可分”和“IP 不得强行通过”两类断言。
后续 `20260914_200044` 的图像求解回放使用其真实归零准备段冻结模型,后续 838 帧可复算通过当前图像检查;
这仍是单轮诊断回放,不是完整四轮精度验收,也不会生成正式 JSON 或 URDF。
`20260915_104446` 正式实采中,拇指三个关节各冻结 10 帧零位并完成四轮,
pitch/IP 的旋转及指令单表子项均通过;yaw 旋转通过,但指令单表 MAE 为 1.0454°,超过 1°。
两场独立会话均观察到稳定方向差,该批观测即使允许每个指令任取一个输出角,MAE 下界仍超过 1°。
这不是通过重新拟合单表能够消除的误差;方向分支的诊断通过也不能冒充单表通过。
详见[拇指实测子项报告](../../calibration_output/O6_RIGHT_001/20260915_104446/motion_holdout_review/README.md)。
后续 `20260915_111401` yaw 保持实测在 64/128/192 各双向保持 10 秒;末 3 秒的方向差
分别为 2.2965°/2.7885°/2.1248°,每个窗口均有 90 帧可靠观测,同一目标的 SDK 反馈相同。
0.5–1 秒至末段的角度变化最多约 0.038°,延长等待没有消除方向差。
本轮仅启动顶部机位,不能据此确定具体机械原因或证明整手精度;继续保留单值 JSON 与原验收门槛。
复现脚本、时间曲线与来源哈希见[双向保持实测](../../calibration_output/O6_RIGHT_001/20260915_111401/yaw_hold_review/README.md)。
`20260915_104446` 会话在小指准备求解超时后停止,未完成整手空间/FK 验收、未发布 JSON/URDF。
小指同源准备段离线求解只需约 3.42 秒,但候选仍未达到统计区分要求;运行时超时来源尚待定位,
详见[小指准备段核查](../../calibration_output/O6_RIGHT_001/20260915_104446/side_preparation_review/README.md)。
后续会话另存 `motion_solver_diagnostic.jsonl`,记录请求来源哈希、版本、求解耗时和提交接受/拒绝,
以区分求解未返回与迟到结果被丢弃;该文件不参与测量授权,也不修改已冻结的原始记录。
此诊断接线在上述实采之后添加,不能补证该次超时原因。
固定基准另有明确的静止约束:锁定后始终使用首次冻结的相机局部姿态,将其投影到当前角点验证,
不因 IPPE 两支重投影误差排名反转而更换姿态或暂停。仍执行原有 1.5 px 重投影和 75° 倾角限制,
并保留当前 IPPE 候选与原有角点漂移监测。当前图像无法验证冻结姿态时丢弃依赖该基准的观测,
记录 `fixed_reference_pose_unverified`,不将几帧图像质量不合格判为不可恢复的姿态分支冲突。
恢复须默认至少 5 个合格帧且跨越至少 0.1 秒,始终核验同一冻结姿态,恢复完成前也禁止缓存兜底。
尚未使用该机位的动作可以继续;当前任务持续缺少有效数据时,仍按原有零位超时、稳态采样和最多一次同速重扫处理。
固定基准实际连续 10 帧漂移超过 5 px 仍使整场暂停,包括当前任务未使用的机位。
此约束仅用于 Profile 声明的 `fixed_reference`
不用于随关节运动的 Tag,也不证明初始分支物理正确或当前外参已经实测验证。
方向结束才检查 ≥40 个有效同步样本、≥32 分箱、内部空白 ≤行程的 1/16、规定的有效行程覆盖,
以及 Profile 声明的端点/双视角观测。数据不足只同速重扫一次;第二次失败暂停。
拟合或最终文件验收失败不重新自动运动、不更新发布指针。
`independent_rotation_holdout_failed` 表示前三轮冻结旋转模型未通过第四轮检查,发生在空间求解前。
错误原因包含失败关节、机位和 MAE/P95/最大误差;`fit_diagnostics.json.rotation_holdout`
另存完整 SO(3)、轴向、偏轴统计及最坏图像身份。验收使用完整旋转误差,不能仅凭投影角度曲线重复判定通过。
新采样行的 `pnp_observation_evidence` 保存父子 Tag 的当前角点、尺寸、已选姿态和候选诊断,
相机矩阵引用同会话的 `rectified_camera_model`。缓存固定基准明确标注来源,不伪造当前帧的角点或候选。
可见且通过冻结姿态投影验证的基准标为 `verified_fixed_reference`,保存当前角点、IPPE 候选和实际投影误差;
完全未检测到时的缓存仍标为 `cached_fixed_reference`。候选回放保持同一冻结基准并复核前者的当前角点证据。
短程诊断报告以 `fixed_reference_summary` 单独统计基准的当前图像投影误差,
不会把固定输出姿态的零跳变误当作独立测量结果。
不合格姿态事件另外记录当时的动作阶段、指令与同步反馈,便于区分准备动作期间的异常与正式采样异常。
角点与候选记录用于诊断;其中分支状态、版本及实际选中姿态也用于核验冻结参考的一致性,
不使历史采集自动获得缺失的观测证据。O12 的轨迹几何评分仍只使用训练轮,
但正式收尾不得用评分更好的另一支替换已冻结的训练或第四轮姿态;超过同一公共等价界时明确拒绝发布。
历史记录仍可在离线诊断入口分析候选,不能借旧格式标记绕过正式产物的冻结分支检查。
所有型号显示相同进度:阶段、任务/轮次/方向、分机位 Tag ID、命令/反馈/单位/速度、
有效数据、基准、断点来源,以及中文原因、建议和原始诊断。
ROS 收尾拟合在独立进程执行:只读取采集结束时已同步的日志前缀,后续图像不进入拟合;
主进程继续接收反馈并检查固定基准,阶段事件按原顺序回传,最终发布仍由主进程持锁完成。
中止时终止计算子进程,不允许子进程更新发布指针。`finalization_timing.json` 保存实际耗时。
标题明确显示型号、左右手和序列号。开始前等待状态列出缺少或过期的相机内参、检测消息和 SDK 条件;
空 Tag 检测消息可以证明视觉链路工作,不要求开始前已经识别全部 Tag。
收到开始请求时重新检查设备状态;实时 CameraInfo 参数必须与受保护外参中的指纹一致。
开始前相机消息的新鲜度窗口为 2 秒,该窗口不用于扫描中的实时暂停。
## 唯一生产代码链
```text
linkerhand_calibration/
├── config/profiles/ 唯一型号定义(受保护 YAML)
├── config/*_product.yaml 设备、序列号、相机、路径与哈希
├── urdf/ 原始 CAD/mesh,不覆盖
└── linkerhand_calibration/
├── profiles/ 加载、静态验证、Tag/CAD 观测关系
├── core/domain/ Profile、样本、CalibrationResult、状态
├── core/geometry/ 相机、变换、PnP 兼容导出
│ └── tag_pose/ IPPE、连续跟踪、冻结前运动证据、刚性组和轨迹选择
├── core/fitting/ 运动曲线、线性耦合、固定安装
│ └── spatial_solver/ 空间观测、基座、零位、统计与独立验收
├── core/urdf/ 授权修正、FK、范围与最终文件验收
├── runtime/runner.py 唯一产品启动器
├── runtime/coordinator.py 会话协调、采集提交、运动和收尾接线
├── runtime/parameters.py 已解析的运行参数;无 ROS 类型
├── runtime/inputs.py 相机/检测输入、时钟与发布接口
├── runtime/cameras.py 受保护相机身份与消息新鲜度
├── runtime/camera_preflight.py 启动前配置检查结果与内外参实测能力边界
├── runtime/session.py 唯一在线状态机
├── runtime/execution.py Profile 任务与运动效果调度
├── runtime/motion_execution.py 平滑轨迹与位移到位判定
├── runtime/capture.py 多机位视觉/反馈采集
├── runtime/branch_initialization.py 最后归零准备段证据与一次性几何确认
├── runtime/branch_tracking.py 冻结约束选候选,原子提交同帧姿态与证据
├── runtime/motion_provenance.py 读取和收尾核验来源内容、零位与样本关联
├── runtime/scan_quality.py 唯一方向数据门
├── runtime/reference_lock.py 开始后固定基准
├── runtime/resume.py 整场断点证据检查
├── runtime/joint_resume.py 逐关节原零位验证后导入旧样本
├── runtime/safety.py 最小实时保护
├── runtime/status.py 统一状态与中文显示
├── runtime/snapshot.py 类型化状态快照与旧状态映射
├── runtime/adapters/ SDK 协议,不包含标定业务
├── runtime/ros/ ROS 参数加载、消息转换、订阅/服务/定时器
├── runtime/artifacts/ 收尾控制、后台 finalizer、serializer、原子发布
└── compat/ 历史格式、布局与离线诊断;不提供旧在线流程
```
`calibrate_hand → runner → runtime/ros/entrypoint → UnifiedCalibrationNode → CalibrationCoordinator → SessionExecution`
是所有型号的生产调用链。节点直接继承 ROS `Node`,组合 `RosCalibrationIO` 和协调器,
不通过父子类业务回调执行标定。SDK 绑定接收明确的反馈新鲜度、时钟、健康订阅和发布接口。
空间入口 `core/fitting/spatial.py` 保留显式兼容导出,计算顺序在
`spatial_solver/solve.py`:输入准备 → 仅训练求解 → 可观测性/轮间统计 → 冻结结果的独立验证 → 结果装配。
`TrainingProblem` 不包含第四轮观测,训练几何中的掌部姿态也按训练轮筛选。
基座策略由 Profile 的几何约束选择;曲线、零位坐标约定、权重、阈值和原始 URDF 授权修正规则不变。
PnP 的原导入路径继续可用。`tag_pose/parameters.py` 集中定义跟踪默认值,
ROS 加载器将角度参数由 deg 转为 rad,离线采集直接使用同一套默认值。
公共分支容差仍为 `0.03 px`,重投影质量和倾角限制保持一致;首次确认、参考冻结与失联恢复
使用同一公共分支策略,ROS 和离线链路均不能绕过它。
`CalibrationSession.phase` 是流程状态来源。协调器在状态锁内开始、中止、推进和提交观测;
耗时姿态计算在锁外进行,提交时核对会话版本、采集对象、运动对象及稳态采样边界。
Start 更换预览采集对象并重新锁定基准,暂停/中止使尚未完成的旧观测失效。
收尾由 `FinalizationController` 组合 worker 和发布器,worker 只上报阶段事件、准备候选文件;
协调器核实全部验收阶段后才能提交。锁顺序固定为状态锁 → worker 锁,
拟合失败维持当前姿态,中止后禁止提交,成功产物只提交一次。
老的单相机拇指/CMC 在线节点和 launch 已退出安装入口;需要局部测量时应增加经过验证的 Profile,
不能重新启用旧状态机。老版本离线诊断不是新的标准 URDF 发布证据。
## 新增型号
必需文件:
```text
config/profiles/new_hand.yaml
config/new_hand_product.yaml
urdf/new_hand/raw.urdf
urdf/new_hand/meshes/*
```
Profile 声明:
1. Adapter、SDK 通道顺序、单位、正负方向、命令/反馈范围、基准值、独立速度槽布局。
2. Tag ID/尺寸/机位/安装 link,以及每次运动的父子观测关系;不明确的 link 不能靠名称猜。
3. 扫描任务、速度、准备/避让/收尾 waypoint;每关节首次零位任务、完整保持指令及固定到达姿态。
4. 声明空间可观测约束、已知 baseline 几何或明确接受的 CAD 零位假设;每个运动关节须独立观测或显式指定对应实测来源。
5. 每关节允许修正的 `origin.rpy``limit.lower/upper``mimic.multiplier/offset`
这仍需要正确的可观测性设计:单轴加一对未知安装角的 Tag 不能自动辨识绝对 CAD 零位。
不能为了减少配置,默认把 SDK 零点或最大指令当成 CAD 零位。可参考最简单的 L6/O6 Profile
删除不适用关系后重新声明;不要盲目复制另一型号的空间约束。
复用 SDK 不增加 Python;新协议只增加 `runtime/adapters/` 实现及协议工厂注册。
没有型号节点、runner、状态机、质量/断点/发布器;新数学只能增加通用拟合原语。
新型号使用 `artifacts.output_schema_version: 3`;产物为
`format: unified_calibration_v3, schema_version: 3`。详细双映射保留在独立标定报告中。
旧 generic v1、v4/v6/v7 只保留历史兼容工具,不把反馈表改名为指令表来伪造兼容。
v1/v2 按原语义读取,不会自动升级为独立实测;保留的 v2 数学入口用于历史回归,不能证明 v3 精度。
配置错字、错误绑定、未知修正字段在启动前失败;修改受保护文件后应审查并更新对应 SHA256。
## 产物与离线验证
会话目录保存原始样本、冻结 Tag 安装、拟合诊断、标准 URDF 验收、JSON、URDF。
有效发布由 `release_manifest.json` 及产品发布指针共同证明;单独的 `PASS` 摘要或存在 URDF 文件不等于已发布。
manifest 记录 JSON、URDF、标定报告三文件 SHA256、受保护输入和独立 holdout 证据。任一验收失败都不会替换原发布指针。
`measurement_contract.json` 提前列出 Tag/link、零位观测关系、CAD 保留/迁移与原始 mimic 限位冲突。
主 JSON 以原始 URDF 关节名为键,单值关节包含 `sdk_channel``angle_rad`
O12 等 rad 输入还包含一一对应的 `input_values`(在有效节点间线性插值,不外推)。
字节型号每条曲线固定 256 项:SDK 指令 `v` 对应所选曲线的第 `v` 项,只能在训练指令实际覆盖 0–255 时导出。
输出均是修正 URDF 的 q_output,不再加零偏。顶层记录身份、单位、完整 `baseline_command` 和共同零位约定。
独立观测的主动和被动关节直接导出各自视觉曲线,同一驱动通道可绑定多个关节。
显式复制的关节使用来源关节的同一组节点与角度,不重新拟合、归零或由 mimic 推导;接收通道保持不变。
默认单值曲线在指定的稳态训练数据中约束 baseline=0,正反向诊断分支共用冻结参考、保留回差。
独立稳态验证分别检验两个方向;超差时禁止发布,不隐藏切换分支。
O6 侧摆 `rh_thumb_cmc_yaw` 通过 `artifacts.directional_command_joints` 显式启用双向指令曲线。
v3 JSON 顶层保存同名声明,所列关节额外包含 `increasing_rad` / `decreasing_rad`
方向指原始 SDK 数值增加/减少,与关节角度正负无关。其他关节继续使用单值表。
两条曲线分别只用所声明的稳态训练数据拟合,共用已冻结零位、各自满足 baseline=0;
独立稳态验证按实采方向读取最终序列化曲线,仍要求 MAE≤1°、P95≤2°、最大≤3°。
报告分别保存运动与指令映射的训练/验证分区及来源图像,限位推导仍不得使用任一验证分区。
`angle_rad` 保留为均值参考;双向关节在非 baseline 输入且方向未知时禁止用均值代替。
读取器须从 baseline 初始化,停住时保持到达方向,允许端点换向;行程中途反转尚无小回环标定证据,会明确拒绝。
字节指令的方向判断与查表共用 `floor(v+0.5)` 量化规则;连续小数步进不能绕过方向切换检查。
两条主回环曲线不构成任意反转、负载或多轴工况的精度证明。旧读取器会拒绝新增关节字段,须同步更新。
修正 URDF 保留原被动关节 `<mimic>`,单独拖动父关节滑条时,末节按
`q_child ≈ multiplier × q_parent + offset` 联动。标准 URDF 不能将查找表表达为 mimic。
前三轮成对视觉角度用于拟合线性近似,优化变量为父关节两端对应的子关节角度;
两端都限制在子关节原有输出范围内,因此整个联动行程合法,无需放大限位或裁剪小负角。
偏移是近似直线在 baseline 的误差,不能当作新的物理零位;报告明确保存
`baseline_error_rad``physical_zero_correction: false` 和独立留出误差。
JSON 指令/反馈曲线、几何零位不因近似拟合改变。需要实测非线性运动时使用完整 JSON,
消费程序显式设置全部关节角,并避免 URDF mimic 再次覆盖末节角度。
原始 CAD 限位默认仍是硬边界。Profile 可逐关节显式声明 `calibrated_motion_envelope`
将该 CAD 限位视为本次要测定的运动范围;只有已授权的 `limit.lower/upper` 可据此修改。
O6 右手采用此策略,范围由前三轮运动反馈映射、独立稳态训练映射的实测包络及共同零位确定,
不增加安全余量,不用第四轮扩展范围,也不为容纳 mimic 越界而放大范围。
SDK 安全指令域不变,IP/DIP 的 CAD 零位不变。复制关节的范围证据仍属于来源关节。
报告和 manifest 的验收数据保存 `motion_range_evidence`,包含来源限位、零偏、训练轮次、
样本身份与新范围,并在收尾重新验证。O6 尚无独立机械安全角度界的实测证据,
该策略不代表对碰撞、过载或任意多轴姿态作出安全认证。
验收分别记录:最终 JSON 对独立稳态验证角度;完整 JSON+URDF FK 对独立验证 Tag 位姿;
仅使用 URDF mimic 的近似损失。前两项决定实测精度通过,第三项如实报告,不冒充非线性精度。
manifest 同时保存 `urdf_joint_angle_source: calibration_json``urdf_mimic_policy` 和两种回放结果。
门限保持角度 MAE≤1°、P95≤2°、最大≤3°,Tag 位置 P95≤3 mm。
平行转轴的零位计算先消除轴上参考点的任意坐标,再比较相机平面中的轴间方向;
改变同一轴线的表示点不能改变标定零位。这是所有型号共用的几何规则。
独立精度结论只覆盖实际观测的关节和 Tag;复制关节另检查来源、查表一致性、通道与机械范围,
不把来源关节的第四轮误差称为接收关节精度。混合产物 manifest 使用
`measured_and_transferred_json_urdf_v3`,列出实测和复制覆盖。
FK 使用训练冻结的安装与基座;第四轮不拟合零位、曲线、Tag 安装或基座。
正式图像采集还要求核验第四轮及独立稳态验证的原始 Tag 四角,沿用冻结的安装、基座
和受哈希保护的相机参数。主观测的独立三维位姿必须解释同帧原始角点,四角重投影 RMS
不得超过原 1.5 px;最终落盘 JSON+修正 URDF 与独立位姿的差异仍按上述角度和位置门限检查。
最终 FK 的总像素误差单独报告,不能将包含允许运动误差的总量等同于图像测量噪声。
额外机位没有独立三维测量时,仅检查是否存在满足原物理误差范围、且重投影误差不超过
1.5 px 的局部刚性位姿。这个数值见证不是真实位姿测量,不写回安装、零位、曲线或相机;
各运动任务分别检查,不能用其他任务中的静止图像稀释误差。它不能替代必要的独立三维验收。
缺少独立测量或原始图像、相机身份不一致、测量像素或物理误差超限均不能发布。
失败时保留 `final_image_diagnostics.json`;通过后在 manifest 记录 `final_file_image_holdout`
及规则版本 `independent_measurement_and_physical_consistency_v2`,分别列出独立精度与额外视角相容性。
训练阶段若 Tag 与整条运动链不满足刚性一致性,`tag_installation_diagnostics.json`
一次列出全部 Tag 的角度/位置 P95、门限和通过状态,不只显示遇到的第一个失败 Tag。
该失败不能直接归因于硬件松动,也不能作为自动重做整场采集的依据。
`calibration_report.json` 保存完整的 `command_to_rad` / `feedback_to_rad`、输入域、方向分支、
实测关节的零位原始参考帧、完整指令/反馈/到达方向、来源和训练/holdout 指标。
复制关节记录 `independently_measured: false``transferred_from_joint` 与来源参考链接,
不生成自己的伪零位样本或独立精度指标。绝对零位来源另由 `zero_method`
`cad_zero_assumption``zero_transferred_from_joint` 表达。报告中的 `applicability` 记录保持姿态;
未采集多姿态证据不能宣称任意多轴工况已验证。主 JSON 用于指令查表,不能误作反馈转换。
旧 unified v1 会话继续只读兼容,不自动覆盖或转换历史产物;新输出单表必须重新通过最终文件验收。
统一消费桥可直接读取通过验证的 manifest(纯消息转换,不会发送实机命令):
```bash
ros2 run linkerhand_calibration calibrated_joint_state_bridge --ros-args \
-p calibration_file:=<session/release_manifest.json> -p input_kind:=command \
-p input_topic:=<SDK原始指令话题> -p output_topic:=/calibration/verification/joint_states
```
需要反馈角度显示时显式选择 `input_kind:=feedback`,桥读取配套报告中的反馈映射。
桥检查三文件哈希、拒绝越域输入,
v3 全部关节角由 JSON 显式计算,不加第二次零偏。完整 FK 使用
`independent_mimic_angles=True`,以 JSON 给出的末节角度覆盖线性近似。
标准 `robot_state_publisher` 和普通 URDF 网页查看器会执行 mimic,适合检查近似联动;
仅把完整 JointState 发给标准 RSP,不能让它停止覆盖被动角。
消费实测非线性曲线的仿真程序必须明确采用 JSON 角度,关闭重复的线性被动更新。
仿真输出不能与实机反馈话题同名。
MuJoCo 是用户后续使用的外部验收项目,本轮不改动它、不增加 MuJoCo 依赖,自动标定不要求打开 GUI。
外部对比须确认加载的模型来自这份 URDF,指令转换使用配套 JSON;不能用实机反馈直接驱动仿真来证明指令映射。
现有 MuJoCo 脚本尚不能视为支持 v3:后续须消费完整关节映射并停止冲突的线性被动更新。
```bash
# 同一算法离线回放新策略的完整采集;不启动硬件,默认不发布
ros2 run linkerhand_calibration calibrate_hand --config <product.yaml> \
--offline-raw <raw_samples.jsonl> --offline-output <新的输出目录>
# 只比较两份 URDF,不修改,不控制机械手
ros2 run linkerhand_calibration compare_calibration_urdfs \
--reference <参考.urdf> --candidate <本次.urdf> --output <新报告.json>
```
离线回放也检查原始会话策略、SDK 输入域、配置/相机证据、每方向完整数据及最终文件。
旧会话不能自动获得当前哈希或新版本认证。`--publish-offline` 必须显式给出,且不会绕过验证。
当前坐标规则、生产职责与验收范围见 [v3 实施记录](CALIBRATION_V3_IMPLEMENTATION.md)。
早期需求审查见 [2026-09-10 需求审查](CALIBRATION_CODE_REVIEW_20260910.md)
当前修改依据、实机结果与未完成的精度验收见 [流程评估](CALIBRATION_PIPELINE_REVIEW.md)。
相机和反馈目前均采用下述测量时间;时间配对正确仍须经过独立空间验收。
### 实时调度边界
公共 ROS 节点顺序处理控制和反馈消息;耗时图像估计由每视角有界工作队列执行。
图像最多在原同步容差内等候真实反馈时间覆盖,队列不反向持锁查询会话。
增加执行器线程数不能提高共享状态的处理能力;反馈超时仍按真实测量判断。
暂停或验收失败的会话仅提供诊断/候选文件,不能当作已发布的标定结果。
调度对照依据与验证范围见 `CALIBRATION_PIPELINE_REVIEW.md` 第七轮评估。
### 相机与反馈时间
连续扫描将 CAN 实测反馈与相机曝光中点配对。MVS 取帧/图像发布时间不参与替代曝光时间;
设备时钟由独立锁存换算到 ROS 时钟,换算依据保存在会话的 `camera_timing_<view>.jsonl`
运行中的时钟校验失败会暂停出图并重新执行独立锁存校验,通过后恢复采集;校验失败期间不会继续使用旧映射,也不会永久退出采集线程。恢复前后的图像时间戳仍须严格递增;若主机时钟向后跳变,早于已发布图像的帧继续丢弃。
四型号共用 `unified_engine_v8_all_view_images`,保留设备曝光时间策略;旧采集只能按其原策略诊断,不能混用。
报告中的 `feedback_u8` 表示来自字节通道的测量值:插值到曝光时间后可以是小数,
拟合、内存映射和报告读取均按声明的分段线性曲线计算,不再次取整。
`command_u8` 仍按实际发送的整数字节查表;旧版整数表保留原读取规则。
相机时钟修正不会将被动关节的非线性曲线转成 URDF mimic;仿真仍须读取 JSON 查找表。
### 同一运动中的所有机位原始角点
四轮扫描、必要稳态点和关节基准采样期间,所有机位均保存 `image_observation_frame`
记录包含曝光时间、完整指令/反馈及方向、轮次与尝试编号、相机矩阵和可见 Tag 的二维角点。
它经过原检测质量与采集窗口检查,但不依赖 PnP 分支或任务的主机位。
缺少同步反馈时保留明确的空值,不能将该帧当成有效角度测量。
例如 O6 顶部 ID7 与正面 ID2 属于同一末节:弯曲和侧摆时两机位的可见角点都应保存,
以便检查整条运动链。其他机位的记录不会增加停点或运动次数,也不会替代主机位的采样覆盖要求。
同机位重复帧不重复入库;暂停、切换运动和结束采集时仍按同一会话身份撤销迟到数据。
最终 JSON/URDF 的图像验收使用通过采集检查的第四轮和独立稳态验证窗口内全部机位角点。
主观测必须能与原始图像记录逐项对应,同一图像不能重复计数;其他动作下的有效角点也直接
投影验收,不先拟合成三维位姿。缺少主观测的原始记录或二者 SDK 条件不一致时拒绝发布。
这不表示联合求解或外参自校准已经通过验证。
@@ -0,0 +1,133 @@
g20_thumb_calibration:
ros__parameters:
command_topic: /g20/cb_left_hand_control_cmd
state_topic: /g20/cb_left_hand_state
info_topic: /g20/cb_left_hand_info
camera_info_topic: /camera/camera/color/camera_info
image_topic: /camera/camera/color/image_rect
detections_topic: /apriltag/detections
tf_topic: /tf
# Use PnP translations as 3-D Tag centres and fit the directly observable
# root/MCP circles. The passive IP output follows the G20 URDF mimic
# relation below; its small residual T5 circle is diagnostic only. PnP
# orientations remain auxiliary quality checks. Command 255 is zero.
angle_estimation_mode: trajectory_center_3d
passive_ip_multiplier: 1.02
publish_debug_image: false
debug_max_rate_hz: 10.0
debug_scale: 0.5
# Default: send one end-to-end command per direction and pair every valid
# AprilTag frame with the timestamp-interpolated actual G20 state.
scan_mode: continuous
continuous_motion_mode: endpoint
repetitions: 1
# Used only by point-mode fallback and validation approach offsets.
command_step: 8
auto_start_tip: true
maximum_state_image_skew_ms: 150.0
continuous_endpoint_tolerance_u8: 2.0
continuous_endpoint_hold_seconds: 1.0
continuous_timeout_seconds: 90.0
continuous_invalid_timeout_seconds: 3.0
continuous_minimum_valid_frames: 40
continuous_minimum_state_span_u8: 240.0
continuous_minimum_bins: 32
continuous_maximum_bin_gap: 16
# Keep the responsive firmware speed, but pace it through the same
# 8-unit grid without waiting for static image captures at each point.
continuous_segment_minimum_seconds: 0.1
continuous_segment_timeout_seconds: 5.0
continuous_prepare_timeout_seconds: 30.0
preflight_frames: 150
minimum_detection_rate: 0.95
minimum_detection_hz: 15.0
maximum_hamming: 0
minimum_decision_margin: 30.0
# Trial threshold for small/far tags (historically observed at 32-38 px).
# Final acceptance is still guarded by static RMS and random validation.
minimum_edge_pixels: 30.0
# Current 30 px tags measure about 0.50-0.53 deg RMS while stationary.
# Keep a small practical margin here; final random validation stays at
# MAE <= 2 deg and P95 <= 3 deg.
# Match the preflight noise gate to the 3 deg robust capture gate below.
# The final calibration is still accepted only by the independent
# validation MAE/P95 limits, not by this readiness check.
maximum_static_std_deg: 3.0
pose_outlier_threshold_deg: 5.0
minimum_pose_inlier_rate: 0.90
pnp_minimum_valid_rate: 0.95
# 30-38 px tags are usable, but only if IPPE gives a tight image fit and
# a pose continuous with the preceding frame.
pnp_maximum_reprojection_error_px: 1.5
# Independent IPPE selection inherits the shared near-tie tolerance from
# core/geometry/pnp.py; continuity cannot override a clear image-fit lead.
pnp_maximum_pose_jump_deg: 35.0
pnp_maximum_translation_jump_m: 0.04
pnp_maximum_tag_tilt_deg: 75.0
# Preserve the branch through short detector gaps. A continuous sweep
# already pauses after 3 s without valid synchronised observations.
pnp_tracker_reset_seconds: 5.0
# Select all four IPPE branches as one kinematic chain. This prevents T4
# and T5 from independently changing mirror branches at the turnaround or
# during validation while still allowing real joint motion frame-to-frame.
pnp_group_relative_rotation_scale_deg: 5.0
pnp_group_relative_translation_scale_m: 0.01
# Reprojection remains a tie-breaker; temporal joint-chain continuity is
# deliberately dominant for the current 30-38 px planar tags.
pnp_group_reprojection_weight: 0.05
# Whole-sweep branch review. During a root sweep T3/T4/T5 should retain
# rigid relative poses; during a tip sweep T0/T3 should remain fixed.
pnp_trajectory_reprojection_scale_px: 0.1
pnp_rigid_rotation_scale_deg: 5.0
pnp_rigid_translation_scale_m: 0.01
# Judge the complete rigid trajectory against a robust sweep reference.
# Reject persistent drift at P95; keep a looser hard maximum so one noisy
# 30 px endpoint frame does not discard an otherwise sound sweep.
pnp_rigid_p95_accepted_drift_deg: 8.0
pnp_rigid_maximum_accepted_drift_deg: 15.0
# Centre-trajectory mode judges branch consistency by the Euclidean
# distance between rigid Tag centres. This is deliberately independent
# of the noisy planar-Tag orientation returned by PnP.
pnp_rigid_p95_accepted_distance_drift_m: 0.003
pnp_rigid_maximum_accepted_distance_drift_m: 0.006
# Three-dimensional centre-trajectory geometry gates. T0 stays on the
# palm as the translation anchor; T3/T4/T5 are the moving thumb points.
trajectory_maximum_plane_rms_m: 0.004
trajectory_maximum_radial_rms_m: 0.004
trajectory_minimum_radius_m: 0.005
trajectory_minimum_arc_deg: 15.0
trajectory_maximum_root_role_disagreement_deg: 5.0
trajectory_maximum_anchor_drift_m: 0.005
trajectory_static_translation_outlier_m: 0.005
trajectory_maximum_static_translation_rms_m: 0.002
# Static captures are now used only for sweep preparation and validation.
stable_frames: 5
capture_frames: 8
minimum_settle_seconds: 0.4
# This only confirms that the hand has stopped before an 8-frame robust
# median capture. The passive T4->T5 pair currently has about 2.3 deg
# peak spread over five 30 px PnP frames, while its two IPPE branches are
# separated by about 5.5 deg. A 3 deg gate accepts measurement jitter but
# still rejects a branch change. Final MAE/P95 limits remain unchanged.
maximum_stable_spread_deg: 3.0
# In centre mode the stationary capture gate is expressed in metres.
maximum_stable_translation_spread_m: 0.003
settle_timeout_seconds: 10.0
capture_timeout_seconds: 10.0
validation_command_count: 5
# The backlash approach point only waits for feedback to reach the target;
# it no longer performs an unnecessary image capture.
validation_approach_minimum_seconds: 0.2
validation_approach_timeout_seconds: 10.0
validation_position_tolerance_u8: 2.0
validation_seed: 20260727
maximum_validation_mae_deg: 2.0
maximum_validation_p95_deg: 3.0
maximum_coupling_drift_deg: 2.0
minimum_ip_coupling_r_squared: 0.98
maximum_monotonic_correction_deg: 2.0
maximum_hysteresis_deg: 5.0
@@ -0,0 +1,59 @@
g20_thumb_cmc_pitch_zero:
ros__parameters:
t0_id: 0
t3_id: 1
joint_name: thumb_cmc_pitch
motor_index: 0
zero_command_u8: 255
measure_travel: false
baseline_command_u8:
[255, 255, 255, 255, 255, 255, 193, 148, 105, 42,
245, 255, 255, 255, 255, 255, 255, 255, 255, 255]
repetitions: 3
zero_capture_frames: 30
# Each round sends one 255->64 endpoint command and one 64->255 return
# command. All valid T3-minus-T0 centres observed during both motions are
# state-binned and fitted to one image-plane circle.
trajectory_command_u8: 64
trajectory_bin_size_u8: 8.0
trajectory_minimum_frames: 45
trajectory_minimum_bins: 18
trajectory_minimum_state_span_u8: 160.0
trajectory_minimum_radius_px: 20.0
trajectory_minimum_arc_deg: 20.0
trajectory_maximum_radial_rms_px: 2.0
trajectory_maximum_p95_radial_error_px: 3.5
trajectory_endpoint_settle_seconds: 0.3
trajectory_timeout_seconds: 30.0
settle_seconds: 0.5
move_timeout_seconds: 20.0
capture_timeout_seconds: 15.0
state_tolerance_u8: 2.0
preflight_frames: 60
minimum_detection_rate: 0.95
maximum_hamming: 0
minimum_decision_margin: 30.0
minimum_edge_pixels: 40.0
maximum_static_position_rms_px: 1.5
maximum_round_difference_deg: 1.0
maximum_return_error_deg: 1.0
# Zero direction is always T3 centre -> fitted circle centre. T3's printed
# orientation and corner +x direction are deliberately not used.
maximum_zero_radial_error_px: 4.0
# Detect a long physical table/reference edge in the lower image. The red
# target and blue detected line are display-only aids for manual alignment;
# they never block preflight or the start service.
camera_alignment_enabled: true
camera_alignment_reference_y_ratio: 0.90
camera_alignment_roi_y_min_ratio: 0.55
camera_alignment_roi_y_max_ratio: 0.98
camera_alignment_minimum_line_length_ratio: 0.30
camera_alignment_max_candidate_angle_deg: 15.0
camera_alignment_max_angle_deg: 0.5
camera_alignment_max_vertical_offset_px: 12.0
camera_alignment_required_frames: 10
camera_alignment_minimum_detection_rate: 0.8
camera_alignment_max_age_seconds: 1.0
publish_debug_image: true
debug_max_rate_hz: 10.0
debug_scale: 0.75
@@ -0,0 +1,60 @@
g20_thumb_cmc_roll_calibration:
ros__parameters:
t0_id: 0
t3_id: 1
joint_name: thumb_cmc_roll
motor_index: 5
zero_command_u8: 255
measure_travel: true
baseline_command_u8:
[255, 255, 255, 255, 255, 255, 193, 148, 105, 42,
245, 255, 255, 255, 255, 255, 255, 255, 255, 255]
repetitions: 3
zero_capture_frames: 30
# Measure the complete motor-5 range. Each round captures both static
# endpoints around one 255->0->255 circle trajectory.
trajectory_command_u8: 0
trajectory_bin_size_u8: 8.0
trajectory_minimum_frames: 65
trajectory_minimum_bins: 30
trajectory_minimum_state_span_u8: 240.0
trajectory_minimum_radius_px: 20.0
trajectory_minimum_arc_deg: 20.0
trajectory_maximum_radial_rms_px: 2.0
trajectory_maximum_p95_radial_error_px: 3.5
trajectory_endpoint_settle_seconds: 0.3
trajectory_timeout_seconds: 35.0
settle_seconds: 0.5
move_timeout_seconds: 25.0
capture_timeout_seconds: 15.0
state_tolerance_u8: 2.0
preflight_frames: 60
minimum_detection_rate: 0.95
maximum_hamming: 0
minimum_decision_margin: 30.0
minimum_edge_pixels: 40.0
maximum_static_position_rms_px: 1.5
maximum_round_difference_deg: 1.0
maximum_travel_difference_deg: 1.0
minimum_travel_deg: 20.0
maximum_return_error_deg: 1.0
# Zero direction is always T3 centre -> fitted circle centre. T3's printed
# orientation and corner +x direction are deliberately not used.
maximum_zero_radial_error_px: 4.0
# Detect a long physical table/reference edge in the lower image. The red
# target and blue detected line are display-only aids for manual alignment;
# they never block preflight or the start service.
camera_alignment_enabled: true
camera_alignment_reference_y_ratio: 0.90
camera_alignment_roi_y_min_ratio: 0.55
camera_alignment_roi_y_max_ratio: 0.98
camera_alignment_minimum_line_length_ratio: 0.30
camera_alignment_max_candidate_angle_deg: 15.0
camera_alignment_max_angle_deg: 0.5
camera_alignment_max_vertical_offset_px: 12.0
camera_alignment_required_frames: 10
camera_alignment_minimum_detection_rate: 0.8
camera_alignment_max_age_seconds: 1.0
publish_debug_image: true
debug_max_rate_hz: 10.0
debug_scale: 0.75
@@ -0,0 +1,32 @@
<?xml version="1.0" encoding="UTF-8" ?>
<dds>
<profiles xmlns="http://www.eprosima.com/XMLSchemas/fastRTPS_Profiles">
<transport_descriptors>
<transport_descriptor>
<transport_id>g20_udp_transport</transport_id>
<type>UDPv4</type>
<sendBufferSize>10485760</sendBufferSize>
<receiveBufferSize>10485760</receiveBufferSize>
</transport_descriptor>
<transport_descriptor>
<transport_id>g20_shm_transport</transport_id>
<type>SHM</type>
<segment_size>67108864</segment_size>
<port_queue_capacity>512</port_queue_capacity>
<healthy_check_timeout_ms>1000</healthy_check_timeout_ms>
</transport_descriptor>
</transport_descriptors>
<participant
profile_name="g20_large_image_participant"
is_default_profile="true">
<rtps>
<userTransports>
<transport_id>g20_udp_transport</transport_id>
<transport_id>g20_shm_transport</transport_id>
</userTransports>
<useBuiltinTransports>false</useBuiltinTransports>
</rtps>
</participant>
</profiles>
</dds>
@@ -0,0 +1,30 @@
/apriltag/apriltag:
ros__parameters:
image_transport: raw
# Live calibration needs the newest frame, not lossless delivery of stale
# frames. BEST_EFFORT prevents a slow full-resolution detection callback
# from back-pressuring image_proc's reliable image publisher.
qos_profile: sensor_data
family: 36h11
size: 0.016
profile: false
max_hamming: 0
detector:
threads: 4
decimate: 1.5
blur: 0.0
refine: true
sharpening: 0.25
debug: false
pose_estimation_method: pnp
tag:
ids: [0, 1, 2, 3]
frames: [tag_t0, tag_t3, tag_t4, tag_t5]
sizes: [0.016, 0.016, 0.016, 0.016]
g20_thumb_calibration:
ros__parameters:
tag_roles: [t0, t3, t4, t5]
tag_ids: [0, 1, 2, 3]
tag_frames: [tag_t0, tag_t3, tag_t4, tag_t5]
tag_sizes_m: [0.016, 0.016, 0.016, 0.016]
@@ -0,0 +1,40 @@
schema_version: 3
profile_id: G20/right/g20_right_19/v1
profile_config: package://linkerhand_calibration/config/profiles/g20_right_19.yaml
profile_config_sha256: 958f74c476730636d2dadb58a30296a7cdf0b25ad00345252754f9c1b6b50c7f
model: G20
side: right
tag_layout: g20_right_19
serial_number: G20_RIGHT_001
can_interface: can0
output_root: calibration_output
cameras:
front:
serial_number: DB2163742
camera_name: hikrobot_front_DB2163742
camera_info: ~/.ros/camera_info/hikrobot_DB2163742.yaml
side:
serial_number: DB2163749
camera_name: hikrobot_side_DB2163749
camera_info: ~/.ros/camera_info/hikrobot_DB2163749.yaml
top:
serial_number: DB2163739
camera_name: hikrobot_top_DB2163739
camera_info: ~/.ros/camera_info/hikrobot_DB2163739.yaml
artifacts:
source_urdf: package://linkerhand_calibration/urdf/g20_right/linkerhand_g20_right.urdf
source_urdf_sha256: eeb6ffb0e95d2a6acd4c26331ae68062e0d74160de4b552b4f6d395cce5ca4e8
camera_extrinsics: config/g20_three_camera_extrinsics.yaml
camera_extrinsics_sha256: dd623572df3cb83fdefcbe92204dab54a60f2c68eb3a8c9bdb08407e8f0e5d80
calibration_config: src/g20_thumb_apriltag_calibration/config/three_camera_calibration.yaml
calibration_config_sha256: 4927506d787d665c16f0209ee4d052654e648bb25beaa1913ed0d416f449d523
tag_config: src/g20_thumb_apriltag_calibration/config/three_camera_tags_g20_right_19.yaml
tag_config_sha256: b1ab45e97ae42d57b0a3a63c725107b5aa3828c06b2ced16f8222d6e9ebadc41
release:
# Each task already contains three training cycles plus an isolated fourth
# holdout, so a second complete hardware session duplicates hours of motion.
required_independent_passes: 1
static_repeatability_deg: 1.0
@@ -0,0 +1,59 @@
/l6_calibration/front/apriltag/apriltag:
ros__parameters:
image_transport: raw
qos_profile: sensor_data
family: 36h11
size: 0.016
max_hamming: 0
detector:
threads: 4
decimate: 1.0
blur: 0.0
refine: true
sharpening: 0.25
debug: false
pose_estimation_method: pnp
tag:
ids: [0, 1, 2]
frames: [front_base, thumb_pitch, thumb_dip]
sizes: [0.016, 0.016, 0.016]
/l6_calibration/side/apriltag/apriltag:
ros__parameters:
image_transport: raw
qos_profile: sensor_data
family: 36h11
size: 0.016
max_hamming: 0
detector:
threads: 4
decimate: 1.0
blur: 0.0
refine: true
sharpening: 0.25
debug: false
pose_estimation_method: pnp
tag:
ids: [3, 4, 5]
frames: [side_base, pinky_pitch, pinky_dip]
sizes: [0.016, 0.016, 0.016]
/l6_calibration/top/apriltag/apriltag:
ros__parameters:
image_transport: raw
qos_profile: sensor_data
family: 36h11
size: 0.016
max_hamming: 0
detector:
threads: 4
decimate: 1.0
blur: 0.0
refine: true
sharpening: 0.25
debug: false
pose_estimation_method: pnp
tag:
ids: [6, 7]
frames: [top_base, thumb_roll]
sizes: [0.016, 0.016]
@@ -0,0 +1,39 @@
schema_version: 3
profile_id: L6/right/l6_right_8/v1
profile_config: package://linkerhand_calibration/config/profiles/l6_right_8.yaml
profile_config_sha256: d241e87ded3a29d416a876b27f1a251f8daf2f98e3676ab277264f1bdfaf5c60
model: L6
side: right
tag_layout: l6_right_8
namespace: /l6_calibration
serial_number: L6_RIGHT_001
can_interface: can0
output_root: calibration_output
cameras:
front:
serial_number: DB2163742
camera_name: hikrobot_front_DB2163742
camera_info: ~/.ros/camera_info/hikrobot_DB2163742.yaml
side:
serial_number: DB2163749
camera_name: hikrobot_side_DB2163749
camera_info: ~/.ros/camera_info/hikrobot_DB2163749.yaml
top:
serial_number: DB2163739
camera_name: hikrobot_top_DB2163739
camera_info: ~/.ros/camera_info/hikrobot_DB2163739.yaml
artifacts:
source_urdf: package://linkerhand_calibration/urdf/l6_right/linkerhand_l6v3.1_right.urdf
source_urdf_sha256: 298c1fbf5189648911426f530b50bdbeea4830cab9c54e20f46c532485df4666
camera_extrinsics: config/g20_three_camera_extrinsics.yaml
camera_extrinsics_sha256: dd623572df3cb83fdefcbe92204dab54a60f2c68eb3a8c9bdb08407e8f0e5d80
calibration_config: package://linkerhand_calibration/config/l6_three_camera_calibration.yaml
calibration_config_sha256: 090d82a5609e8b981c1b8c6ebeaf4c2f477323209ae3db3b9d8713ac6d6169b5
tag_config: package://linkerhand_calibration/config/l6_right_8_tags.yaml
tag_config_sha256: be1499eb947b61d2fe360ae2c92307a87710480fae8a9dd4cd171fc959fdcbf5
release:
required_independent_passes: 1
static_repeatability_deg: 1.0
@@ -0,0 +1,66 @@
l6_calibration:
ros__parameters:
command_topic: /l6/cb_right_hand_control_cmd
state_topic: /l6/cb_right_hand_state
setting_topic: /l6/cb_hand_setting_cmd
front_camera_info_topic: /l6_calibration/front/camera/camera_info
front_detections_topic: /l6_calibration/front/apriltag/detections
side_camera_info_topic: /l6_calibration/side/camera/camera_info
side_detections_topic: /l6_calibration/side/apriltag/detections
top_camera_info_topic: /l6_calibration/top/camera/camera_info
top_detections_topic: /l6_calibration/top/apriltag/detections
baseline_command_u8: [255, 255, 255, 255, 255, 255]
# L6_RIGHT_001 measured a 250->5 travel of only ~0.9 s at speed 10,
# which left fewer than 32 useful feedback bins. Speed 1 is still only a
# firmware ceiling: different L6 motors complete a full stroke in 0.7-1.3 s.
# A 100 Hz cosine trajectory therefore sets the actual, model-level pace.
preflight_speed_u8: 1
formal_speed_u8: 1
speed_settle_seconds: 0.2
command_trajectory_full_range_seconds: 6.0
torque_u8: 80
repetitions: 4
preflight_checkpoints_u8: [255, 127, 0]
tag_size_m: 0.016
tag_size_override_ids: [0, 1, 2, 3, 4, 5, 6, 7]
tag_size_overrides_m: [0.016, 0.016, 0.016, 0.016, 0.016, 0.016, 0.016, 0.016]
minimum_detection_rate: 0.95
# Per-Tag quality remains >=95%. With three independently detected Tags,
# the fully joined frame rate may be 0.95^3 ~= 85.7%.
minimum_joint_frame_rate: 0.85
minimum_feedback_hz: 25.0
maximum_state_image_skew_ms: 50.0
maximum_hamming: 0
minimum_decision_margin: 30.0
minimum_edge_pixels: 30.0
pnp_maximum_reprojection_error_px: 1.5
pnp_maximum_pose_jump_deg: 35.0
pnp_maximum_translation_jump_m: 0.04
pnp_maximum_tag_tilt_deg: 75.0
pnp_tracker_reset_seconds: 5.0
minimum_sweep_frames: 40
minimum_state_span_u8: 240.0
minimum_sweep_bins: 32
maximum_bin_gap: 16
maximum_monotonic_correction_deg: 2.0
passive_maximum_monotonic_correction_deg: 3.0
maximum_validation_mae_deg: 1.0
maximum_validation_p95_deg: 2.0
maximum_validation_error_deg: 3.0
mimic_minimum_multiplier: 0.5
mimic_maximum_multiplier: 1.5
mimic_maximum_cycle_range: 0.03
mimic_maximum_residual_p95_deg: 2.0
endpoint_tolerance_u8: 2.0
endpoint_hold_seconds: 1.0
motor_stall_timeout_seconds: 2.0
position_timeout_seconds: 30.0
sweep_timeout_seconds: 90.0
automatic_sweep_retry_limit: 1
non_target_motion_tolerance_u8: 3.0
fixed_base_maximum_corner_drift_px: 5.0
fixed_base_movement_confirmation_frames: 10
@@ -0,0 +1,41 @@
/o12_calibration/front/apriltag/apriltag:
ros__parameters:
image_transport: raw
qos_profile: sensor_data
family: 36h11
size: 0.016
max_hamming: 0
detector: {threads: 4, decimate: 1.0, blur: 0.0, refine: true, sharpening: 0.25, debug: false}
pose_estimation_method: pnp
tag:
ids: [0, 1, 2, 3, 12, 13]
frames: [front_base, thumb_cmc, thumb_mcp, thumb_dip, middle_roll, index_roll]
sizes: [0.016, 0.016, 0.016, 0.016, 0.016, 0.016]
/o12_calibration/side/apriltag/apriltag:
ros__parameters:
image_transport: raw
qos_profile: sensor_data
family: 36h11
size: 0.016
max_hamming: 0
detector: {threads: 4, decimate: 1.0, blur: 0.0, refine: true, sharpening: 0.25, debug: false}
pose_estimation_method: pnp
tag:
ids: [4, 5, 6, 7, 8, 9, 10, 11]
frames: [side_base, pinky_mcp, pinky_pip, pinky_dip, middle_pip, middle_dip, index_pip, index_dip]
sizes: [0.016, 0.016, 0.016, 0.016, 0.016, 0.016, 0.016, 0.016]
/o12_calibration/top/apriltag/apriltag:
ros__parameters:
image_transport: raw
qos_profile: sensor_data
family: 36h11
size: 0.016
max_hamming: 0
detector: {threads: 4, decimate: 1.0, blur: 0.0, refine: true, sharpening: 0.25, debug: false}
pose_estimation_method: pnp
tag:
ids: [14, 15]
frames: [top_base, thumb_yaw]
sizes: [0.016, 0.016]
@@ -0,0 +1,47 @@
schema_version: 3
profile_id: O12/right/o12_right_16/v1
profile_config: package://linkerhand_calibration/config/profiles/o12_right_16.yaml
profile_config_sha256: 652c2a82dfc7532b4f62c42cebdad57b169435cf950c5f2ee910cc67c7972621
model: O12
side: right
tag_layout: o12_right_16
namespace: /o12_calibration
serial_number: O12_RIGHT_001
output_root: calibration_output
sdk:
driver: o12_sdk_bridge
python_package: src/agillink_omnihand_sdk/linux/x64/python/omnihand-1.1.8-cp312-cp312-linux_x86_64.whl
package_sha256: cae7a0d5bce7e7c9d72cc90a0dd15e152f11e2a1ecf04cb6761e8171399ff170
transport: hcan
setup: src/agillink_omnihand_sdk/linux/x64/ros2/jazzy/setup.bash
config: package://linkerhand_calibration/config/o12_sdk.yaml
config_sha256: 1e3c0942b32128943fbba27846a69a8483af6da9551d99ed687e45c4321ade9c
cameras:
front:
serial_number: DB2163742
camera_name: hikrobot_front_DB2163742
camera_info: ~/.ros/camera_info/hikrobot_DB2163742.yaml
side:
serial_number: DB2163749
camera_name: hikrobot_side_DB2163749
camera_info: ~/.ros/camera_info/hikrobot_DB2163749.yaml
top:
serial_number: DB2163739
camera_name: hikrobot_top_DB2163739
camera_info: ~/.ros/camera_info/hikrobot_DB2163739.yaml
artifacts:
source_urdf: package://linkerhand_calibration/urdf/o12_right/linkerhand_o12_t3_right-0703.urdf
source_urdf_sha256: 75b3c18992a3640d2d7d36719da5ff75428ded477b97d9cf29adbe43a6eb27eb
camera_extrinsics: config/o12_three_camera_extrinsics.yaml
camera_extrinsics_sha256: 5a515d0706f4e67d5e26bfddb348b519817bd72e885ea9e43997e41016176e53
calibration_config: package://linkerhand_calibration/config/o12_three_camera_calibration.yaml
calibration_config_sha256: c2fcdd532e22c15013413e655311de2ca87107c23a96d5c043e8f340f232eebd
tag_config: package://linkerhand_calibration/config/o12_right_16_tags.yaml
tag_config_sha256: 41001c3afba74cc01eb524a75dc58561a37e9364fab029ea12d879156a008dab
release:
required_independent_passes: 1
static_repeatability_deg: 1.0
@@ -0,0 +1,45 @@
# OmniHand Pro 2025 (O12) Configuration
# connection_type: "zlgcan" | "hcan" | "socketcan" | "rs485" | "usb"
# request_interval_ms: min gap between requests, SDK default 0 = no throttling, range 0..100
# frame_recv_timeout_ms: single-frame reply wait, SDK default 50, range 10..1000
/**:
ros__parameters:
# This workstation uses one O12 right hand through HCAN device 0/channel 0.
# Leave left_hand undefined so the node does not try to open a second hand.
right_hand:
hand_device_id: 1
connection_type: "hcan"
canfd_device_id: 0
canfd_channel_id: 0
# O12 POSITION streams set-points and has no G20-style firmware speed
# interpolation, so do not throttle the 50 Hz calibration command stream.
request_interval_ms: 0
frame_recv_timeout_ms: 100
show_data_details: false
# Alternative ZLG CANFD example:
# right_hand:
# hand_device_id: 1
# connection_type: "zlgcan"
# canfd_device_id: 0
# canfd_channel_id: 0
# request_interval_ms: 0
# frame_recv_timeout_ms: 50
# show_data_details: false
# socketcan example
# left_hand:
# hand_device_id: 1
# connection_type: "socketcan"
# can_interface: "can0"
# request_interval_ms: 0
# frame_recv_timeout_ms: 50
# show_data_details: true
# right_hand:
# hand_device_id: 1
# connection_type: "socketcan"
# can_interface: "can1"
# request_interval_ms: 0
# frame_recv_timeout_ms: 50
# show_data_details: true
@@ -0,0 +1,53 @@
o12_calibration:
ros__parameters:
command_topic: /o12/right/joint_cmd
state_topic: /o12/right/joint_states
front_camera_info_topic: /o12_calibration/front/camera/camera_info
front_detections_topic: /o12_calibration/front/apriltag/detections
side_camera_info_topic: /o12_calibration/side/camera/camera_info
side_detections_topic: /o12_calibration/side/apriltag/detections
top_camera_info_topic: /o12_calibration/top/camera/camera_info
top_detections_topic: /o12_calibration/top/apriltag/detections
# O12 standard interface is radians. Scan the complete reviewed SDK range;
# the source CAD limits are outputs to correct, not acquisition limits.
# Raw 0..2000 mixed control is forbidden.
command_rate_hz: 50.0
repetitions: 4
tag_size_m: 0.016
tag_size_override_ids: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]
tag_size_overrides_m: [0.016, 0.016, 0.016, 0.016, 0.016, 0.016, 0.016, 0.016, 0.016, 0.016, 0.016, 0.016, 0.016, 0.016, 0.016, 0.016]
minimum_detection_rate: 0.95
minimum_joint_frame_rate: 0.85
minimum_feedback_hz: 35.0
maximum_state_image_skew_ms: 50.0
minimum_sweep_frames: 40
minimum_state_span_fraction: 0.90
minimum_sweep_bins: 32
maximum_bin_gap: 2
endpoint_tolerance_rad: 0.01
endpoint_hold_seconds: 0.5
motor_stall_timeout_seconds: 2.0
position_timeout_seconds: 60.0
sweep_timeout_seconds: 180.0
automatic_sweep_retry_limit: 1
non_target_motion_tolerance_rad: 0.015
maximum_temperature_c: 70
# 当前 O12 固件返回空温度报告。仍发起查询;5秒无结果后依赖已验证的
# joint_error_states bit1 过热保护,并在会话记录中明确标注降级。
temperature_report_required: false
temperature_fallback_after_seconds: 5.0
# 快速标定档:各任务按4倍请求,但拇指pitch、侧摆、屈曲及避让均有独立硬上限。
motion_speed_scale: 4.0
# 正弦速度加减速时间;中段保持关节限速,避免位置余弦全程低速。
trajectory_ramp_seconds: 0.4
maximum_hamming: 0
minimum_decision_margin: 30.0
minimum_edge_pixels: 30.0
fixed_base_maximum_corner_drift_px: 5.0
fixed_base_movement_confirmation_frames: 10
pnp_maximum_reprojection_error_px: 1.5
pnp_maximum_pose_jump_deg: 35.0
pnp_maximum_translation_jump_m: 0.04
pnp_maximum_tag_tilt_deg: 75.0
pnp_tracker_reset_seconds: 5.0
@@ -0,0 +1,59 @@
/o6_calibration/front/apriltag/apriltag:
ros__parameters:
image_transport: raw
qos_profile: sensor_data
family: 36h11
size: 0.0165
max_hamming: 0
detector:
threads: 4
decimate: 1.0
blur: 0.0
refine: true
sharpening: 0.25
debug: false
pose_estimation_method: pnp
tag:
ids: [0, 1, 2]
frames: [front_base, thumb_pitch, thumb_ip]
sizes: [0.0165, 0.0165, 0.0165]
/o6_calibration/side/apriltag/apriltag:
ros__parameters:
image_transport: raw
qos_profile: sensor_data
family: 36h11
size: 0.0165
max_hamming: 0
detector:
threads: 4
decimate: 1.0
blur: 0.0
refine: true
sharpening: 0.25
debug: false
pose_estimation_method: pnp
tag:
ids: [3, 4, 5]
frames: [side_base, pinky_pitch, pinky_dip]
sizes: [0.0165, 0.0165, 0.0165]
/o6_calibration/top/apriltag/apriltag:
ros__parameters:
image_transport: raw
qos_profile: sensor_data
family: 36h11
size: 0.0165
max_hamming: 0
detector:
threads: 4
decimate: 1.0
blur: 0.0
refine: true
sharpening: 0.25
debug: false
pose_estimation_method: pnp
tag:
ids: [6, 7]
frames: [top_base, thumb_yaw]
sizes: [0.0165, 0.0165]
@@ -0,0 +1,39 @@
schema_version: 3
profile_id: O6/right/o6_right_8/v1
profile_config: package://linkerhand_calibration/config/profiles/o6_right_8.yaml
profile_config_sha256: 72bf56b13eb2a0d4d14a5be42838b2ffc176722f470fe1a3ae5fd84604e1df4d
model: O6
side: right
tag_layout: o6_right_8
namespace: /o6_calibration
serial_number: O6_RIGHT_001
can_interface: can0
output_root: calibration_output
cameras:
front:
serial_number: DB2163742
camera_name: hikrobot_front_DB2163742
camera_info: config/o6_camera_intrinsics_20260915/hikrobot_DB2163742.yaml
side:
serial_number: DB2163749
camera_name: hikrobot_side_DB2163749
camera_info: config/o6_camera_intrinsics_20260915/hikrobot_DB2163749.yaml
top:
serial_number: DB2163739
camera_name: hikrobot_top_DB2163739
camera_info: config/o6_camera_intrinsics_20260915/hikrobot_DB2163739.yaml
artifacts:
source_urdf: package://linkerhand_calibration/urdf/o6_right/linkerhand_o6_right.urdf
source_urdf_sha256: 8f184faad699fbf771e388f109a4e8793b5cb190c33a87b2eba8491a3a37dd62
camera_extrinsics: config/o6_three_camera_extrinsics_20260915_192517.yaml
camera_extrinsics_sha256: d057183eaba592149ec4c63c4ceab8a70e1a0c9d7cf2460a1c8b7e86a6ca6638
calibration_config: package://linkerhand_calibration/config/o6_three_camera_calibration.yaml
calibration_config_sha256: 49ce8a0d317695994c5b306e3a3a28d0470b8769c77f1173ca935178430587db
tag_config: package://linkerhand_calibration/config/o6_right_8_tags.yaml
tag_config_sha256: 219b8aa906fc9e2a13f3bbede46f542ff176ef2c7f60190f0dd2508993f04593
release:
required_independent_passes: 1
static_repeatability_deg: 1.0
@@ -0,0 +1,63 @@
o6_calibration:
ros__parameters:
command_topic: /o6/cb_right_hand_control_cmd
state_topic: /o6/cb_right_hand_state
setting_topic: /o6/cb_hand_setting_cmd
front_camera_info_topic: /o6_calibration/front/camera/camera_info
front_detections_topic: /o6_calibration/front/apriltag/detections
side_camera_info_topic: /o6_calibration/side/camera/camera_info
side_detections_topic: /o6_calibration/side/apriltag/detections
top_camera_info_topic: /o6_calibration/top/camera/camera_info
top_detections_topic: /o6_calibration/top/apriltag/detections
baseline_command_u8: [255, 255, 255, 255, 255, 255]
# O6 has a different speed scale from L6. Motion is still bounded by the
# six-second cosine command trajectory; these values are firmware limits.
baseline_speed_u8: 80
preflight_speed_u8: 60
formal_speed_u8: 40
speed_settle_seconds: 0.2
command_trajectory_full_range_seconds: 6.0
torque_u8: 80
repetitions: 4
preflight_checkpoints_u8: [255, 127, 0]
tag_size_m: 0.0165
tag_size_override_ids: [0, 1, 2, 3, 4, 5, 6, 7]
tag_size_overrides_m: [0.0165, 0.0165, 0.0165, 0.0165, 0.0165, 0.0165, 0.0165, 0.0165]
minimum_detection_rate: 0.95
minimum_joint_frame_rate: 0.85
minimum_feedback_hz: 25.0
maximum_state_image_skew_ms: 50.0
maximum_hamming: 0
minimum_decision_margin: 30.0
minimum_edge_pixels: 30.0
pnp_maximum_reprojection_error_px: 1.5
pnp_maximum_pose_jump_deg: 35.0
pnp_maximum_translation_jump_m: 0.04
pnp_maximum_tag_tilt_deg: 75.0
pnp_tracker_reset_seconds: 5.0
minimum_sweep_frames: 40
minimum_state_span_u8: 240.0
minimum_sweep_bins: 32
maximum_bin_gap: 16
maximum_monotonic_correction_deg: 2.0
passive_maximum_monotonic_correction_deg: 3.0
maximum_validation_mae_deg: 1.0
maximum_validation_p95_deg: 2.0
maximum_validation_error_deg: 3.0
mimic_minimum_multiplier: 0.5
mimic_maximum_multiplier: 2.2
mimic_maximum_cycle_range: 0.03
mimic_maximum_residual_p95_deg: 2.0
endpoint_tolerance_u8: 2.0
endpoint_hold_seconds: 1.0
motor_stall_timeout_seconds: 2.0
position_timeout_seconds: 30.0
sweep_timeout_seconds: 90.0
automatic_sweep_retry_limit: 1
non_target_motion_tolerance_u8: 3.0
fixed_base_maximum_corner_drift_px: 5.0
fixed_base_movement_confirmation_frames: 10
File diff suppressed because it is too large Load Diff
@@ -0,0 +1,418 @@
schema_version: 1
profile_id: L6/right/l6_right_8/v1
namespace: /l6_calibration
sdk_adapter: legacy_byte_sdk
command:
sdk_to_joint_direction: [-1, -1, -1, -1, -1, -1]
names:
- thumb_cmc_pitch
- thumb_cmc_roll
- index_mcp_pitch
- middle_mcp_pitch
- ring_mcp_pitch
- pinky_mcp_pitch
baseline_u8:
- 255
- 255
- 255
- 255
- 255
- 255
command_index_by_joint:
rh_thumb_cmc_pitch: 0
rh_thumb_cmc_roll: 1
rh_index_mcp_pitch: 2
rh_middle_mcp_pitch: 3
rh_ring_mcp_pitch: 4
rh_pinky_mcp_pitch: 5
disabled_indices: []
urdf_joint_by_joint:
rh_thumb_cmc_pitch: rh_thumb_cmc_pitch
rh_thumb_cmc_roll: rh_thumb_cmc_roll
rh_index_mcp_pitch: rh_index_mcp_pitch
rh_middle_mcp_pitch: rh_middle_mcp_pitch
rh_ring_mcp_pitch: rh_ring_mcp_pitch
rh_pinky_mcp_pitch: rh_pinky_mcp_pitch
feedback_name_aliases:
thumb_cmc_yaw: thumb_cmc_roll
speed_slot_by_command_index:
'0': 0
'1': 1
'2': 2
'3': 3
'4': 4
'5': 5
unit: u8
baseline: []
lower_bounds: []
upper_bounds: []
feedback_lower_bounds: []
feedback_upper_bounds: []
feedback_by_index: false
vision:
views:
- name: front
tags:
- role: front_base
fixed_reference: true
id: 0
- role: thumb_pitch
fixed_reference: false
id: 1
- role: thumb_dip
fixed_reference: false
id: 2
- name: side
tags:
- role: side_base
fixed_reference: true
id: 3
- role: pinky_pitch
fixed_reference: false
id: 4
- role: pinky_dip
fixed_reference: false
id: 5
- name: top
tags:
- role: top_base
fixed_reference: true
id: 6
- role: thumb_roll
fixed_reference: false
id: 7
common_frame: calibration_common
extrinsic_reference_view: front
extrinsics_quality_limits:
reprojection_rms_px: 1.2
maximum_rotation_repeatability_deg: 0.3
maximum_translation_repeatability_m: 0.0015
minimum_capture_counts:
front_side_captures: 15
front_top_captures: 15
motion:
tasks:
- key: thumb_roll_top
view: top
command_index: 1
joints:
- rh_thumb_cmc_roll
auxiliary_commands:
- - 0
- 255
validation_only: false
start_u8: 255
end_u8: 0
preflight_speed_u8: 1
formal_speed_u8: 1
start: null
end: null
preflight_speed: null
formal_speed: null
- key: thumb_pitch_dip_front
view: front
command_index: 0
joints:
- rh_thumb_cmc_pitch
- rh_thumb_dip
auxiliary_commands:
- - 1
- 255
validation_only: false
start_u8: 255
end_u8: 0
preflight_speed_u8: 1
formal_speed_u8: 1
start: null
end: null
preflight_speed: null
formal_speed: null
- key: pinky_pitch_dip_side
view: side
command_index: 5
joints:
- rh_pinky_mcp_pitch
- rh_pinky_dip
auxiliary_commands: []
validation_only: false
start_u8: 255
end_u8: 0
preflight_speed_u8: 1
formal_speed_u8: 1
start: null
end: null
preflight_speed: null
formal_speed: null
preparation_waypoints_u8: []
safe_return_waypoints_u8: []
speed_parameters:
preflight_u8: 1
formal_u8: 1
speed_settle_seconds: 0.2
command_trajectory_full_range_seconds: 6.0
torque_u8: 80
endpoint_hold_seconds: 1.0
stall_timeout_seconds: 2.0
precheck_sweeps: false
steady_command_checkpoints: false
joint_zero_references:
rh_thumb_cmc_roll:
task_key: thumb_roll_top
command: [255, 255, 255, 255, 255, 255]
approach_commands:
- [255, 0, 255, 255, 255, 255]
rh_thumb_cmc_pitch:
task_key: thumb_pitch_dip_front
command: [255, 255, 255, 255, 255, 255]
approach_commands:
- [0, 255, 255, 255, 255, 255]
rh_thumb_dip:
task_key: thumb_pitch_dip_front
command: [255, 255, 255, 255, 255, 255]
approach_commands:
- [0, 255, 255, 255, 255, 255]
rh_pinky_mcp_pitch:
task_key: pinky_pitch_dip_side
command: [255, 255, 255, 255, 255, 255]
approach_commands:
- [255, 255, 255, 255, 255, 0]
rh_pinky_dip:
task_key: pinky_pitch_dip_side
command: [255, 255, 255, 255, 255, 255]
approach_commands:
- [255, 255, 255, 255, 255, 0]
measurement:
transferred_motion_sources:
rh_index_mcp_pitch: rh_pinky_mcp_pitch
rh_index_dip: rh_pinky_dip
rh_middle_mcp_pitch: rh_pinky_mcp_pitch
rh_middle_dip: rh_pinky_dip
rh_ring_mcp_pitch: rh_pinky_mcp_pitch
rh_ring_dip: rh_pinky_dip
measurements:
rh_thumb_cmc_roll:
joint: rh_thumb_cmc_roll
kind: relative_rotation
view: top
parent_role: top_base
child_role: thumb_roll
validation_source: null
pose_axis_line_required: true
rh_thumb_cmc_pitch:
joint: rh_thumb_cmc_pitch
kind: relative_rotation
view: front
parent_role: front_base
child_role: thumb_pitch
validation_source: null
pose_axis_line_required: true
rh_thumb_dip:
joint: rh_thumb_dip
kind: relative_rotation
view: front
parent_role: thumb_pitch
child_role: thumb_dip
validation_source: null
pose_axis_line_required: true
rh_pinky_mcp_pitch:
joint: rh_pinky_mcp_pitch
kind: relative_rotation
view: side
parent_role: side_base
child_role: pinky_pitch
validation_source: null
pose_axis_line_required: true
rh_pinky_dip:
joint: rh_pinky_dip
kind: relative_rotation
view: side
parent_role: pinky_pitch
child_role: pinky_dip
validation_source: null
pose_axis_line_required: true
cross_view_sources: {}
image_curve_joints: []
directional_zero: true
cross_view_roll_curve: false
stable_cross_view_cone_bias: false
zero:
cad_zero_assumptions:
rh_index_dip: User accepts original CAD joint zero as the physical baseline; not
independently measured.
rh_middle_dip: User accepts original CAD joint zero as the physical baseline; not
independently measured.
rh_pinky_dip: User accepts original CAD joint zero as the physical baseline; not
independently measured.
rh_ring_dip: User accepts original CAD joint zero as the physical baseline; not
independently measured.
rh_thumb_dip: User accepts original CAD joint zero as the physical baseline; not
independently measured.
active_joints:
- rh_index_mcp_pitch
- rh_middle_mcp_pitch
- rh_pinky_mcp_pitch
- rh_ring_mcp_pitch
- rh_thumb_cmc_pitch
- rh_thumb_cmc_roll
passive_joints:
- rh_index_dip
- rh_middle_dip
- rh_pinky_dip
- rh_ring_dip
- rh_thumb_dip
direct_zero_joints:
- rh_pinky_mcp_pitch
- rh_thumb_cmc_pitch
- rh_thumb_cmc_roll
axis_joints:
- rh_pinky_dip
- rh_pinky_mcp_pitch
- rh_thumb_cmc_pitch
- rh_thumb_cmc_roll
- rh_thumb_dip
mechanical_endpoint_joints: []
post_solve_endpoint_joints: []
mimic_source_by_joint:
rh_thumb_dip: rh_thumb_cmc_pitch
rh_index_dip: rh_index_mcp_pitch
rh_middle_dip: rh_middle_mcp_pitch
rh_ring_dip: rh_ring_mcp_pitch
rh_pinky_dip: rh_pinky_mcp_pitch
cad_frozen_joints:
- rh_index_dip
- rh_middle_dip
- rh_pinky_dip
- rh_ring_dip
- rh_thumb_dip
endpoint_anchor_by_joint: {}
fitted_mimic_joints:
- rh_pinky_dip
- rh_thumb_dip
coupling_model_by_joint:
rh_thumb_dip: linear_mimic
rh_pinky_dip: linear_mimic
rh_index_dip: linear_mimic
rh_middle_dip: linear_mimic
rh_ring_dip: linear_mimic
transferred_zero_sources: {rh_index_mcp_pitch: rh_pinky_mcp_pitch, rh_middle_mcp_pitch: rh_pinky_mcp_pitch, rh_ring_mcp_pitch: rh_pinky_mcp_pitch}
transferred_mimic_sources: {rh_index_dip: rh_pinky_dip, rh_middle_dip: rh_pinky_dip, rh_ring_dip: rh_pinky_dip}
spatial:
base_pose_strategy: thumb_serial
root_anchor_joints: [rh_thumb_cmc_roll]
orientation_anchor_joint: rh_pinky_mcp_pitch
directed_base_axis_joints: [rh_thumb_cmc_roll, rh_pinky_mcp_pitch]
depth_free_axis_projection: true
axis_order: [rh_thumb_cmc_roll, rh_thumb_cmc_pitch, rh_thumb_dip, rh_pinky_mcp_pitch, rh_pinky_dip]
axis_parent_joint: {rh_thumb_cmc_pitch: rh_thumb_cmc_roll}
phase_parent_joint: {rh_thumb_dip: rh_thumb_cmc_pitch, rh_pinky_dip: rh_pinky_mcp_pitch}
offset_observer_joint: {rh_thumb_cmc_roll: rh_thumb_cmc_pitch, rh_thumb_cmc_pitch: rh_thumb_dip, rh_pinky_mcp_pitch: rh_pinky_dip}
quality:
training_cycles:
- 0
- 1
- 2
holdout_cycle: 3
hard_threshold_keys:
- maximum_mimic_residual_rad
- maximum_state_image_skew_ms
- maximum_validation_error_rad
- minimum_detection_rate
retry_metric_scope: {}
isolated_holdout: true
scope:
calibrate_joints:
partial:
- rh_pinky_mcp_pitch
- rh_thumb_cmc_pitch
- rh_thumb_cmc_roll
frozen_joints:
partial:
- rh_index_mcp_pitch
- rh_middle_mcp_pitch
- rh_ring_mcp_pitch
default_scope: partial
artifacts:
output_schema_version: 3
calibration_filename: l6_right_{serial_number}_partial_calibration.json
corrected_urdf_filename: linkerhand_l6_right_{serial_number}_partial_zero_calibrated.urdf
protected_input_fields:
- calibration_config_sha256
- camera_extrinsics_sha256
- profile_config_sha256
- source_urdf_sha256
- tag_config_sha256
publication_pointer: latest_partial_passed
session_compatibility_tokens:
- feedback_curves_v6
- l6_partial_v1
publish_corrected_urdf: true
acquisition:
command_capture_mode: separate
joint_zero_timeout_seconds: 5.0
policy_version: unified_engine_v8_all_view_images
mapping_probe_maximum_rad: 0.0
automatic_rescan_limit: 1
minimum_valid_samples: 40
minimum_bins: 32
maximum_unobserved_fraction: 0.0625
legacy_minimum_span_01: 0.9411764705882353
physical_first_cycle_minimum_span_01: 0.85
physical_repeat_minimum_fraction: 0.9
stall_timeout_seconds: 2.0
feedback_stale_seconds: 1.0
fixed_reference_minimum_frames: 10
fixed_reference_maximum_drift_px: 5.0
fixed_reference_confirmation_frames: 10
urdf:
authorized_fields:
rh_pinky_mcp_pitch:
- limit.lower
- limit.upper
- origin.rpy
rh_thumb_cmc_roll:
- limit.lower
- limit.upper
- origin.rpy
rh_index_mcp_pitch:
- limit.lower
- limit.upper
- origin.rpy
rh_thumb_cmc_pitch:
- limit.lower
- limit.upper
- origin.rpy
rh_middle_mcp_pitch:
- limit.lower
- limit.upper
- origin.rpy
rh_ring_mcp_pitch:
- limit.lower
- limit.upper
- origin.rpy
rh_ring_dip:
- mimic.multiplier
- mimic.offset
rh_thumb_dip:
- mimic.multiplier
- mimic.offset
rh_index_dip:
- mimic.multiplier
- mimic.offset
rh_pinky_dip:
- mimic.multiplier
- mimic.offset
rh_middle_dip:
- mimic.multiplier
- mimic.offset
joint_coverage:
rh_pinky_mcp_pitch: measured_static_dynamic
rh_thumb_cmc_roll: measured_static_dynamic
rh_thumb_cmc_pitch: measured_static_dynamic
rh_middle_mcp_pitch: transferred_static_dynamic
rh_index_mcp_pitch: transferred_static_dynamic
rh_ring_mcp_pitch: transferred_static_dynamic
rh_pinky_dip: measured_dynamic_cad_static
rh_middle_dip: transferred_dynamic_cad_static
rh_ring_dip: transferred_dynamic_cad_static
rh_thumb_dip: measured_dynamic_cad_static
rh_index_dip: transferred_dynamic_cad_static
File diff suppressed because it is too large Load Diff
@@ -0,0 +1,472 @@
schema_version: 1
profile_id: O6/right/o6_right_8/v1
namespace: /o6_calibration
sdk_adapter: legacy_byte_sdk
command:
sdk_to_joint_direction: [-1, -1, -1, -1, -1, -1]
names:
- thumb_cmc_pitch
- thumb_cmc_yaw
- index_mcp_pitch
- middle_mcp_pitch
- ring_mcp_pitch
- pinky_mcp_pitch
baseline_u8:
- 255
- 255
- 255
- 255
- 255
- 255
command_index_by_joint:
rh_thumb_cmc_pitch: 0
rh_thumb_cmc_yaw: 1
rh_index_mcp_pitch: 2
rh_middle_mcp_pitch: 3
rh_ring_mcp_pitch: 4
rh_pinky_mcp_pitch: 5
disabled_indices: []
urdf_joint_by_joint:
rh_thumb_cmc_pitch: rh_thumb_cmc_pitch
rh_thumb_cmc_yaw: rh_thumb_cmc_yaw
rh_index_mcp_pitch: rh_index_mcp_pitch
rh_middle_mcp_pitch: rh_middle_mcp_pitch
rh_ring_mcp_pitch: rh_ring_mcp_pitch
rh_pinky_mcp_pitch: rh_pinky_mcp_pitch
feedback_name_aliases: {}
speed_slot_by_command_index:
'0': 0
'1': 1
'2': 2
'3': 3
'4': 4
'5': 5
unit: u8
baseline: []
lower_bounds: []
upper_bounds: []
feedback_lower_bounds: []
feedback_upper_bounds: []
feedback_by_index: false
vision:
views:
- name: front
tags:
- role: front_base
fixed_reference: true
id: 0
size_m: 0.0165
- role: thumb_pitch
fixed_reference: false
id: 1
size_m: 0.0165
- role: thumb_ip
fixed_reference: false
id: 2
size_m: 0.0165
- name: side
tags:
- role: side_base
fixed_reference: true
id: 3
size_m: 0.0165
- role: pinky_pitch
fixed_reference: false
id: 4
size_m: 0.0165
- role: pinky_dip
fixed_reference: false
id: 5
size_m: 0.0165
- name: top
tags:
- role: top_base
fixed_reference: true
id: 6
size_m: 0.0165
- role: thumb_yaw
fixed_reference: false
id: 7
size_m: 0.0165
link: rh_thumb_distal
common_frame: calibration_common
extrinsic_reference_view: front
extrinsics_quality_limits:
reprojection_rms_px: 1.5
maximum_rotation_repeatability_deg: 0.3
maximum_translation_repeatability_m: 0.0015
minimum_capture_counts:
front_side_captures: 15
front_top_captures: 15
motion:
tasks:
- key: thumb_yaw_top
view: top
command_index: 1
joints:
- rh_thumb_cmc_yaw
auxiliary_commands: []
validation_only: false
start_u8: 255
end_u8: 0
preflight_speed_u8: 60
formal_speed_u8: 40
start: null
end: null
preflight_speed: null
formal_speed: null
- key: thumb_pitch_ip_front
view: front
command_index: 0
joints:
- rh_thumb_cmc_pitch
- rh_thumb_ip
auxiliary_commands: []
validation_only: false
start_u8: 255
end_u8: 0
preflight_speed_u8: 60
formal_speed_u8: 40
start: null
end: null
preflight_speed: null
formal_speed: null
- key: pinky_pitch_dip_side
view: side
command_index: 5
joints:
- rh_pinky_mcp_pitch
- rh_pinky_dip
auxiliary_commands: []
validation_only: false
start_u8: 255
end_u8: 0
preflight_speed_u8: 60
formal_speed_u8: 40
start: null
end: null
preflight_speed: null
formal_speed: null
preparation_waypoints_u8: []
safe_return_waypoints_u8: []
speed_parameters:
baseline_u8: 80
preflight_u8: 60
formal_u8: 40
speed_settle_seconds: 0.2
command_trajectory_full_range_seconds: 6.0
torque_u8: 80
endpoint_hold_seconds: 1.0
stall_timeout_seconds: 2.0
precheck_sweeps: false
steady_command_checkpoints: false
joint_zero_references:
rh_thumb_cmc_yaw:
task_key: thumb_yaw_top
command: [255, 255, 255, 255, 255, 255]
approach_commands:
- [255, 0, 255, 255, 255, 255]
rh_thumb_cmc_pitch:
task_key: thumb_pitch_ip_front
command: [255, 255, 255, 255, 255, 255]
approach_commands:
- [0, 255, 255, 255, 255, 255]
rh_thumb_ip:
task_key: thumb_pitch_ip_front
command: [255, 255, 255, 255, 255, 255]
approach_commands:
- [0, 255, 255, 255, 255, 255]
rh_pinky_mcp_pitch:
task_key: pinky_pitch_dip_side
command: [255, 255, 255, 255, 255, 255]
approach_commands:
- [255, 255, 255, 255, 255, 0]
rh_pinky_dip:
task_key: pinky_pitch_dip_side
command: [255, 255, 255, 255, 255, 255]
approach_commands:
- [255, 255, 255, 255, 255, 0]
measurement:
transferred_motion_sources:
rh_index_mcp_pitch: rh_pinky_mcp_pitch
rh_index_dip: rh_pinky_dip
rh_middle_mcp_pitch: rh_pinky_mcp_pitch
rh_middle_dip: rh_pinky_dip
rh_ring_mcp_pitch: rh_pinky_mcp_pitch
rh_ring_dip: rh_pinky_dip
measurements:
rh_thumb_cmc_yaw:
joint: rh_thumb_cmc_yaw
kind: relative_rotation
view: top
parent_role: top_base
child_role: thumb_yaw
validation_source: null
pose_axis_line_required: true
rh_thumb_cmc_pitch:
joint: rh_thumb_cmc_pitch
kind: relative_rotation
view: front
parent_role: front_base
child_role: thumb_pitch
validation_source: null
pose_axis_line_required: true
rh_thumb_ip:
joint: rh_thumb_ip
kind: relative_rotation
view: front
parent_role: thumb_pitch
child_role: thumb_ip
validation_source: null
pose_axis_line_required: true
rh_pinky_mcp_pitch:
joint: rh_pinky_mcp_pitch
kind: relative_rotation
view: side
parent_role: side_base
child_role: pinky_pitch
validation_source: null
pose_axis_line_required: true
rh_pinky_dip:
joint: rh_pinky_dip
kind: relative_rotation
view: side
parent_role: pinky_pitch
child_role: pinky_dip
validation_source: null
pose_axis_line_required: true
cross_view_sources: {}
image_curve_joints: []
directional_zero: true
cross_view_roll_curve: false
stable_cross_view_cone_bias: false
zero:
cad_zero_assumptions:
rh_index_dip: User accepts original CAD joint zero as the physical baseline; not
independently measured.
rh_middle_dip: User accepts original CAD joint zero as the physical baseline; not
independently measured.
rh_pinky_dip: User accepts original CAD joint zero as the physical baseline; not
independently measured.
rh_ring_dip: User accepts original CAD joint zero as the physical baseline; not
independently measured.
rh_thumb_ip: User accepts original CAD joint zero as the physical baseline; not
independently measured.
active_joints:
- rh_index_mcp_pitch
- rh_middle_mcp_pitch
- rh_pinky_mcp_pitch
- rh_ring_mcp_pitch
- rh_thumb_cmc_pitch
- rh_thumb_cmc_yaw
passive_joints:
- rh_index_dip
- rh_middle_dip
- rh_pinky_dip
- rh_ring_dip
- rh_thumb_ip
direct_zero_joints:
- rh_pinky_mcp_pitch
- rh_thumb_cmc_pitch
- rh_thumb_cmc_yaw
axis_joints:
- rh_pinky_dip
- rh_pinky_mcp_pitch
- rh_thumb_cmc_pitch
- rh_thumb_cmc_yaw
- rh_thumb_ip
mechanical_endpoint_joints: []
post_solve_endpoint_joints: []
mimic_source_by_joint:
rh_thumb_ip: rh_thumb_cmc_pitch
rh_index_dip: rh_index_mcp_pitch
rh_middle_dip: rh_middle_mcp_pitch
rh_ring_dip: rh_ring_mcp_pitch
rh_pinky_dip: rh_pinky_mcp_pitch
cad_frozen_joints:
- rh_index_dip
- rh_middle_dip
- rh_pinky_dip
- rh_ring_dip
- rh_thumb_ip
endpoint_anchor_by_joint: {}
fitted_mimic_joints:
- rh_pinky_dip
- rh_thumb_ip
coupling_model_by_joint:
rh_thumb_ip: linear_mimic
rh_index_dip: linear_mimic
rh_middle_dip: linear_mimic
rh_ring_dip: linear_mimic
rh_pinky_dip: linear_mimic
transferred_zero_sources: {rh_index_mcp_pitch: rh_pinky_mcp_pitch, rh_middle_mcp_pitch: rh_pinky_mcp_pitch, rh_ring_mcp_pitch: rh_pinky_mcp_pitch}
transferred_mimic_sources: {rh_index_dip: rh_pinky_dip, rh_middle_dip: rh_pinky_dip, rh_ring_dip: rh_pinky_dip}
spatial:
base_pose_strategy: thumb_serial
root_anchor_joints: [rh_thumb_cmc_yaw]
orientation_anchor_joint: rh_pinky_mcp_pitch
directed_base_axis_joints: [rh_thumb_cmc_yaw, rh_pinky_mcp_pitch]
depth_free_axis_projection: true
axis_order: [rh_thumb_cmc_yaw, rh_thumb_cmc_pitch, rh_thumb_ip, rh_pinky_mcp_pitch, rh_pinky_dip]
axis_parent_joint: {rh_thumb_cmc_pitch: rh_thumb_cmc_yaw}
phase_parent_joint: {rh_thumb_ip: rh_thumb_cmc_pitch, rh_pinky_dip: rh_pinky_mcp_pitch}
offset_observer_joint: {rh_thumb_cmc_yaw: rh_thumb_cmc_pitch, rh_thumb_cmc_pitch: rh_thumb_ip, rh_pinky_mcp_pitch: rh_pinky_dip}
quality:
training_cycles:
- 0
- 1
- 2
holdout_cycle: 3
hard_threshold_keys:
- maximum_mimic_residual_rad
- maximum_state_image_skew_ms
- maximum_validation_error_rad
- minimum_detection_rate
retry_metric_scope: {}
isolated_holdout: true
scope:
calibrate_joints:
partial:
- rh_pinky_mcp_pitch
- rh_thumb_cmc_pitch
- rh_thumb_cmc_yaw
frozen_joints:
partial:
- rh_index_mcp_pitch
- rh_middle_mcp_pitch
- rh_ring_mcp_pitch
default_scope: partial
artifacts:
output_schema_version: 3
directional_command_joints:
- rh_thumb_cmc_yaw
calibration_filename: o6_right_{serial_number}_partial_calibration.json
corrected_urdf_filename: linkerhand_o6_right_{serial_number}_partial_zero_calibrated.urdf
protected_input_fields:
- calibration_config_sha256
- camera_extrinsics_sha256
- profile_config_sha256
- source_urdf_sha256
- tag_config_sha256
publication_pointer: latest_partial_passed
session_compatibility_tokens:
- feedback_curves_v6
- o6_partial_v1
publish_corrected_urdf: true
acquisition:
joint_zero_timeout_seconds: 5.0
command_capture_mode: separate
steady_training_nodes: 9
steady_extra_training_nodes:
pinky_pitch_dip_side: [239.0]
policy_version: unified_engine_v8_all_view_images
mapping_probe_maximum_rad: 0.0
automatic_rescan_limit: 1
minimum_valid_samples: 40
minimum_bins: 32
maximum_unobserved_fraction: 0.0625
legacy_minimum_span_01: 0.9411764705882353
physical_first_cycle_minimum_span_01: 0.85
physical_repeat_minimum_fraction: 0.9
stall_timeout_seconds: 2.0
feedback_stale_seconds: 1.0
fixed_reference_minimum_frames: 10
fixed_reference_maximum_drift_px: 5.0
fixed_reference_confirmation_frames: 10
urdf:
limit_policies:
rh_pinky_mcp_pitch:
source_limits: calibrated_motion_envelope
evidence_source: User accepts original CAD zeros but requests measured SDK motion amplitudes; source CAD limits are nominal motion envelopes.
rh_thumb_cmc_pitch:
source_limits: calibrated_motion_envelope
evidence_source: User accepts original CAD zeros but requests measured SDK motion amplitudes; source CAD limits are nominal motion envelopes.
rh_middle_mcp_pitch:
source_limits: calibrated_motion_envelope
evidence_source: User accepts original CAD zeros but requests measured SDK motion amplitudes; source CAD limits are nominal motion envelopes.
rh_index_mcp_pitch:
source_limits: calibrated_motion_envelope
evidence_source: User accepts original CAD zeros but requests measured SDK motion amplitudes; source CAD limits are nominal motion envelopes.
rh_ring_mcp_pitch:
source_limits: calibrated_motion_envelope
evidence_source: User accepts original CAD zeros but requests measured SDK motion amplitudes; source CAD limits are nominal motion envelopes.
rh_thumb_cmc_yaw:
source_limits: calibrated_motion_envelope
evidence_source: User accepts original CAD zeros but requests measured SDK motion amplitudes; source CAD limits are nominal motion envelopes.
rh_pinky_dip:
source_limits: calibrated_motion_envelope
evidence_source: User accepts original CAD zeros but requests measured SDK motion amplitudes; source CAD limits are nominal motion envelopes.
rh_thumb_ip:
source_limits: calibrated_motion_envelope
evidence_source: User accepts original CAD zeros but requests measured SDK motion amplitudes; source CAD limits are nominal motion envelopes.
rh_middle_dip:
source_limits: calibrated_motion_envelope
evidence_source: User accepts original CAD zeros but requests measured SDK motion amplitudes; source CAD limits are nominal motion envelopes.
rh_ring_dip:
source_limits: calibrated_motion_envelope
evidence_source: User accepts original CAD zeros but requests measured SDK motion amplitudes; source CAD limits are nominal motion envelopes.
rh_index_dip:
source_limits: calibrated_motion_envelope
evidence_source: User accepts original CAD zeros but requests measured SDK motion amplitudes; source CAD limits are nominal motion envelopes.
authorized_fields:
rh_pinky_mcp_pitch:
- limit.lower
- limit.upper
- origin.rpy
rh_index_mcp_pitch:
- limit.lower
- limit.upper
- origin.rpy
rh_thumb_cmc_pitch:
- limit.lower
- limit.upper
- origin.rpy
rh_middle_mcp_pitch:
- limit.lower
- limit.upper
- origin.rpy
rh_ring_mcp_pitch:
- limit.lower
- limit.upper
- origin.rpy
rh_thumb_cmc_yaw:
- limit.lower
- limit.upper
- origin.rpy
rh_ring_dip:
- mimic.multiplier
- mimic.offset
- limit.lower
- limit.upper
rh_index_dip:
- mimic.multiplier
- mimic.offset
- limit.lower
- limit.upper
rh_pinky_dip:
- mimic.multiplier
- mimic.offset
- limit.lower
- limit.upper
rh_thumb_ip:
- mimic.multiplier
- mimic.offset
- limit.lower
- limit.upper
rh_middle_dip:
- mimic.multiplier
- mimic.offset
- limit.lower
- limit.upper
joint_coverage:
rh_pinky_mcp_pitch: measured_static_dynamic
rh_thumb_cmc_pitch: measured_static_dynamic
rh_middle_mcp_pitch: transferred_static_dynamic
rh_index_mcp_pitch: transferred_static_dynamic
rh_ring_mcp_pitch: transferred_static_dynamic
rh_thumb_cmc_yaw: measured_static_dynamic
rh_pinky_dip: measured_dynamic_cad_static
rh_thumb_ip: measured_dynamic_cad_static
rh_middle_dip: transferred_dynamic_cad_static
rh_ring_dip: transferred_dynamic_cad_static
rh_index_dip: transferred_dynamic_cad_static
@@ -0,0 +1,44 @@
$schema: https://json-schema.org/draft/2020-12/schema
title: LinkerHand calibration profile
type: object
additionalProperties: false
required:
- schema_version
- profile_id
- namespace
- sdk_adapter
- command
- vision
- motion
- measurement
- zero
- quality
- scope
- artifacts
- urdf
properties:
schema_version: {const: 1}
profile_id: {type: string, pattern: "^[A-Za-z0-9_-]+/(left|right)/[a-z0-9_-]+/v[1-9][0-9]*$"}
namespace: {type: string, pattern: "^/[^/].*[^/]$"}
sdk_adapter: {type: string, minLength: 1}
command: {type: object}
vision: {type: object}
motion: {type: object}
measurement: {type: object}
zero: {type: object}
quality: {type: object}
scope: {type: object}
artifacts: {type: object}
acquisition: {type: object}
urdf:
type: object
required: [authorized_fields]
properties:
authorized_fields:
type: object
additionalProperties:
type: array
uniqueItems: true
items:
enum: [origin.rpy, limit.lower, limit.upper, mimic.multiplier, mimic.offset]
joint_coverage: {type: object}
@@ -0,0 +1,14 @@
$schema: https://json-schema.org/draft/2020-12/schema
title: LinkerHand calibration product
type: object
required: [schema_version, profile_id, serial_number, cameras, artifacts]
properties:
schema_version: {type: integer, minimum: 2}
profile_id: {type: string}
profile_config: {type: string}
serial_number: {type: string, pattern: "^[A-Za-z0-9_.-]+$"}
namespace: {type: string}
sdk: {type: object}
cameras: {type: object, minProperties: 1}
artifacts: {type: object}
release: {type: object}
@@ -0,0 +1,191 @@
g20_calibration:
ros__parameters:
command_topic: /g20/cb_left_hand_control_cmd
state_topic: /g20/cb_left_hand_state
info_topic: /g20/cb_left_hand_info
setting_topic: /g20/cb_hand_setting_cmd
front_camera_info_topic: /g20_calibration/front/camera/camera_info
front_detections_topic: /g20_calibration/front/apriltag/detections
side_camera_info_topic: /g20_calibration/side/camera/camera_info
side_detections_topic: /g20_calibration/side/apriltag/detections
top_camera_info_topic: /g20_calibration/top/camera/camera_info
top_detections_topic: /g20_calibration/top/apriltag/detections
# /start先下发并确认这个20通道基准姿态,稳定后才进入第一条扫描。
baseline_command_u8: [255, 255, 255, 255, 255, 255, 127, 127, 127, 127, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255]
normal_calibration_speed: 15
index_roll_calibration_speed: 5
index_flex_calibration_speed: 10
# 19-Tag产品预检仍使用上面保守速度;只有正反预检都留出至少双倍正式分箱余量,
# 才把非roll任务正式扫描最多提速1.5倍。四指roll受0.5°回差门限约束,
# 始终保持速度5;任一方向采样余量不足也保持原速度。
adaptive_formal_speed_enabled: true
adaptive_formal_speed_max_scale: 1.5
adaptive_formal_speed_minimum_bins: 64
adaptive_formal_speed_maximum_bin_gap: 8
speed_setting_settle_seconds: 0.25
# tag36h11尺寸是检测角点围成的黑色正方形边长,不包含外围白边。
# 19张Tag的黑色码区外边长均为16 mm。自定义PnP必须与
# apriltag_ros逐ID尺寸一致,禁止用纸张/白边尺寸代替码区尺寸。
tag_size_m: 0.016
# ROS 2无法从YAML空数组推断整数/浮点数组类型。这四个
# 末端Tag仍显式写16 mm,防止节点启动时得到未初始化参数。
tag_size_override_ids: [7, 14, 16, 18]
tag_size_overrides_m: [0.016, 0.016, 0.016, 0.016]
repetitions: 3
# 19-Tag产品正式零位使用前三轮训练、最后一轮完全留出;旧11-Tag仍读取repetitions=3。
g20_right_19_repetitions: 4
preflight_frames: 60
minimum_detection_rate: 0.95
minimum_detection_hz: 15.0
minimum_feedback_hz: 25.0
maximum_hamming: 0
minimum_decision_margin: 30.0
minimum_edge_pixels: 30.0
pnp_maximum_reprojection_error_px: 1.5
pnp_maximum_pose_jump_deg: 35.0
pnp_maximum_translation_jump_m: 0.04
pnp_maximum_tag_tilt_deg: 75.0
pnp_tracker_reset_seconds: 5.0
# 标定任务不再用第一帧决定平面Tag的IPPE分支;静止端点联合8帧选择整组最稳定解。
pnp_group_initialization_frames: 8
# 侧面当前任务所需Tag在初始化端点的贴面法向应一致;用此先验消除静态IPPE镜像双解。
pnp_group_normal_alignment_scale_deg: 5.0
pnp_group_maximum_normal_alignment_deg: 15.0
# 三个拇指顶部任务共用预检时冻结的Tag 8位姿。Tag 8仍须实时可见;
# 任一角点相对会话基准漂移超过2 px并连续5帧时,判定标定中基准被移动。
fixed_base_maximum_corner_drift_px: 5.0
fixed_base_movement_confirmation_frames: 10
# 仅在拇指MCP/IP同步运动且至少一个候选落入可信区间时,用源URDF mimic
# 关系辅助选择IPPE分支;若全部候选超限则退回纯视觉,绝不丢帧,也不生成、
# 缩放或替代被动IP的自身Tag实测曲线。
thumb_ip_pnp_coupling_multiplier: 1.03
thumb_ip_pnp_coupling_scale_deg: 3.0
thumb_ip_pnp_maximum_coupling_residual_deg: 7.5
top_pnp_invalid_reset_seconds: 1.0
# 三维位姿必须与实测20通道状态严格按时间戳配对。
maximum_state_image_skew_ms: 50.0
axis_maximum_plane_rms_m: 0.003
# 被动耦合轴只用轨迹确定轴线位置,允许更大的轴向深度噪声;径向和跨轮
# 轴线一致性仍沿用严格检查。
passive_axis_maximum_plane_rms_m: 0.004
axis_maximum_radial_rms_m: 0.003
# 整段相对SE(3)运动拟合轴线点;端视关节会投影掉单目PnP光轴深度。
axis_maximum_pose_line_rms_m: 0.001
# 仅用于运动平面在三维中可观测的斜视关节;近图像平面关节使用姿态轴
# 约束三维圆,不让单目平面Tag的深度噪声自由决定转轴方向。
axis_maximum_rotation_circle_difference_deg: 1.0
# 单轴模型残差与跨轮重复误差分开判定:主动刚性关节要求更严;被动耦合
# 关节允许可重复的非理想单轴分量,但仍须通过0.75°跨轮轴差及最终轮留出。
active_maximum_rotation_orthogonal_rms_deg: 2.5
passive_maximum_rotation_orthogonal_rms_deg: 7.5
zero_maximum_axis_cycle_difference_deg: 0.75
# 零位无法改变父子轴夹角;超过该值属于CAD/PnP几何错误,不能吸收到零位。
zero_maximum_axis_cone_mismatch_deg: 5.0
zero_maximum_observability_condition_number: 10000000000.0
zero_maximum_offset_deg: 20.0
# 四指MCP roll保留严格的装配保护范围。thumb CMC三轴由多轴视觉几何
# 求解且不假定电气端点等于CAD上限;thumb_mcp及四指MCP pitch/PIP
# 静态零位由实测全行程与CAD机械端点联合求解,不写死为0。
zero_finger_maximum_offset_deg: 3.0
# 只对实物已确认等同CAD端点的关节使用该限制;CMC电气端点不作此假设。
mechanical_endpoint_maximum_offset_deg: 5.0
endpoint_tolerance_u8: 2.0
# 请求命令与固件反馈是两个标定域。稳态检查点允许小幅死区,但反馈
# 必须已经稳定;大残差仍由机械卡滞保护处理。
steady_checkpoint_command_feedback_tolerance_u8: 8.0
steady_checkpoint_maximum_feedback_range_u8: 2.0
# 电机10在命令0时实测会稳定反馈为4;该0端使用±4。
thumb_yaw_zero_endpoint_tolerance_u8: 4.0
# 右手电机10在命令255时多次实测稳定反馈为250;仅右手该端点使用±5。
right_thumb_yaw_255_endpoint_tolerance_u8: 5.0
# 右手小指PIP电机19在命令0时固件反馈稳定饱和为5;仅其0端使用±5。
pinky_pip_zero_endpoint_tolerance_u8: 5.0
endpoint_hold_seconds: 0.5
# roll零位127必须从两个方向到位并静止采集,禁止用运动中经过127的帧判回差。
baseline_hold_seconds: 0.5
minimum_baseline_hold_frames: 10
# unified_engine_v2 不执行每任务全行程预检;保留参数仅兼容旧配置读取。
task_precheck_hold_seconds: 2.0
position_timeout_seconds: 30.0
sweep_timeout_seconds: 90.0
# 启动宽限1秒后,反馈连续2秒没有至少1个u8的进展,按机械卡滞立即暂停;
# 这类硬故障不自动重试。
motor_stall_timeout_seconds: 2.0
motor_stall_startup_grace_seconds: 1.0
motor_stall_minimum_progress_u8: 1.0
invalid_timeout_seconds: 3.0
minimum_sweep_frames: 40
minimum_state_span_u8: 240.0
minimum_sweep_bins: 32
maximum_bin_gap: 16
# 可恢复的采样失败自动重扫当前方向;超过次数才暂停等待人工处理。
automatic_sweep_retry_limit: 1
# 轨迹拟合失败优先只重扫失败轮次;零位/URDF模型失败不重复运动。
automatic_fit_retry_limit: 0
automatic_motion_retry_limit: 0
# 留空为正式标定;设为pinky/ring/middle/index时只采该指正面+侧面roll
# 即使正面baseline回差失败也继续完成侧面对照,并永久锁定本会话URDF发布。
cross_view_roll_diagnostic_finger: ""
# 过程检查允许25%的黄色预警带,最终验收仍使用下面的严格门限。
provisional_warning_ratio: 1.25
retry_minimum_speed: 3
retry_speed_scales: [1.0]
retry_endpoint_hold_seconds: [0.5]
trajectory_maximum_plane_rms_m: 0.004
trajectory_maximum_radial_rms_m: 0.004
trajectory_minimum_radius_m: 0.003
trajectory_minimum_arc_deg: 15.0
# 以下二维参数只供旧轨迹工具兼容,三机位v4零位不使用二维投影。
image_trajectory_maximum_radial_rms_px: 2.0
image_trajectory_maximum_radial_p95_px: 3.5
image_trajectory_minimum_radius_px: 20.0
trajectory_maximum_cycle_travel_difference_deg: 3.0
passive_maximum_cycle_travel_difference_deg: 10.0
maximum_monotonic_correction_deg: 2.0
# 旧布局仍用连续扫描正反程差门限;19-Tag产品的连续运动包含速度相关滞后,
# 由方向曲线和最终留出验证建模,不再重复硬判。其绝对正反程门禁使用下面
# 的九点稳态command_maximum_direction_gap_deg。
maximum_hysteresis_deg: 2.0
# 19-Tag产品模式额外要求每轮正反方向在各自baseline处绕实测关节轴的角度差
# 不超过0.5°;四指roll例外:127以255→127为唯一物理零位,反向分支
# 保留实测偏差,并改为检查分支间隙上限及跨轮稳定性。
baseline_maximum_hysteresis_deg: 0.5
directional_zero_maximum_branch_gap_deg: 2.0
directional_zero_maximum_branch_gap_range_deg: 0.3
cross_view_roll_maximum_branch_gap_difference_deg: 0.3
# 正面roll是Tag中心的二维投影角,侧面roll是三维姿态角。允许一个有界的
# 固定比例吸收Tag安装倾角/偏置带来的投影缩放,再严格比较两条曲线形状;
# 比例过大、方向相反、形状RMS及两视角各自的四轮重复性仍会失败。
cross_view_roll_maximum_shape_rms_deg: 1.25
cross_view_roll_maximum_projection_scale_ratio: 1.5
passive_maximum_monotonic_correction_deg: 3.0
passive_maximum_hysteresis_deg: 2.0
# 九点姿态先按实测反馈重新对齐,再用此门限检查固有机械回差;请求命令域
# 的两条运行曲线仍原样保留固件方向死区,不能把command/feedback差算成回差。
command_maximum_direction_gap_deg: 2.0
# 默认无额外随机动作;19-Tag产品最终一轮始终作为不可关闭的留出验证。
validation_enabled: false
# 第四轮留出求解后必须再走8个固定安全组合姿态;三机位规定Tag全部可见
# 且实测20通道到位才允许发布。只保存Tag位姿,不保存原始图像。
# Developer diagnostic only. The formal fourth sweep cycle already gives
# every isolated PIP/DIP pair an independent holdout.
combination_validation_enabled: false
combination_validation_frames: 10
combination_maximum_position_p95_m: 0.003
combination_maximum_orientation_p95_deg: 2.0
validation_command_count: 3
validation_frames: 10
validation_seed: 20260804
validation_timeout_seconds: 20.0
maximum_validation_mae_deg: 1.0
maximum_validation_p95_deg: 2.0
# 19-Tag产品模式使用更严格的任一点及静态零偏95%置信区间门限。
maximum_validation_error_deg: 3.0
zero_maximum_confidence_half_width_deg: 1.5
@@ -0,0 +1,62 @@
/g20_calibration/front/apriltag/apriltag:
ros__parameters:
image_transport: raw
qos_profile: sensor_data
family: 36h11
size: 0.016
profile: false
max_hamming: 0
detector:
threads: 4
decimate: 1.5
blur: 0.0
refine: true
sharpening: 0.25
debug: false
pose_estimation_method: pnp
tag:
ids: [0, 1, 2, 3, 10]
frames: [front_base, thumb_cmc, thumb_mcp, thumb_ip, index_roll]
sizes: [0.016, 0.016, 0.016, 0.016, 0.016]
/g20_calibration/side/apriltag/apriltag:
ros__parameters:
image_transport: raw
qos_profile: sensor_data
family: 36h11
size: 0.016
profile: false
max_hamming: 0
detector:
threads: 4
decimate: 1.5
blur: 0.0
refine: true
sharpening: 0.25
debug: false
pose_estimation_method: pnp
tag:
ids: [4, 5, 6, 7]
frames: [side_base, index_mcp, index_pip, index_dip]
sizes: [0.016, 0.016, 0.016, 0.016]
/g20_calibration/top/apriltag/apriltag:
ros__parameters:
image_transport: raw
qos_profile: sensor_data
family: 36h11
size: 0.016
profile: false
max_hamming: 0
detector:
threads: 4
decimate: 1.5
blur: 0.0
refine: true
sharpening: 0.25
debug: false
pose_estimation_method: pnp
tag:
ids: [8, 9]
frames: [top_base, thumb_yaw]
sizes: [0.016, 0.016]
@@ -0,0 +1,62 @@
/g20_calibration/front/apriltag/apriltag:
ros__parameters:
image_transport: raw
qos_profile: sensor_data
family: 36h11
size: 0.016
profile: false
max_hamming: 0
detector:
threads: 4
decimate: 1.5
blur: 0.0
refine: true
sharpening: 0.25
debug: false
pose_estimation_method: pnp
tag:
ids: [0, 1, 2, 3, 10, 11, 12, 13]
frames: [front_base, thumb_cmc, thumb_mcp, thumb_ip, pinky_roll, ring_roll, middle_roll, index_roll]
sizes: [0.016, 0.016, 0.016, 0.016, 0.016, 0.016, 0.016, 0.016]
/g20_calibration/side/apriltag/apriltag:
ros__parameters:
image_transport: raw
qos_profile: sensor_data
family: 36h11
size: 0.016
profile: false
max_hamming: 0
detector:
threads: 4
decimate: 1.5
blur: 0.0
refine: true
sharpening: 0.25
debug: false
pose_estimation_method: pnp
tag:
ids: [4, 5, 6, 15, 17]
frames: [side_base, ring_pip, pinky_pip, middle_pip, index_pip]
sizes: [0.016, 0.016, 0.016, 0.016, 0.016]
/g20_calibration/top/apriltag/apriltag:
ros__parameters:
image_transport: raw
qos_profile: sensor_data
family: 36h11
size: 0.016
profile: false
max_hamming: 0
detector:
threads: 4
decimate: 1.5
blur: 0.0
refine: true
sharpening: 0.25
debug: false
pose_estimation_method: pnp
tag:
ids: [8, 9]
frames: [top_base, thumb_yaw]
sizes: [0.016, 0.016]
@@ -0,0 +1,63 @@
/g20_calibration/front/apriltag/apriltag:
ros__parameters:
image_transport: raw
qos_profile: sensor_data
family: 36h11
size: 0.016
profile: false
max_hamming: 0
detector:
threads: 4
decimate: 1.5
blur: 0.0
refine: true
sharpening: 0.25
debug: false
pose_estimation_method: pnp
tag:
ids: [0, 1, 2, 3, 10, 11, 12, 13]
frames: [front_base, thumb_cmc, thumb_mcp, thumb_ip, pinky_roll, ring_roll, middle_roll, index_roll]
sizes: [0.016, 0.016, 0.016, 0.016, 0.016, 0.016, 0.016, 0.016]
/g20_calibration/side/apriltag/apriltag:
ros__parameters:
image_transport: raw
qos_profile: sensor_data
family: 36h11
size: 0.016
profile: false
max_hamming: 0
detector:
threads: 4
# Distal Tags use the same measured 16 mm black-code edge as all others.
decimate: 1.0
blur: 0.0
refine: true
sharpening: 0.25
debug: false
pose_estimation_method: pnp
tag:
ids: [4, 5, 6, 7, 14, 15, 16, 17, 18]
frames: [side_base, ring_pip, pinky_pip, pinky_dip, ring_dip, middle_pip, middle_dip, index_pip, index_dip]
sizes: [0.016, 0.016, 0.016, 0.016, 0.016, 0.016, 0.016, 0.016, 0.016]
/g20_calibration/top/apriltag/apriltag:
ros__parameters:
image_transport: raw
qos_profile: sensor_data
family: 36h11
size: 0.016
profile: false
max_hamming: 0
detector:
threads: 4
decimate: 1.5
blur: 0.0
refine: true
sharpening: 0.25
debug: false
pose_estimation_method: pnp
tag:
ids: [8, 9]
frames: [top_base, thumb_yaw]
sizes: [0.016, 0.016]
@@ -0,0 +1,21 @@
"""One-release compatibility surface for the former Python package name.
New code must import :mod:`linkerhand_calibration`. Only the documented
configuration loader is re-exported here; calibration algorithms continue to
have a single implementation in the renamed package.
"""
from __future__ import annotations
import warnings
warnings.warn(
"g20_thumb_apriltag_calibration is deprecated; "
"import linkerhand_calibration instead",
DeprecationWarning,
stacklevel=2,
)
from linkerhand_calibration.product import ProductConfig, load_product_config
__all__ = ["ProductConfig", "load_product_config"]
@@ -0,0 +1,9 @@
"""Deprecated forwarding entry point for the runtime joint-state bridge."""
from linkerhand_calibration.calibrated_joint_state_bridge import main
__all__ = ["main"]
if __name__ == "__main__":
main()
@@ -0,0 +1,9 @@
"""Deprecated forwarding entry point for offline replay."""
from linkerhand_calibration.offline_replay import main
__all__ = ["main"]
if __name__ == "__main__":
main()
@@ -0,0 +1,9 @@
"""Deprecated forwarding entry point for the former Python package."""
from linkerhand_calibration.one_command import main
__all__ = ["main"]
if __name__ == "__main__":
main()
@@ -0,0 +1,37 @@
"""Publish profile-calibrated URDF angles from raw command/feedback u8 values."""
from launch import LaunchDescription
from launch.actions import DeclareLaunchArgument
from launch.substitutions import LaunchConfiguration
from launch_ros.actions import Node
def generate_launch_description() -> LaunchDescription:
return LaunchDescription(
[
DeclareLaunchArgument("hand_type", default_value="right"),
DeclareLaunchArgument("calibration_file"),
DeclareLaunchArgument("input_topic", default_value=""),
DeclareLaunchArgument("output_topic", default_value=""),
Node(
package="linkerhand_calibration",
executable="calibrated_joint_state_bridge",
name=[
"calibrated_joint_state_bridge_",
LaunchConfiguration("hand_type"),
],
output="screen",
emulate_tty=True,
parameters=[
{
"hand_type": LaunchConfiguration("hand_type"),
"calibration_file": LaunchConfiguration(
"calibration_file"
),
"input_topic": LaunchConfiguration("input_topic"),
"output_topic": LaunchConfiguration("output_topic"),
}
],
),
]
)
@@ -0,0 +1,462 @@
"""Launch Profile-declared Hikrobot views and one calibration owner."""
from __future__ import annotations
from datetime import datetime
import hashlib
from pathlib import Path
import re
from ament_index_python.packages import get_package_share_directory
from launch import LaunchDescription
from launch.actions import (
DeclareLaunchArgument,
ExecuteProcess,
LogInfo,
OpaqueFunction,
SetEnvironmentVariable,
)
from launch.conditions import IfCondition
from launch.substitutions import LaunchConfiguration
from launch_ros.actions import ComposableNodeContainer, Node
from launch_ros.descriptions import ComposableNode
from launch_ros.parameter_descriptions import ParameterValue
def _launch_stack(context):
from linkerhand_calibration.product import (
ProductCalibrationContract,
)
from linkerhand_calibration.profiles import load_hand_profile
from linkerhand_calibration.runtime.adapters.ros_topics import sdk_topics
from linkerhand_calibration.runtime.diagnostic_capture import resolve_diagnostic_capture
from linkerhand_calibration.runtime.observation_scope import required_observation_views
model = LaunchConfiguration("model").perform(context).strip().upper()
hand_type = LaunchConfiguration("hand_type").perform(context).lower()
if hand_type not in {"left", "right"}:
raise RuntimeError("hand_type must be left or right")
tag_layout = LaunchConfiguration("tag_layout").perform(context).lower()
try:
profile_path = LaunchConfiguration("profile_config").perform(context).strip()
if not profile_path:
raise ValueError("online calibration requires a protected YAML Profile; use calibrate_hand --config")
expected = LaunchConfiguration("profile_config_expected_sha256").perform(context).strip()
if not expected or hashlib.sha256(Path(profile_path).read_bytes()).hexdigest() != expected:
raise ValueError("Profile changed between product validation and launch")
contract = ProductCalibrationContract(declarative=load_hand_profile(profile_path))
key = contract.typed_profile.key
if (key.model, key.side, key.layout) != (model, hand_type, tag_layout):
raise ValueError("launch identity differs from the protected Profile")
except ValueError as error:
raise RuntimeError(str(error)) from error
diagnostic = resolve_diagnostic_capture(
contract.typed_profile,
LaunchConfiguration("diagnostic_capture").perform(context).strip(),
)
views = required_observation_views(contract.typed_profile, diagnostic)
requested_tag_config = LaunchConfiguration("tag_config").perform(context)
if not requested_tag_config:
raise RuntimeError("tag_config is required; use calibrate_hand --config")
tag_config = Path(requested_tag_config).expanduser().resolve()
if not tag_config.is_file():
raise RuntimeError(f"tag config does not exist: {tag_config}")
topic_prefix = f"/{model.lower()}"
uses_hcan = contract.typed_profile.sdk_adapter == "o12_hcan_sdk"
if contract.typed_profile.sdk_adapter not in {"legacy_byte_sdk", "o12_hcan_sdk"}:
raise RuntimeError("SDK adapter has no ROS launch binding")
topics = sdk_topics(contract.typed_profile)
command_topic, state_topic = topics.command, topics.feedback
requested_source = LaunchConfiguration("source_urdf_path").perform(context)
if not requested_source:
raise RuntimeError("source_urdf_path is required; use calibrate_hand --config")
source_urdf = Path(requested_source).expanduser().resolve()
if not source_urdf.is_file():
raise RuntimeError(f"source URDF does not exist: {source_urdf}")
expected_source_hash = LaunchConfiguration(
"source_urdf_expected_sha256"
).perform(context).strip().lower()
if contract.typed_profile.artifacts.publish_corrected_urdf:
if re.fullmatch(r"[0-9a-f]{64}", expected_source_hash) is None:
raise RuntimeError(
"this profile requires source_urdf_expected_sha256 confirmed "
"by the CAD/hardware owner"
)
actual_source_hash = hashlib.sha256(source_urdf.read_bytes()).hexdigest()
if actual_source_hash != expected_source_hash:
raise RuntimeError(
"source_urdf_expected_sha256 does not match source_urdf_path"
)
hand_serial = LaunchConfiguration("serial_number").perform(context)
if (
not hand_serial
or hand_serial == "UNSET"
or re.fullmatch(r"[A-Za-z0-9_.-]+", hand_serial) is None
or hand_serial in {".", ".."}
):
raise RuntimeError("serial_number must be a safe non-empty hand serial")
requested_session = LaunchConfiguration("session_dir").perform(context)
output_root = Path(
LaunchConfiguration("output_root").perform(context)
).expanduser().resolve()
if requested_session:
session_dir = Path(requested_session).expanduser().resolve()
else:
session_dir = (
output_root
/ hand_serial
/ datetime.now().strftime("%Y%m%d_%H%M%S")
)
session_dir.mkdir(parents=True, exist_ok=True)
camera_serials = {
view: LaunchConfiguration(f"{view}_camera_serial").perform(context)
for view in views
}
if any(not serial for serial in camera_serials.values()):
raise RuntimeError("all required observation camera serial numbers are required")
if len(set(camera_serials.values())) != len(views):
raise RuntimeError("required observation camera serial numbers must be unique")
cameras = []
components = []
raw_topics = []
info_topics = []
detection_topics = []
calibration_namespace = contract.typed_profile.namespace
for view in views:
namespace = f"{calibration_namespace}/{view}/camera"
raw_topic = f"{namespace}/image_raw"
info_topic = f"{namespace}/camera_info"
rect_topic = f"{namespace}/image_rect"
detector_namespace = f"{calibration_namespace}/{view}/apriltag"
detection_topic = f"{detector_namespace}/detections"
raw_topics.append(raw_topic)
info_topics.append(info_topic)
detection_topics.append(detection_topic)
cameras.append(
Node(
package="linkerhand_calibration",
executable="hikrobot_camera_node",
name="hikrobot_camera",
namespace=namespace,
output="screen",
emulate_tty=True,
condition=IfCondition(LaunchConfiguration("start_cameras")),
parameters=[
{
"serial_number": LaunchConfiguration(
f"{view}_camera_serial"
),
"expected_model": LaunchConfiguration("camera_model"),
"camera_name": LaunchConfiguration(
f"{view}_camera_name"
),
"frame_id": (
f"{model.lower()}_calibration_{view}_optical_frame"
),
"image_width": 1624,
"image_height": 1240,
"timestamp_journal_path": str(Path(LaunchConfiguration("session_dir").perform(context))
/ f"camera_timing_{view}.jsonl"),
"frame_rate": ParameterValue(
LaunchConfiguration("camera_frame_rate"),
value_type=float,
),
"exposure_time_us": ParameterValue(
LaunchConfiguration("exposure_time_us"),
value_type=float,
),
"gain_db": ParameterValue(
LaunchConfiguration("gain_db"), value_type=float
),
"auto_exposure": ParameterValue(
LaunchConfiguration("auto_exposure"), value_type=bool
),
"camera_info_url": LaunchConfiguration(
f"{view}_camera_info_url"
),
}
],
)
)
components.extend(
[
ComposableNode(
package="image_proc",
plugin="image_proc::RectifyNode",
name=f"rectify_{view}",
namespace=namespace,
remappings=[
("image", raw_topic),
("camera_info", info_topic),
("image_rect", rect_topic),
],
parameters=[{"queue_size": 1}],
extra_arguments=[{"use_intra_process_comms": True}],
),
ComposableNode(
package="apriltag_ros",
plugin="AprilTagNode",
name="apriltag",
namespace=detector_namespace,
# Detector settings belong to the protected model YAML.
# A generic launch default must not silently replace its
# full-resolution setting for the small calibration Tags.
parameters=[str(tag_config)],
remappings=[
("image_rect", rect_topic),
("camera_info", info_topic),
],
extra_arguments=[{"use_intra_process_comms": True}],
),
]
)
vision = ComposableNodeContainer(
name=f"{model.lower()}_calibration_vision",
namespace="/",
package="rclcpp_components",
executable="component_container_mt",
composable_node_descriptions=components,
output="screen",
emulate_tty=True,
)
sdk = (
Node(
package="linkerhand_calibration",
executable="o12_sdk_bridge",
name="o12_sdk_bridge",
output="screen",
condition=IfCondition(LaunchConfiguration("start_sdk")),
parameters=[{
"vendor_config": LaunchConfiguration("vendor_sdk_config"),
"vendor_config_sha256": LaunchConfiguration("sdk_config_expected_sha256"),
"vendor_python_package": LaunchConfiguration("vendor_sdk_python_package"),
"vendor_package_sha256": LaunchConfiguration("sdk_package_expected_sha256"),
"hand_type": hand_type,
"topic_prefix": f"/{model.lower()}/{hand_type}",
}],
)
if uses_hcan
else Node(
package="linker_hand_ros2_sdk",
executable="linker_hand_sdk",
name="linker_hand_sdk",
output="screen",
condition=IfCondition(LaunchConfiguration("start_sdk")),
parameters=[{
"hand_type": hand_type,
"hand_joint": model,
"can": LaunchConfiguration("can_interface"),
"modbus": "None",
"topic_prefix": topic_prefix,
"move_on_startup": False,
"startup_speed": ParameterValue(
LaunchConfiguration("calibration_speed"), value_type=int
),
"startup_torque": 80,
# Match 30 Hz cameras so state/image p95 skew stays below 50 ms.
"state_poll_rate": 30.0,
# Calibration does not consume measured joint velocity. A
# G20 velocity read sends another five synchronous CAN
# queries, so keep it off the trajectory-critical path.
"velocity_poll_rate": 1.0,
# G20 sends an endpoint and L6 streams a bounded trajectory.
# Keep polling the real motor state during either command path;
# otherwise the SDK republishes stale state and creates large
# command-unit holes in the trajectory bins.
"defer_state_reads_while_commanding": False,
"repeat_position_commands": False,
"is_touch": False,
}],
)
)
calibration = Node(
package="linkerhand_calibration",
executable="three_camera_calibration_node",
name=f"{model.lower()}_calibration",
output="screen",
emulate_tty=True,
arguments=[
"--profile-id",
contract.typed_profile.key.profile_id,
"--profile-config", profile_path, "--profile-sha256", expected,
],
parameters=[
LaunchConfiguration("calibration_config"),
{
"serial_number": hand_serial,
"session_dir": str(session_dir),
"resume_raw_samples_path": LaunchConfiguration(
"resume_raw_samples_path"
),
"command_topic": command_topic,
"state_topic": state_topic,
"camera_extrinsics_file": LaunchConfiguration(
"camera_extrinsics_file"
),
"source_urdf_path": str(source_urdf),
"source_urdf_expected_sha256": LaunchConfiguration(
"source_urdf_expected_sha256"
),
"camera_extrinsics_expected_sha256": LaunchConfiguration(
"camera_extrinsics_expected_sha256"
),
"calibration_config_expected_sha256": LaunchConfiguration(
"calibration_config_expected_sha256"
),
"tag_config_expected_sha256": LaunchConfiguration(
"tag_config_expected_sha256"
),
"sdk_config_expected_sha256": LaunchConfiguration(
"sdk_config_expected_sha256"
),
"sdk_package_expected_sha256": LaunchConfiguration("sdk_package_expected_sha256"),
"profile_config_expected_sha256": LaunchConfiguration(
"profile_config_expected_sha256"
),
"commands_enabled": ParameterValue(
LaunchConfiguration("commands_enabled"), value_type=bool
),
"diagnostic_capture": ParameterValue(
LaunchConfiguration("diagnostic_capture"), value_type=str
),
},
],
)
bag = ExecuteProcess(
condition=IfCondition(LaunchConfiguration("record_bag")),
cmd=[
"ros2",
"bag",
"record",
"--storage",
"mcap",
"--storage-preset-profile",
"zstd_fast",
"--max-bag-size",
"10737418240",
"--output",
str(session_dir / "rosbag"),
*raw_topics,
*info_topics,
*detection_topics,
command_topic,
state_topic,
*(
[
f"/{model.lower()}/{hand_type}/calibration_health",
]
if uses_hcan else []
),
f"{calibration_namespace}/status",
],
output="screen",
)
return [
LogInfo(
msg=(
f"{model} {hand_type} {tag_layout} calibration session: {session_dir}; "
f"source_urdf={source_urdf}"
)
),
LogInfo(
msg=(
"Camera mapping: " + " ".join(f"{view}={serial}" for view, serial in camera_serials.items())
)
),
*cameras,
vision,
Node(package="linkerhand_calibration", executable="tag_border_filter_node",
name="tag_border_filter", output="screen",
arguments=["--profile-config", profile_path, "--profile-sha256", expected,
"--views", *views]),
sdk,
calibration,
bag,
]
def generate_launch_description() -> LaunchDescription:
package_share = Path(
get_package_share_directory("linkerhand_calibration")
)
return LaunchDescription(
[
# Camera processes publish ~2 MB frames across DDS. Force the
# matching RMW and provide both current and legacy profile names
# so the configured 64 MB shared-memory segment is actually used.
SetEnvironmentVariable(
name="RMW_IMPLEMENTATION",
value="rmw_fastrtps_cpp",
),
SetEnvironmentVariable(
name="FASTDDS_DEFAULT_PROFILES_FILE",
value=str(package_share / "config" / "fastdds_large_images.xml"),
),
SetEnvironmentVariable(
name="FASTRTPS_DEFAULT_PROFILES_FILE",
value=str(package_share / "config" / "fastdds_large_images.xml"),
),
DeclareLaunchArgument("model", default_value=""),
DeclareLaunchArgument("hand_type", default_value=""),
DeclareLaunchArgument("tag_layout", default_value=""),
DeclareLaunchArgument("serial_number", default_value="UNSET"),
DeclareLaunchArgument("camera_model", default_value="MV-CS020-10U"),
DeclareLaunchArgument("camera_frame_rate", default_value="30.0"),
DeclareLaunchArgument("exposure_time_us", default_value="5000.0"),
DeclareLaunchArgument("gain_db", default_value="0.0"),
DeclareLaunchArgument("auto_exposure", default_value="false"),
DeclareLaunchArgument("can_interface", default_value="can0"),
DeclareLaunchArgument("calibration_speed", default_value="15"),
DeclareLaunchArgument("camera_extrinsics_file", default_value=""),
DeclareLaunchArgument(
"source_urdf_path", default_value=""
),
DeclareLaunchArgument(
"source_urdf_expected_sha256", default_value=""
),
DeclareLaunchArgument(
"camera_extrinsics_expected_sha256", default_value=""
),
DeclareLaunchArgument(
"calibration_config_expected_sha256", default_value=""
),
DeclareLaunchArgument(
"tag_config_expected_sha256", default_value=""
),
DeclareLaunchArgument("vendor_sdk_config", default_value=""),
DeclareLaunchArgument("vendor_sdk_python_package", default_value=""),
DeclareLaunchArgument("sdk_package_expected_sha256", default_value=""),
DeclareLaunchArgument(
"sdk_config_expected_sha256", default_value=""
),
DeclareLaunchArgument(
"profile_config_expected_sha256", default_value=""
),
DeclareLaunchArgument("profile_config", default_value=""),
DeclareLaunchArgument("commands_enabled", default_value="true"),
DeclareLaunchArgument("diagnostic_capture", default_value=""),
DeclareLaunchArgument("start_cameras", default_value="true"),
DeclareLaunchArgument("start_sdk", default_value="true"),
DeclareLaunchArgument("record_bag", default_value="false"),
DeclareLaunchArgument(
"output_root",
default_value=str(Path.cwd() / "calibration_output"),
),
DeclareLaunchArgument("session_dir", default_value=""),
DeclareLaunchArgument("resume_raw_samples_path", default_value=""),
DeclareLaunchArgument(
"calibration_config",
default_value="",
),
DeclareLaunchArgument(
"tag_config",
default_value="",
),
OpaqueFunction(function=_launch_stack),
]
)
@@ -0,0 +1,209 @@
"""Launch three Hikrobot cameras for one-time checkerboard extrinsics."""
from pathlib import Path
from ament_index_python.packages import get_package_share_directory
from launch import LaunchDescription
from launch.actions import DeclareLaunchArgument, OpaqueFunction, SetEnvironmentVariable
from launch.substitutions import LaunchConfiguration
from launch_ros.actions import ComposableNodeContainer, Node
from launch_ros.descriptions import ComposableNode
from launch_ros.parameter_descriptions import ParameterValue
VIEWS = ("front", "side", "top")
def _launch(context):
cameras = []
rectifiers = []
serials = {}
for view in VIEWS:
serial = LaunchConfiguration(f"{view}_camera_serial").perform(context)
if not serial:
raise RuntimeError(f"{view}_camera_serial is required")
serials[view] = serial
namespace = f"/g20_extrinsics/{view}/camera"
cameras.append(
Node(
package="linkerhand_calibration",
executable="hikrobot_camera_node",
name="hikrobot_camera",
namespace=namespace,
output="screen",
emulate_tty=True,
parameters=[
{
"serial_number": serial,
"expected_model": LaunchConfiguration("camera_model"),
"camera_name": LaunchConfiguration(
f"{view}_camera_name"
),
"frame_id": f"g20_extrinsics_{view}_optical_frame",
"image_width": 1624,
"image_height": 1240,
"frame_rate": ParameterValue(
LaunchConfiguration("camera_frame_rate"),
value_type=float,
),
"exposure_time_us": ParameterValue(
LaunchConfiguration("exposure_time_us"),
value_type=float,
),
"gain_db": ParameterValue(
LaunchConfiguration("gain_db"), value_type=float
),
"auto_exposure": False,
"camera_info_url": LaunchConfiguration(
f"{view}_camera_info_url"
),
}
],
)
)
rectifiers.append(
ComposableNode(
package="image_proc",
plugin="image_proc::RectifyNode",
name=f"rectify_{view}",
namespace=namespace,
remappings=[
("image", f"{namespace}/image_raw"),
("camera_info", f"{namespace}/camera_info"),
("image_rect", f"{namespace}/image_rect"),
],
parameters=[{"queue_size": 1}],
extra_arguments=[{"use_intra_process_comms": True}],
)
)
container = ComposableNodeContainer(
name="g20_extrinsics_vision",
namespace="/",
package="rclcpp_components",
executable="component_container_mt",
composable_node_descriptions=rectifiers,
output="screen",
)
solver = Node(
package="linkerhand_calibration",
executable="three_camera_extrinsics_node",
name="g20_camera_extrinsics",
output="screen",
emulate_tty=True,
parameters=[
{
"output_file": LaunchConfiguration("output_file"),
"checkerboard_columns": ParameterValue(
LaunchConfiguration("checkerboard_columns"), value_type=int
),
"checkerboard_rows": ParameterValue(
LaunchConfiguration("checkerboard_rows"), value_type=int
),
"square_size_m": ParameterValue(
LaunchConfiguration("square_size_m"), value_type=float
),
"enable_gui": ParameterValue(
LaunchConfiguration("enable_gui"), value_type=bool
),
"gui_refresh_hz": ParameterValue(
LaunchConfiguration("gui_refresh_hz"), value_type=float
),
"maximum_reprojection_rms_px": ParameterValue(
LaunchConfiguration("maximum_reprojection_rms_px"),
value_type=float,
),
"maximum_candidate_pair_reprojection_rms_px": ParameterValue(
LaunchConfiguration(
"maximum_candidate_pair_reprojection_rms_px"
),
value_type=float,
),
"maximum_single_camera_reprojection_rms_px": ParameterValue(
LaunchConfiguration(
"maximum_single_camera_reprojection_rms_px"
),
value_type=float,
),
"auto_capture_default": ParameterValue(
LaunchConfiguration("auto_capture_default"),
value_type=bool,
),
"auto_capture_stable_seconds": ParameterValue(
LaunchConfiguration("auto_capture_stable_seconds"),
value_type=float,
),
**{
f"{view}_camera_serial": serials[view]
for view in VIEWS
},
}
],
)
return [*cameras, container, solver]
def generate_launch_description() -> LaunchDescription:
package_share = Path(
get_package_share_directory("linkerhand_calibration")
)
camera_info = Path.home() / ".ros" / "camera_info"
return LaunchDescription(
[
# Keep the large-image transport deterministic even when the
# calling shell selected another ROS 2 RMW implementation.
SetEnvironmentVariable(
name="RMW_IMPLEMENTATION",
value="rmw_fastrtps_cpp",
),
SetEnvironmentVariable(
name="FASTDDS_DEFAULT_PROFILES_FILE",
value=str(
package_share / "config" / "fastdds_large_images.xml"
),
),
SetEnvironmentVariable(
name="FASTRTPS_DEFAULT_PROFILES_FILE",
value=str(
package_share / "config" / "fastdds_large_images.xml"
),
),
DeclareLaunchArgument("front_camera_serial", default_value="DB2163742"),
DeclareLaunchArgument("side_camera_serial", default_value="DB2163749"),
DeclareLaunchArgument("top_camera_serial", default_value="DB2163739"),
DeclareLaunchArgument("camera_model", default_value="MV-CS020-10U"),
DeclareLaunchArgument("front_camera_name", default_value="hikrobot_front_DB2163742"),
DeclareLaunchArgument("side_camera_name", default_value="hikrobot_side_DB2163749"),
DeclareLaunchArgument("top_camera_name", default_value="hikrobot_top_DB2163739"),
DeclareLaunchArgument("front_camera_info_url", default_value=str(camera_info / "hikrobot_DB2163742.yaml")),
DeclareLaunchArgument("side_camera_info_url", default_value=str(camera_info / "hikrobot_DB2163749.yaml")),
DeclareLaunchArgument("top_camera_info_url", default_value=str(camera_info / "hikrobot_DB2163739.yaml")),
DeclareLaunchArgument("camera_frame_rate", default_value="15.0"),
DeclareLaunchArgument("exposure_time_us", default_value="5000.0"),
DeclareLaunchArgument("gain_db", default_value="0.0"),
DeclareLaunchArgument("checkerboard_columns", default_value="8"),
DeclareLaunchArgument("checkerboard_rows", default_value="5"),
DeclareLaunchArgument("square_size_m", default_value="0.027"),
DeclareLaunchArgument("enable_gui", default_value="true"),
DeclareLaunchArgument("gui_refresh_hz", default_value="2.0"),
DeclareLaunchArgument(
"maximum_reprojection_rms_px", default_value="1.2"
),
DeclareLaunchArgument(
"maximum_candidate_pair_reprojection_rms_px",
default_value="1.5",
),
DeclareLaunchArgument(
"maximum_single_camera_reprojection_rms_px",
default_value="1.5",
),
DeclareLaunchArgument("auto_capture_default", default_value="false"),
DeclareLaunchArgument(
"auto_capture_stable_seconds", default_value="1.0"
),
DeclareLaunchArgument(
"output_file",
default_value=str(Path.cwd() / "config" / "g20_three_camera_extrinsics.yaml"),
),
OpaqueFunction(function=_launch),
]
)
@@ -0,0 +1,18 @@
"""Stable launch name for the profile-driven calibration stack."""
from __future__ import annotations
import importlib.util
from pathlib import Path
def generate_launch_description():
implementation = Path(__file__).with_name("three_camera_calibration.launch.py")
spec = importlib.util.spec_from_file_location(
"linkerhand_unified_calibration_launch", implementation
)
if spec is None or spec.loader is None:
raise RuntimeError("unified calibration launch implementation is missing")
module = importlib.util.module_from_spec(spec)
spec.loader.exec_module(module)
return module.generate_launch_description()
@@ -0,0 +1,5 @@
"""Profile-driven LinkerHand calibration and validated URDF correction."""
from .core import CalibrationProfile, ProfileKey
__all__ = ["CalibrationProfile", "ProfileKey"]
@@ -0,0 +1,748 @@
"""Hardware-independent point acquisition state."""
from __future__ import annotations
from bisect import bisect_left
from collections import deque
from dataclasses import dataclass, field
from typing import Any, Mapping, Sequence
import numpy as np
from .core import robust_rotation_summary
from .pnp import SquareTagPose
TAG_PAIR_ROLES: dict[str, tuple[str, str]] = {
"t0_t3": ("t0", "t3"),
"t3_t4": ("t3", "t4"),
"t4_t5": ("t4", "t5"),
}
PAIR_NAMES: tuple[str, ...] = tuple(TAG_PAIR_ROLES)
@dataclass(frozen=True)
class TagQuality:
hamming: int
decision_margin: float
edge_pixels: float
reprojection_error_px: float | None = None
def tag_quality_is_valid(
quality: TagQuality,
*,
maximum_hamming: int,
minimum_decision_margin: float,
minimum_edge_pixels: float,
maximum_reprojection_error_px: float | None = None,
) -> bool:
detection_valid = (
quality.hamming <= maximum_hamming
and quality.decision_margin >= minimum_decision_margin
and quality.edge_pixels >= minimum_edge_pixels
)
if not detection_valid:
return False
if maximum_reprojection_error_px is None:
return True
return (
quality.reprojection_error_px is not None
and quality.reprojection_error_px <= maximum_reprojection_error_px
)
def update_pnp_reset_watchdog(
*,
detection_good: bool,
pnp_valid: bool,
now: float,
invalid_since: float | None,
reset_after_seconds: float,
) -> tuple[float | None, bool]:
"""Track continuous PnP-only failures and request a throttled reset."""
reset_after = float(reset_after_seconds)
if reset_after <= 0.0:
raise ValueError("reset_after_seconds must be positive")
if not detection_good or pnp_valid:
return None, False
since = float(now) if invalid_since is None else float(invalid_since)
if float(now) - since >= reset_after:
# Start a new interval so a permanently bad view is not reset on every
# frame. The next valid frame clears the interval.
return float(now), True
return since, False
def required_resume_views(active_view: str | None) -> tuple[str, ...]:
"""Require only the active view on resume; start still checks all views."""
all_views = ("front", "side", "top")
if active_view is None:
return all_views
view = str(active_view)
if view not in all_views:
raise ValueError(f"unknown calibration view: {view}")
return (view,)
@dataclass(frozen=True)
class Observation:
stamp_ns: int
received_at: float
relative_quaternion_xyzw: Mapping[str, tuple[float, float, float, float]]
tag_quality: Mapping[str, TagQuality]
state_u8: tuple[float, ...] = ()
state_stamp_ns: int | None = None
state_sync_error_ns: int | None = None
tag_quaternion_xyzw: Mapping[
str, tuple[float, float, float, float]
] = field(default_factory=dict)
tag_translation_xyz_m: Mapping[
str, tuple[float, float, float]
] = field(default_factory=dict)
tag_pose_candidates: Mapping[
str, tuple[SquareTagPose, ...]
] = field(default_factory=dict)
@dataclass(frozen=True)
class StateSample:
stamp_ns: int
position_u8: tuple[float, ...]
channel_stamps_ns: tuple[int, ...] = ()
def interpolate_state_u8(
samples: Sequence[StateSample],
stamp_ns: int,
*,
maximum_skew_ns: int,
) -> tuple[tuple[float, ...], int] | None:
"""Interpolate a profile-sized hand state at an image timestamp.
The SDK publishes state independently from the camera. Continuous
calibration must therefore use the image timestamp instead of whichever
state happened to arrive most recently in the ROS callback thread.
"""
if maximum_skew_ns < 0:
raise ValueError("maximum_skew_ns must be non-negative")
if not samples:
return None
if any(sample.channel_stamps_ns for sample in samples):
# A multi-frame SDK does not observe all motors at the same instant.
# De-duplicate each channel by its CAN receipt, then interpolate it
# independently. Repeated publication never creates extra support.
samples = tuple(sample for sample in samples if sample.stamp_ns >= stamp_ns-maximum_skew_ns
and sample.channel_stamps_ns and min(sample.channel_stamps_ns) <= stamp_ns+maximum_skew_ns)
if not samples:
return None
count = len(samples[0].position_u8)
channels = [dict() for _ in range(count)]
for sample in samples:
if len(sample.position_u8) != count or len(sample.channel_stamps_ns) != count:
return None
for index, (stamp, value) in enumerate(zip(sample.channel_stamps_ns, sample.position_u8)):
if stamp in channels[index] and channels[index][stamp] != value:
return None
channels[index][stamp] = value
matches = [interpolate_state_u8(tuple(StateSample(stamp, (value,))
for stamp, value in sorted(channel.items())), stamp_ns, maximum_skew_ns=maximum_skew_ns)
for channel in channels]
if any(match is None for match in matches):
return None
return tuple(match[0][0] for match in matches), max(match[1] for match in matches)
stamps = [int(sample.stamp_ns) for sample in samples]
index = bisect_left(stamps, int(stamp_ns))
if index < len(samples) and stamps[index] == int(stamp_ns):
state = samples[index].position_u8
return (tuple(float(value) for value in state), 0)
before = samples[index - 1] if index > 0 else None
after = samples[index] if index < len(samples) else None
if before is not None and after is not None:
before_gap = int(stamp_ns) - int(before.stamp_ns)
after_gap = int(after.stamp_ns) - int(stamp_ns)
nearest_gap = min(before_gap, after_gap)
if nearest_gap > maximum_skew_ns:
return None
denominator = int(after.stamp_ns) - int(before.stamp_ns)
if denominator <= 0:
return (
tuple(float(value) for value in before.position_u8),
nearest_gap,
)
fraction = before_gap / denominator
before_values = np.asarray(before.position_u8, dtype=float)
after_values = np.asarray(after.position_u8, dtype=float)
if (
before_values.ndim != 1
or before_values.size == 0
or after_values.shape != before_values.shape
):
return None
interpolated = before_values + fraction * (after_values - before_values)
return (
tuple(float(value) for value in interpolated),
nearest_gap,
)
nearest = before if before is not None else after
if nearest is None:
return None
gap = abs(int(stamp_ns) - int(nearest.stamp_ns))
if gap > maximum_skew_ns or not nearest.position_u8:
return None
return (tuple(float(value) for value in nearest.position_u8), gap)
# Physical-angle profiles use the same timestamp interpolation. Keep the old
# public name for compatibility and offer a unit-neutral spelling to new code.
interpolate_state = interpolate_state_u8
class ContinuousSweepCollector:
"""Collect timestamp-synchronised observations during one end-to-end move."""
def __init__(
self,
*,
endpoint_tolerance_u8: float = 2.0,
endpoint_hold_seconds: float = 1.0,
timeout_seconds: float = 90.0,
invalid_timeout_seconds: float = 2.0,
minimum_valid_frames: int = 40,
minimum_state_span_u8: float = 240.0,
) -> None:
if endpoint_tolerance_u8 < 0.0:
raise ValueError("endpoint_tolerance_u8 must be non-negative")
if endpoint_hold_seconds <= 0.0:
raise ValueError("endpoint_hold_seconds must be positive")
if timeout_seconds <= 0.0 or invalid_timeout_seconds <= 0.0:
raise ValueError("sweep timeouts must be positive")
if minimum_valid_frames < 3:
raise ValueError("minimum_valid_frames must be at least 3")
if minimum_state_span_u8 <= 0.0:
raise ValueError("minimum_state_span_u8 must be positive")
self.endpoint_tolerance_u8 = float(endpoint_tolerance_u8)
self.endpoint_hold_seconds = float(endpoint_hold_seconds)
self.timeout_seconds = float(timeout_seconds)
self.invalid_timeout_seconds = float(invalid_timeout_seconds)
self.minimum_valid_frames = int(minimum_valid_frames)
self.minimum_state_span_u8 = float(minimum_state_span_u8)
self.observations: list[Observation] = []
self.motor_index = 0
self.start_u8 = 255.0
self.target_u8 = 0.0
self.started_at: float | None = None
self.last_valid_at: float | None = None
self.endpoint_since: float | None = None
self.state = "idle"
self.reason = ""
def start(
self,
now: float,
*,
motor_index: int,
start_u8: int,
target_u8: int,
) -> None:
if motor_index not in (0, 15):
raise ValueError("continuous thumb sweep only permits motor 0 or 15")
if {int(start_u8), int(target_u8)} != {0, 255}:
raise ValueError("continuous sweep endpoints must be 0 and 255")
self.observations.clear()
self.motor_index = int(motor_index)
self.start_u8 = float(start_u8)
self.target_u8 = float(target_u8)
self.started_at = float(now)
self.last_valid_at = float(now)
self.endpoint_since = None
self.state = "collecting"
self.reason = ""
@property
def active(self) -> bool:
return self.state == "collecting"
@property
def valid_frames_seen(self) -> int:
return len(self.observations)
@property
def state_span_u8(self) -> float:
if not self.observations:
return 0.0
values = [
float(observation.state_u8[self.motor_index])
for observation in self.observations
]
return float(max(values) - min(values))
def add(
self, observation: Observation, now: float
) -> list[Observation] | None:
if not self.active:
return None
if (
len(observation.state_u8) != 20
or observation.state_sync_error_ns is None
):
return None
value = float(observation.state_u8[self.motor_index])
if not np.isfinite(value) or not -3.0 <= value <= 258.0:
return None
now = float(now)
self.observations.append(observation)
self.last_valid_at = now
if abs(value - self.target_u8) <= self.endpoint_tolerance_u8:
if self.endpoint_since is None:
self.endpoint_since = now
else:
self.endpoint_since = None
enough_endpoint_hold = (
self.endpoint_since is not None
and now - self.endpoint_since >= self.endpoint_hold_seconds
)
if (
enough_endpoint_hold
and len(self.observations) >= self.minimum_valid_frames
and self.state_span_u8 >= self.minimum_state_span_u8
):
self.state = "complete"
return list(self.observations)
return None
def poll(self, now: float) -> None:
if not self.active:
return
now = float(now)
if now - float(self.started_at) > self.timeout_seconds:
self.state = "failed"
self.reason = "sweep_timeout"
elif now - float(self.last_valid_at) > self.invalid_timeout_seconds:
self.state = "failed"
self.reason = "synchronised_tag_state_timeout"
def aggregate_sweep_observations(
observations: Sequence[Observation],
*,
motor_index: int,
start_u8: int,
target_u8: int,
endpoint_tolerance_u8: float,
) -> dict[int, dict[str, Any]]:
"""Robustly aggregate continuous observations into integer motor bins."""
if not observations:
raise ValueError("cannot aggregate an empty continuous sweep")
bins: dict[int, list[Observation]] = {}
for observation in observations:
if len(observation.state_u8) != 20:
continue
value = float(observation.state_u8[motor_index])
if abs(value - float(start_u8)) <= endpoint_tolerance_u8:
command = int(start_u8)
elif abs(value - float(target_u8)) <= endpoint_tolerance_u8:
command = int(target_u8)
else:
command = int(np.clip(np.rint(value), 0, 255))
bins.setdefault(command, []).append(observation)
return {
command: aggregate_observations(values)
for command, values in sorted(bins.items())
}
class PointCollector:
"""Wait for a stable pose, then aggregate a fixed number of frames."""
def __init__(
self,
*,
stable_frames: int = 15,
capture_frames: int = 30,
minimum_settle_seconds: float = 0.4,
maximum_stable_spread_rad: float = np.deg2rad(0.3),
stability_mode: str = "rotation",
maximum_stable_translation_spread_m: float = 0.003,
settle_timeout_seconds: float = 5.0,
capture_timeout_seconds: float = 5.0,
) -> None:
if stable_frames < 3 or capture_frames < 3:
raise ValueError("stable_frames and capture_frames must be at least 3")
self.stable_frames = int(stable_frames)
self.capture_frames = int(capture_frames)
self.minimum_settle_seconds = float(minimum_settle_seconds)
self.maximum_stable_spread_rad = float(maximum_stable_spread_rad)
self.stability_mode = str(stability_mode)
self.maximum_stable_translation_spread_m = float(
maximum_stable_translation_spread_m
)
if self.stability_mode not in {"rotation", "translation"}:
raise ValueError(
"stability_mode must be rotation or translation"
)
if self.maximum_stable_translation_spread_m <= 0.0:
raise ValueError(
"maximum_stable_translation_spread_m must be positive"
)
self.settle_timeout_seconds = float(settle_timeout_seconds)
self.capture_timeout_seconds = float(capture_timeout_seconds)
self._stable: deque[Observation] = deque(maxlen=self.stable_frames)
self._captured: list[Observation] = []
self._consecutive_invalid_frames = 0
self.started_at: float | None = None
self.capture_started_at: float | None = None
self.state = "idle"
self.reason = ""
self.stable_spread_rad: dict[str, float] = {}
self.stable_spread_m: dict[str, float] = {}
self.required_state_index: int | None = None
self.required_state_u8: float | None = None
self.maximum_state_error_u8: float | None = None
def start(
self,
now: float,
*,
required_state_index: int | None = None,
required_state_u8: float | None = None,
maximum_state_error_u8: float | None = None,
) -> None:
state_constraints = (
required_state_index,
required_state_u8,
maximum_state_error_u8,
)
if any(value is not None for value in state_constraints) and not all(
value is not None for value in state_constraints
):
raise ValueError(
"point state constraint parameters must be provided together"
)
if required_state_index is not None:
if not 0 <= int(required_state_index) < 20:
raise ValueError("required_state_index must be in [0, 19]")
if not np.isfinite(float(required_state_u8)):
raise ValueError("required_state_u8 must be finite")
if float(maximum_state_error_u8) < 0.0:
raise ValueError(
"maximum_state_error_u8 must be non-negative"
)
self._stable.clear()
self._captured.clear()
self._consecutive_invalid_frames = 0
self.started_at = float(now)
self.capture_started_at = None
self.state = "settling"
self.reason = ""
self.stable_spread_rad = {}
self.stable_spread_m = {}
self.required_state_index = (
None
if required_state_index is None
else int(required_state_index)
)
self.required_state_u8 = (
None if required_state_u8 is None else float(required_state_u8)
)
self.maximum_state_error_u8 = (
None
if maximum_state_error_u8 is None
else float(maximum_state_error_u8)
)
@property
def active(self) -> bool:
return self.state in {"settling", "capturing"}
@property
def stable_frames_seen(self) -> int:
return len(self._stable)
@property
def capture_frames_seen(self) -> int:
return len(self._captured)
def _window_is_stable(self) -> bool:
if len(self._stable) < self.stable_frames:
return False
if self.stability_mode == "translation":
spreads: dict[str, float] = {}
for pair, (parent, child) in TAG_PAIR_ROLES.items():
if any(
parent not in observation.tag_translation_xyz_m
or child not in observation.tag_translation_xyz_m
for observation in self._stable
):
self.reason = f"{pair}_translation_missing"
return False
vectors = np.asarray(
[
np.asarray(
observation.tag_translation_xyz_m[child],
dtype=float,
)
- np.asarray(
observation.tag_translation_xyz_m[parent],
dtype=float,
)
for observation in self._stable
],
dtype=float,
)
reference = np.median(vectors, axis=0)
spreads[pair] = float(
np.max(np.linalg.norm(vectors - reference, axis=1))
)
self.stable_spread_m = spreads
for pair, spread in spreads.items():
if spread > self.maximum_stable_translation_spread_m:
self.reason = f"{pair}_not_stable"
return False
return True
spreads: dict[str, float] = {}
for pair in PAIR_NAMES:
quaternions = [
observation.relative_quaternion_xyzw[pair]
for observation in self._stable
]
_, spread = robust_rotation_summary(quaternions)
spreads[pair] = float(spread)
self.stable_spread_rad = spreads
for pair, spread in spreads.items():
if spread > self.maximum_stable_spread_rad:
self.reason = f"{pair}_not_stable"
return False
return True
def _state_is_acceptable(self, observation: Observation) -> bool:
if self.required_state_index is None:
return True
if (
len(observation.state_u8) != 20
or observation.state_sync_error_ns is None
):
return False
value = float(observation.state_u8[self.required_state_index])
return bool(
np.isfinite(value)
and abs(value - float(self.required_state_u8))
<= float(self.maximum_state_error_u8)
)
def _return_to_settling(self, reason: str) -> None:
self._stable.clear()
self._captured.clear()
self._consecutive_invalid_frames = 0
self.capture_started_at = None
self.state = "settling"
self.reason = str(reason)
self.stable_spread_rad = {}
self.stable_spread_m = {}
def add(
self, observation: Observation, now: float
) -> dict[str, Any] | None:
if not self.active:
return None
self._consecutive_invalid_frames = 0
now = float(now)
if not self._state_is_acceptable(observation):
self._return_to_settling("motor_position_out_of_tolerance")
return None
if self.state == "settling":
self._stable.append(observation)
elapsed = now - float(self.started_at)
if elapsed >= self.minimum_settle_seconds and self._window_is_stable():
self.state = "capturing"
self.capture_started_at = now
self._captured.clear()
self.reason = ""
return None
self._captured.append(observation)
if len(self._captured) < self.capture_frames:
return None
# Validate continuity across the boundary as well as inside the
# capture block. A planar branch can switch immediately after the
# stable window and then look perfectly stable for every capture
# frame; checking only the captured frames would accept that jump.
stability_aggregate = aggregate_observations(
[*self._stable, *self._captured]
)
if self.stability_mode == "translation":
unstable_pairs = [
pair
for pair, spread in stability_aggregate[
"maximum_translation_spread_m"
].items()
if float(spread)
> self.maximum_stable_translation_spread_m
]
else:
unstable_pairs = [
pair
for pair, spread in stability_aggregate[
"maximum_spread_rad"
].items()
if float(spread) > self.maximum_stable_spread_rad
]
if unstable_pairs:
self._return_to_settling(
f"{unstable_pairs[0]}_capture_not_stable"
)
return None
self.state = "complete"
return aggregate_observations(self._captured)
def poll(self, now: float) -> None:
if not self.active:
return
now = float(now)
if self.state == "settling":
if now - float(self.started_at) > self.settle_timeout_seconds:
self.state = "failed"
self.reason = self.reason or "settle_timeout"
elif self.state == "capturing":
if now - float(self.capture_started_at) > self.capture_timeout_seconds:
self.state = "failed"
self.reason = "capture_timeout"
def mark_invalid_frame(self) -> None:
"""Skip one invalid frame while retaining the recent valid window."""
if self.state == "capturing":
self._return_to_settling("invalid_tag_frame")
return
if self.state == "settling":
self._consecutive_invalid_frames += 1
if self._consecutive_invalid_frames >= 3:
self._stable.clear()
self.reason = "invalid_tag_frame"
def aggregate_observations(
observations: Sequence[Observation],
) -> dict[str, Any]:
if not observations:
raise ValueError("cannot aggregate an empty observation sequence")
relative: dict[str, list[float]] = {}
spread: dict[str, float] = {}
for pair in PAIR_NAMES:
quaternion, maximum = robust_rotation_summary(
[
observation.relative_quaternion_xyzw[pair]
for observation in observations
]
)
relative[pair] = [float(value) for value in quaternion]
spread[pair] = float(maximum)
quality: dict[str, dict[str, float]] = {}
tag_names = sorted(observations[0].tag_quality)
for tag_name in tag_names:
values = [
observation.tag_quality[tag_name] for observation in observations
]
quality[tag_name] = {
"minimum_decision_margin": float(
min(value.decision_margin for value in values)
),
"minimum_edge_pixels": float(min(value.edge_pixels for value in values)),
"maximum_hamming": int(max(value.hamming for value in values)),
}
reprojection_errors = [
float(value.reprojection_error_px)
for value in values
if value.reprojection_error_px is not None
]
if reprojection_errors:
quality[tag_name]["maximum_reprojection_error_px"] = float(
max(reprojection_errors)
)
states = [
observation.state_u8
for observation in observations
if len(observation.state_u8) == 20
]
state_median: list[float] = []
if states:
state_median = [
float(value)
for value in np.median(np.asarray(states, dtype=float), axis=0)
]
sync_errors = [
int(observation.state_sync_error_ns)
for observation in observations
if observation.state_sync_error_ns is not None
]
tag_translations: dict[str, list[float]] = {}
translation_spread: dict[str, float] = {}
translation_roles = sorted(
set.intersection(
*(
set(observation.tag_translation_xyz_m)
for observation in observations
)
)
if observations
else set()
)
for role in translation_roles:
values = np.asarray(
[
observation.tag_translation_xyz_m[role]
for observation in observations
],
dtype=float,
)
if values.shape == (len(observations), 3) and np.all(
np.isfinite(values)
):
tag_translations[role] = [
float(value)
for value in np.median(values, axis=0)
]
for pair, (parent, child) in TAG_PAIR_ROLES.items():
if parent not in translation_roles or child not in translation_roles:
continue
vectors = np.asarray(
[
np.asarray(
observation.tag_translation_xyz_m[child],
dtype=float,
)
- np.asarray(
observation.tag_translation_xyz_m[parent],
dtype=float,
)
for observation in observations
],
dtype=float,
)
reference = np.median(vectors, axis=0)
translation_spread[pair] = float(
np.max(np.linalg.norm(vectors - reference, axis=1))
)
return {
"stamp_start_ns": int(observations[0].stamp_ns),
"stamp_end_ns": int(observations[-1].stamp_ns),
"valid_frames": len(observations),
"relative_quaternion_xyzw": relative,
"maximum_spread_rad": spread,
"tag_quality": quality,
"state_u8_median": state_median,
"tag_translation_xyz_m": tag_translations,
"maximum_translation_spread_m": translation_spread,
"maximum_state_sync_error_ms": (
None
if not sync_errors
else float(max(sync_errors)) / 1_000_000.0
),
}
@@ -0,0 +1,358 @@
"""Visual roll-alignment aid for one G20 calibration camera."""
from __future__ import annotations
from collections import deque
import math
import time
from typing import Any, Mapping, Sequence
from apriltag_msgs.msg import AprilTagDetectionArray
import cv2
from cv_bridge import CvBridge
import numpy as np
import rclpy
from rclpy.node import Node
from rclpy.qos import qos_profile_sensor_data
from sensor_msgs.msg import Image
from .full_hand import VIEW_TAGS
from .hikrobot_camera import configure_fastdds_large_image_transport
from .zero_calibration import detect_reference_alignment_line
VIEWS = ("front", "side", "top")
def summarize_alignment_measurements(
measurements: Sequence[Mapping[str, Any] | None],
) -> dict[str, Any] | None:
"""Return a median-smoothed physical reference-line measurement."""
valid = [measurement for measurement in measurements if measurement]
if not valid:
return None
return {
"line_xyxy_px": np.median(
np.asarray(
[measurement["line_xyxy_px"] for measurement in valid],
dtype=float,
),
axis=0,
).tolist(),
"angle_rad": float(
np.median(
[float(measurement["angle_rad"]) for measurement in valid]
)
),
"vertical_offset_px": float(
np.median(
[
float(measurement["vertical_offset_px"])
for measurement in valid
]
)
),
"detected_frames": len(valid),
"window_frames": len(measurements),
}
class G20CameraAlignmentView(Node):
"""Publish a red/blue roll aid based on a physical scene edge."""
def __init__(self) -> None:
"""Configure one view without taking ownership of hand commands."""
super().__init__("g20_camera_alignment_view")
self.declare_parameter("view", "front")
view = str(self.get_parameter("view").value).strip().lower()
if view not in VIEWS:
raise ValueError(f"view must be one of {VIEWS}")
self.view = view
namespace = f"/g20_calibration/{view}"
self.required_tag_ids = {
int(value) for value in VIEW_TAGS[view].values()
}
self.declare_parameter("image_topic", f"{namespace}/camera/image_rect")
self.declare_parameter(
"detections_topic", f"{namespace}/apriltag/detections"
)
self.declare_parameter("reference_y_ratio", 0.90)
self.declare_parameter("roi_y_min_ratio", 0.55)
self.declare_parameter("roi_y_max_ratio", 0.98)
self.declare_parameter("minimum_line_length_ratio", 0.30)
self.declare_parameter("maximum_candidate_angle_deg", 15.0)
self.declare_parameter("maximum_alignment_error_deg", 0.5)
self.declare_parameter("maximum_vertical_offset_px", 12.0)
self.declare_parameter("maximum_hamming", 0)
self.declare_parameter("minimum_decision_margin", 20.0)
self.declare_parameter("minimum_edge_pixels", 20.0)
self.declare_parameter("smoothing_frames", 10)
self.declare_parameter("maximum_line_age_seconds", 1.0)
self.declare_parameter("maximum_tag_age_seconds", 1.0)
self.declare_parameter("maximum_publish_rate_hz", 10.0)
self.declare_parameter("output_scale", 0.75)
def value(name: str) -> Any:
return self.get_parameter(name).value
self.image_topic = str(value("image_topic"))
self.detections_topic = str(value("detections_topic"))
self.reference_y_ratio = float(value("reference_y_ratio"))
self.roi_y_min_ratio = float(value("roi_y_min_ratio"))
self.roi_y_max_ratio = float(value("roi_y_max_ratio"))
self.minimum_line_length_ratio = float(
value("minimum_line_length_ratio")
)
self.maximum_candidate_angle_rad = math.radians(
float(value("maximum_candidate_angle_deg"))
)
self.maximum_alignment_error_rad = math.radians(
float(value("maximum_alignment_error_deg"))
)
self.maximum_vertical_offset_px = float(
value("maximum_vertical_offset_px")
)
self.maximum_hamming = int(value("maximum_hamming"))
self.minimum_decision_margin = float(value("minimum_decision_margin"))
self.minimum_edge_pixels = float(value("minimum_edge_pixels"))
self.maximum_line_age_seconds = float(
value("maximum_line_age_seconds")
)
self.maximum_tag_age_seconds = float(value("maximum_tag_age_seconds"))
self.maximum_publish_rate_hz = float(value("maximum_publish_rate_hz"))
self.output_scale = float(value("output_scale"))
smoothing_frames = int(value("smoothing_frames"))
if not (
0.0
<= self.roi_y_min_ratio
< self.reference_y_ratio
< self.roi_y_max_ratio
<= 1.0
):
raise ValueError(
"ratios must satisfy 0 <= roi_min < reference < roi_max <= 1"
)
if not 0.0 < self.minimum_line_length_ratio <= 1.0:
raise ValueError("minimum_line_length_ratio must be in (0, 1]")
if not (
0.0
< self.maximum_alignment_error_rad
< self.maximum_candidate_angle_rad
< math.pi / 2.0
):
raise ValueError(
"angle limits must satisfy 0 < alignment < candidate < 90"
)
if self.maximum_vertical_offset_px <= 0.0:
raise ValueError("maximum_vertical_offset_px must be positive")
if smoothing_frames < 1:
raise ValueError("smoothing_frames must be positive")
if self.maximum_line_age_seconds <= 0.0:
raise ValueError("maximum_line_age_seconds must be positive")
if self.maximum_tag_age_seconds <= 0.0:
raise ValueError("maximum_tag_age_seconds must be positive")
if self.maximum_publish_rate_hz <= 0.0:
raise ValueError("maximum_publish_rate_hz must be positive")
if not 0.1 <= self.output_scale <= 1.0:
raise ValueError("output_scale must be in [0.1, 1.0]")
self.bridge = CvBridge()
self.line_history: deque[dict[str, Any] | None] = deque(
maxlen=smoothing_frames
)
self.last_line_at = 0.0
self.latest_tag_corners: dict[int, np.ndarray] = {}
self.latest_tag_at: dict[int, float] = {}
self.last_publish_at = 0.0
self.publisher = self.create_publisher(
Image, "~/image", qos_profile_sensor_data
)
self.create_subscription(
AprilTagDetectionArray,
self.detections_topic,
self._detections_callback,
qos_profile_sensor_data,
)
self.create_subscription(
Image,
self.image_topic,
self._image_callback,
qos_profile_sensor_data,
)
self.get_logger().info(
f"{view} alignment view uses physical long-edge detection; "
f"Tag orientation is ignored; input={self.image_topic}; "
f"output={self.get_name()}/image"
)
def _detections_callback(self, message: AprilTagDetectionArray) -> None:
now = time.monotonic()
for detection in message.detections:
tag_id = int(detection.id)
if tag_id not in self.required_tag_ids:
continue
corners = np.asarray(
[
[float(point.x), float(point.y)]
for point in detection.corners
],
dtype=float,
)
if corners.shape != (4, 2) or not np.all(np.isfinite(corners)):
continue
edges = np.linalg.norm(
corners - np.roll(corners, -1, axis=0), axis=1
)
if (
int(detection.hamming) > self.maximum_hamming
or float(detection.decision_margin)
< self.minimum_decision_margin
or float(np.mean(edges)) < self.minimum_edge_pixels
):
continue
self.latest_tag_corners[tag_id] = corners
self.latest_tag_at[tag_id] = now
def _draw_tags(self, image: np.ndarray, now: float) -> None:
for tag_id in sorted(self.required_tag_ids):
corners = self.latest_tag_corners.get(tag_id)
detected_at = self.latest_tag_at.get(tag_id, 0.0)
if (
corners is None
or now - detected_at > self.maximum_tag_age_seconds
):
continue
points = np.rint(corners * self.output_scale).astype(np.int32)
cv2.polylines(image, [points], True, (0, 220, 0), 2)
centre = np.rint(np.mean(points, axis=0)).astype(int)
cv2.putText(
image,
f"ID {tag_id}",
(int(centre[0]) + 5, int(centre[1]) - 7),
cv2.FONT_HERSHEY_SIMPLEX,
0.55,
(0, 220, 0),
2,
)
def _image_callback(self, message: Image) -> None:
# Avoid conversion and Hough work until an image viewer subscribes.
if self.publisher.get_subscription_count() < 1:
return
now = time.monotonic()
if now - self.last_publish_at < 1.0 / self.maximum_publish_rate_hz:
return
self.last_publish_at = now
try:
image = self.bridge.imgmsg_to_cv2(message, desired_encoding="bgr8")
except Exception as error:
self.get_logger().warning(
f"alignment image conversion failed: {error}"
)
return
if self.output_scale != 1.0:
image = cv2.resize(
image,
None,
fx=self.output_scale,
fy=self.output_scale,
interpolation=cv2.INTER_AREA,
)
height, width = image.shape[:2]
reference_y = self.reference_y_ratio * float(height - 1)
detected = detect_reference_alignment_line(
image,
reference_y_px=reference_y,
roi_y_min_ratio=self.roi_y_min_ratio,
roi_y_max_ratio=self.roi_y_max_ratio,
minimum_length_ratio=self.minimum_line_length_ratio,
maximum_candidate_angle_rad=self.maximum_candidate_angle_rad,
)
self.line_history.append(detected)
if detected is not None:
self.last_line_at = now
measurement = summarize_alignment_measurements(self.line_history)
if now - self.last_line_at > self.maximum_line_age_seconds:
measurement = None
red_y = int(round(reference_y))
cv2.line(
image,
(15, red_y),
(max(15, width - 15), red_y),
(0, 0, 255),
4,
)
if measurement is not None:
line = np.rint(measurement["line_xyxy_px"]).astype(int)
blue_ok, blue_start, blue_end = cv2.clipLine(
(0, 0, width, height),
(int(line[0]), int(line[1])),
(int(line[2]), int(line[3])),
)
if blue_ok:
cv2.line(image, blue_start, blue_end, (255, 0, 0), 3)
angle_rad = float(measurement["angle_rad"])
offset_px = float(measurement["vertical_offset_px"])
aligned = bool(
abs(angle_rad) <= self.maximum_alignment_error_rad
and abs(offset_px) <= self.maximum_vertical_offset_px
)
status = "ALIGNED" if aligned else "ADJUST CAMERA"
status_text = (
f"{self.view.upper()} red-blue "
f"{math.degrees(angle_rad):+.2f} deg "
f"dy {offset_px:+.1f}px {status}"
)
status_color = (0, 220, 0) if aligned else (0, 165, 255)
else:
status_text = (
f"{self.view.upper()} PHYSICAL REFERENCE LINE NOT DETECTED"
)
status_color = (0, 165, 255)
cv2.putText(
image,
status_text,
(20, 34),
cv2.FONT_HERSHEY_SIMPLEX,
0.72,
status_color,
2,
)
self._draw_tags(image, now)
cv2.putText(
image,
"RED=target BLUE=physical edge GREEN=Tags (angle ignored)",
(20, max(64, height - 24)),
cv2.FONT_HERSHEY_SIMPLEX,
0.60,
(255, 255, 255),
2,
)
output = self.bridge.cv2_to_imgmsg(image, encoding="bgr8")
output.header = message.header
self.publisher.publish(output)
def main(args: list[str] | None = None) -> None:
"""Run the single-view alignment helper."""
configure_fastdds_large_image_transport()
rclpy.init(args=args)
node: G20CameraAlignmentView | None = None
try:
node = G20CameraAlignmentView()
rclpy.spin(node)
except KeyboardInterrupt:
pass
finally:
if node is not None:
node.destroy_node()
if rclpy.ok():
rclpy.shutdown()
if __name__ == "__main__":
main()
@@ -0,0 +1,377 @@
"""Map explicit SDK commands or feedback to corrected URDF joint coordinates.
Unified artifacts load their certified manifest and standard URDF. The old
readers below are used only when a historical payload is explicitly supplied.
The static encoder-zero corrections in ``zero_angles`` are already baked into
the corrected URDF joint origins. This bridge therefore publishes only the
dynamic ``angle_rad`` values and never adds the static offsets a second time.
Schema-v5 trajectories are fitted against timestamp-synchronised hardware
feedback, not controller set-points. They must therefore be queried with the
SDK ``hand_state`` topic. The retained schema-v4 path is command-indexed for
backwards compatibility only.
"""
from __future__ import annotations
import json
import math
from pathlib import Path
from typing import Any, Mapping, Sequence
import numpy as np
import rclpy
from rclpy.node import Node
from sensor_msgs.msg import JointState
G20_COMMAND_NAMES: tuple[str, ...] = (
"thumb_cmc_pitch",
"index_mcp_pitch",
"middle_mcp_pitch",
"ring_mcp_pitch",
"pinky_mcp_pitch",
"thumb_cmc_roll",
"index_mcp_roll",
"middle_mcp_roll",
"ring_mcp_roll",
"pinky_mcp_roll",
"thumb_cmc_yaw",
"reserved_11",
"reserved_12",
"reserved_13",
"reserved_14",
"thumb_mcp",
"index_pip",
"middle_pip",
"ring_pip",
"pinky_pip",
)
# Match the stable ordering used by the existing MuJoCo bridge. JointState
# consumers must use names, but retaining the ordering also keeps logs and
# direct comparisons deterministic.
G20_URDF_JOINT_NAMES: tuple[str, ...] = (
"index_dip",
"index_mcp_pitch",
"index_mcp_roll",
"index_pip",
"middle_dip",
"middle_mcp_pitch",
"middle_mcp_roll",
"middle_pip",
"pinky_dip",
"pinky_mcp_pitch",
"pinky_mcp_roll",
"pinky_pip",
"ring_dip",
"ring_mcp_pitch",
"ring_mcp_roll",
"ring_pip",
"thumb_cmc_pitch",
"thumb_cmc_roll",
"thumb_cmc_yaw",
"thumb_ip",
"thumb_mcp",
)
class CalibratedCommandMapper:
"""Validated, profile-specific lookup from SDK u8 values to URDF radians."""
def __init__(
self, payload: Mapping[str, Any], *, expected_side: str | None = None
) -> None:
from .full_hand import get_hand_calibration_profile, infer_compact_payload_layout, validate_compact_payload
from .compat.legacy_diagnostic_tools.models import get_default_registry, validate_schema_v6_runtime_payload
from .compat.legacy_diagnostic_tools.models.o12.artifacts import validate_o12_runtime_payload
from .core import ProfileKey
schema_version = int(payload["schema_version"])
if schema_version == 7:
validate_o12_runtime_payload(payload)
elif schema_version == 6:
validate_schema_v6_runtime_payload(payload)
else:
validate_compact_payload(payload)
side = str(payload["side"]).lower()
if expected_side is not None and side != str(expected_side).lower():
raise ValueError(
f"calibration side {side!r} does not match requested side "
f"{str(expected_side).lower()!r}"
)
quality = payload["quality"]
if quality.get("passed") is not True:
raise ValueError("calibration quality.passed must be true")
layout_id = (
str(payload["layout_id"])
if schema_version in {6, 7}
else infer_compact_payload_layout(payload)
)
self.side = side
self.layout_id = layout_id
self.model = str(payload["model"]).upper()
self.profile_id = str(
payload.get("profile_id", f"G20/{side}/{layout_id}/v1")
)
self.serial_number = str(payload["serial_number"])
self.input_domain = str(
payload.get(
"curve_input_domain",
"command_u8" if schema_version == 4 else "",
)
)
if self.input_domain not in {"command_u8", "feedback_u8", "feedback_rad"}:
raise ValueError("calibration curve_input_domain is invalid")
if schema_version in {6, 7}:
self.command_names = tuple(str(value) for value in payload["command_names"])
self.urdf_joint_names = tuple(str(name) for name in payload["joints"])
self._motor_by_joint = {
name: int(payload["joints"][name]["motor_index"])
for name in self.urdf_joint_names
}
registered = get_default_registry().get(
ProfileKey.parse(self.profile_id)
)
self.feedback_name_aliases = dict(
registered.profile.command.feedback_name_aliases
)
self.feedback_by_index = bool(
registered.profile.command.feedback_by_index
)
else:
profile = get_hand_calibration_profile(side, layout_id)
self.command_names = G20_COMMAND_NAMES
self.urdf_joint_names = G20_URDF_JOINT_NAMES
self._motor_by_joint = {
name: int(profile.joint_specs[name].motor_index)
for name in self.urdf_joint_names
}
self.feedback_name_aliases = {}
self.feedback_by_index = False
self._curves = {
name: tuple(
float(value)
for value in payload["joints"][name]["angle_rad"]
)
for name in self.urdf_joint_names
}
self._decreasing_curves = {
name: tuple(
float(value)
for value in payload["joints"][name].get(
"decreasing_rad", payload["joints"][name]["angle_rad"]
)
)
for name in self.urdf_joint_names
}
self._increasing_curves = {
name: tuple(
float(value)
for value in payload["joints"][name].get(
"increasing_rad", payload["joints"][name]["angle_rad"]
)
)
for name in self.urdf_joint_names
}
self._previous_by_motor: dict[int, float] = {}
self._direction_by_motor: dict[int, str] = {}
self.direction_deadband_u8 = 0.002 if schema_version == 7 else 0.5
self._knots = {
name: tuple(float(value) for value in payload["joints"][name].get(
"curve_input_knots_rad", ()
))
for name in self.urdf_joint_names
}
self._raw_increasing_branch = {
name: str(payload["joints"][name].get(
"raw_increasing_curve_branch", "increasing"
))
for name in self.urdf_joint_names
}
@staticmethod
def _command_index(value: float) -> int:
command = float(value)
if not math.isfinite(command):
raise ValueError("calibrated command positions must be finite")
return max(0, min(255, int(math.floor(command + 0.5))))
def map_positions(
self, positions: Sequence[float], names: Sequence[str] = ()
) -> tuple[float, ...]:
values = tuple(float(value) for value in positions)
if names and not self.feedback_by_index:
if len(names) != len(values):
raise ValueError(
"JointState names and positions must have equal length"
)
if len(set(names)) != len(names):
raise ValueError("JointState names must be unique")
by_name = dict(zip((str(name) for name in names), values))
for alias, canonical in self.feedback_name_aliases.items():
if alias in by_name and canonical not in by_name:
by_name[canonical] = by_name[alias]
missing = [name for name in self.command_names if name not in by_name]
if missing:
raise ValueError(
f"{self.model} feedback is missing named channels: "
+ ",".join(missing)
)
command = tuple(by_name[name] for name in self.command_names)
else:
if len(values) != len(self.command_names):
raise ValueError(
f"unnamed {self.model} feedback must contain exactly "
f"{len(self.command_names)} positions"
)
command = values
indices = (
() if self.input_domain == "feedback_rad"
else tuple(self._command_index(value) for value in command)
)
direction_by_motor: dict[int, str | None] = {}
for motor, value in enumerate(command):
previous = self._previous_by_motor.get(motor)
direction = self._direction_by_motor.get(motor)
if previous is not None:
if value > previous + self.direction_deadband_u8:
direction = "increasing"
elif value < previous - self.direction_deadband_u8:
direction = "decreasing"
direction_by_motor[motor] = direction
result: list[float] = []
for name in self.urdf_joint_names:
motor = self._motor_by_joint[name]
direction = direction_by_motor[motor]
if self.input_domain == "feedback_rad" and direction is not None:
raw_increasing = self._raw_increasing_branch[name]
direction = (
raw_increasing
if direction == "increasing"
else "increasing" if raw_increasing == "decreasing" else "decreasing"
)
curves = (
self._increasing_curves
if direction == "increasing"
else self._decreasing_curves
if direction == "decreasing"
else self._curves
)
if self.input_domain == "feedback_rad":
result.append(float(np.interp(command[motor], self._knots[name], curves[name])))
else:
result.append(curves[name][indices[motor]])
for motor, value in enumerate(command):
self._previous_by_motor[motor] = value
direction = direction_by_motor[motor]
if direction is not None:
self._direction_by_motor[motor] = direction
return tuple(result)
def load_calibrated_command_mapper(
calibration_file: str | Path, *, expected_side: str | None = None, input_kind="command"
) -> CalibratedCommandMapper:
path = Path(calibration_file).expanduser().resolve()
if not path.is_file():
raise ValueError(f"calibration JSON does not exist: {path}")
payload = json.loads(path.read_text(encoding="utf-8"))
if path.name == "release_manifest.json" or payload.get("format") in {"unified_calibration_v1", "unified_calibration_v2", "unified_calibration_v3"}:
from .runtime.artifacts.reader import load_unified_mapper
return load_unified_mapper(path, expected_side=expected_side, input_kind=input_kind)
return CalibratedCommandMapper(payload, expected_side=expected_side)
def default_input_topic(
hand_type: str, input_domain: str, model: str = "G20"
) -> str:
side = str(hand_type).lower()
if side not in {"left", "right"}:
raise ValueError("hand_type must be left or right")
if input_domain == "feedback_u8":
return f"/{str(model).lower()}/cb_{side}_hand_state"
if input_domain == "command_u8":
return f"/{str(model).lower()}/cb_{side}_hand_control_cmd"
if input_domain == "feedback_rad":
return f"/{str(model).lower()}/{side}/joint_states"
if input_domain == "command_rad":
return f"/{str(model).lower()}/{side}/joint_cmd"
raise ValueError("calibration curve_input_domain is invalid")
class CalibratedJointStateBridge(Node):
def __init__(self) -> None:
super().__init__("calibrated_joint_state_bridge")
self.declare_parameter("hand_type", "right")
self.declare_parameter("calibration_file", "")
self.declare_parameter("input_topic", "")
self.declare_parameter("output_topic", "")
self.declare_parameter("input_kind", "command")
hand_type = str(self.get_parameter("hand_type").value).lower()
if hand_type not in {"left", "right"}:
raise ValueError("hand_type must be left or right")
calibration_file = str(self.get_parameter("calibration_file").value)
if not calibration_file:
raise ValueError("calibration_file is required")
self.mapper = load_calibrated_command_mapper(
calibration_file, expected_side=hand_type, input_kind=str(self.get_parameter("input_kind").value)
)
input_topic = str(self.get_parameter("input_topic").value).strip()
output_topic = str(self.get_parameter("output_topic").value).strip()
self.input_topic = input_topic or default_input_topic(
hand_type, self.mapper.input_domain, self.mapper.model
)
self.output_topic = (
output_topic
or f"/sim/mujoco/{self.mapper.model.lower()}/{hand_type}/joint_state"
)
if self.output_topic in {self.input_topic, default_input_topic(hand_type,
"feedback_rad" if self.mapper.input_domain.endswith("rad") else "feedback_u8", self.mapper.model)}:
raise ValueError("mapped output must not overwrite the real SDK command/feedback topic")
self.publisher = self.create_publisher(JointState, self.output_topic, 10)
self.subscription = self.create_subscription(
JointState, self.input_topic, self._command_callback, 10
)
self._last_error = ""
self.get_logger().info(
f"loaded {self.mapper.profile_id} calibration for "
f"{self.mapper.serial_number}: "
f"{self.input_topic} ({self.mapper.input_domain}) -> "
f"{self.output_topic}"
)
def _command_callback(self, command: JointState) -> None:
try:
positions = self.mapper.map_positions(command.position, command.name)
except ValueError as error:
message = str(error)
if message != self._last_error:
self.get_logger().error(message)
self._last_error = message
return
self._last_error = ""
result = JointState()
result.header = command.header
result.name = list(self.mapper.urdf_joint_names)
result.position = list(positions)
self.publisher.publish(result)
def main(args: Sequence[str] | None = None) -> None:
rclpy.init(args=args)
node: CalibratedJointStateBridge | None = None
try:
node = CalibratedJointStateBridge()
rclpy.spin(node)
except KeyboardInterrupt:
pass
finally:
if node is not None:
node.destroy_node()
if rclpy.ok():
rclpy.shutdown()
if __name__ == "__main__":
main()
@@ -0,0 +1,182 @@
"""Map camera frame clocks to host time before pairing images with feedback.
The MVS frame counter timestamps exposure start. USB delivery/publication is
later. Clock latches bracket the device/host correspondence independently of
hand motion, images, or fitted calibration curves.
"""
from dataclasses import asdict, dataclass
import json
import math
from pathlib import Path
import time
CAMERA_TIMING_POLICY = "mvs_latched_exposure_time_v1"
@dataclass(frozen=True)
class ClockLatch:
device_tick: int
before_ns: int
after_ns: int
@property
def midpoint_ns(self):
return (self.before_ns+self.after_ns)//2
@property
def half_width_ns(self):
return (self.after_ns-self.before_ns)/2
class DeviceFrameClock:
"""A nominal device clock with short, measured host-time brackets.
A new latch is checked against the previous one before it can change the
mapping. This rejects wrong counter units, resets and host clock jumps.
Periodic updates bound accumulated drift; none depend on image delivery.
"""
def __init__(self, ticks_per_second):
if not math.isfinite(ticks_per_second) or ticks_per_second <= 0:
raise ValueError("invalid camera clock frequency")
self.ticks_per_second = ticks_per_second
self.anchor = None
self.last_frame_tick = None
self.last_stamp_ns = None
def elapsed_ns(self, ticks):
return round(ticks*1_000_000_000/self.ticks_per_second)
def update(self, latches):
if not latches or any(x.device_tick <= 0 or x.before_ns <= 0
or not 0 <= x.after_ns-x.before_ns <= 2_000_000 for x in latches):
raise ValueError("camera clock latch is invalid or too uncertain")
previous = self.anchor or latches[0]
if self.anchor is None and latches[-1].before_ns-latches[0].after_ns < 50_000_000:
raise ValueError("camera clock units need an independent elapsed-time check")
for current in latches:
elapsed = self.elapsed_ns(current.device_tick-previous.device_tick)
host_elapsed = current.midpoint_ns-previous.midpoint_ns
tolerance = current.half_width_ns+previous.half_width_ns+1_000_000
if elapsed < 0 or abs(elapsed-host_elapsed) > tolerance:
raise ValueError("camera/host clock discontinuity or counter unit mismatch")
self.anchor = min(latches, key=lambda x: x.half_width_ns)
def timestamp(self, tick, *, received_ns, exposure_us):
if (self.anchor is None or not 0 <= received_ns-self.anchor.midpoint_ns <= 2_000_000_000
or not math.isfinite(exposure_us) or exposure_us < 0):
raise ValueError("camera frame has no fresh clock or valid exposure")
if self.last_frame_tick is not None and tick <= self.last_frame_tick:
raise ValueError("camera frame counter reset or duplicate frame")
stamp = self.anchor.midpoint_ns+self.elapsed_ns(tick-self.anchor.device_tick)+round(exposure_us*500)
if stamp > received_ns+2_000_000 or (self.last_stamp_ns is not None and stamp <= self.last_stamp_ns):
raise ValueError("camera image time is future or out of order")
self.last_frame_tick, self.last_stamp_ns = tick, stamp
return stamp
class MvsCameraTiming:
"""The camera-specific clock latch and its optional append-only journal."""
def __init__(self, camera, mvs, clock_ns, *, exposure_us, journal_path=""):
self.camera, self.mvs, self.clock_ns = camera, mvs, clock_ns
self.exposure_us = exposure_us
self.frame_clock = None
self.last_stamp_ns = None
self.next_refresh = 0.
self.journal = None
if journal_path:
path = Path(journal_path)
path.parent.mkdir(parents=True, exist_ok=True)
self.journal = path.open("a", encoding="utf-8")
def _integer(self, name):
value = self.mvs.MVCC_INTVALUE_EX()
status = self.camera.MV_CC_GetIntValueEx(name, value)
if status != 0:
raise ValueError(f"camera clock cannot read {name}:0x{status:08x}")
return int(value.nCurValue)
def _latch(self):
before = self.clock_ns()
status = self.camera.MV_CC_SetCommandValue("DeviceTimestampLatch")
if status != 0:
raise ValueError(f"camera clock latch failed:0x{status:08x}")
tick = self._integer("DeviceTimestamp")
return ClockLatch(tick, before, self.clock_ns())
def _write(self, row):
if self.journal is not None:
self.journal.write(json.dumps(row, separators=(",", ":"))+"\n")
def start(self):
# The installed USB camera firmware returns ticks/second for this
# feature (despite its XML unit label). Verify that interpretation
# against elapsed host time; do not assume it for another firmware.
self.frame_clock = DeviceFrameClock(self._integer("DeviceTimestampIncrement"))
latches = []
for index in range(40):
if index:
time.sleep(.02)
latches.append(self._latch())
valid = [x for x in latches if 0 <= x.after_ns-x.before_ns <= 2_000_000]
if len(valid) >= 8 and valid[-1].before_ns-valid[0].after_ns >= 50_000_000:
self._update(latches)
return
raise ValueError("camera clock has no bounded-latency startup latches")
def _update(self, latches):
valid = [x for x in latches if 0 <= x.after_ns-x.before_ns <= 2_000_000]
if not valid:
# A delayed USB control transaction provides no new clock evidence.
# Keep the original two-second freshness bound while trying again.
self._write(dict(kind="camera_clock_latch_rejected", latches=[asdict(x) for x in latches]))
self.next_refresh = time.monotonic()+.02
return
self.frame_clock.update(valid)
self._write(dict(kind="camera_clock_sync", policy=CAMERA_TIMING_POLICY,
ticks_per_second=self.frame_clock.ticks_per_second,
latches=[asdict(x) for x in latches], anchor=asdict(self.frame_clock.anchor)))
if self.journal is not None:
self.journal.flush()
self.next_refresh = time.monotonic()+.5
def refresh_if_due(self):
if self.frame_clock is None:
self.start()
elif time.monotonic() >= self.next_refresh:
self._update([self._latch() for _ in range(3)])
def invalidate(self, reason):
"""Stop using an uncertain mapping until a new startup check succeeds."""
self.frame_clock = None
self.next_refresh = 0.
self._write(dict(kind="camera_clock_invalidated", reason=str(reason),
last_stamp_ns=self.last_stamp_ns))
if self.journal is not None:
self.journal.flush()
def timestamp(self, info, received_ns):
if self.frame_clock is None:
raise ValueError("camera frame has no synchronized clock")
tick = (int(info.nDevTimeStampHigh)<<32)|int(info.nDevTimeStampLow)
# Auto-exposure preview uses the exact exposure-start timestamp when
# the SDK supplies no per-frame duration. Calibration fixes exposure.
measured_exposure = float(info.fExposureTime)
exposure = measured_exposure if measured_exposure > 0 else self.exposure_us
stamp = self.frame_clock.timestamp(tick, received_ns=received_ns, exposure_us=exposure)
if self.last_stamp_ns is not None and stamp <= self.last_stamp_ns:
raise ValueError("camera exposure time is not increasing after clock synchronization")
self.last_stamp_ns = stamp
self._write(dict(kind="camera_frame_time", frame_number=int(info.nFrameNum),
device_tick=tick, sdk_host_stamp_ms=int(info.nHostTimeStamp),
received_ns=received_ns, stamp_ns=stamp, exposure_us=exposure,
timestamp_reference="exposure_midpoint" if exposure else "exposure_start"))
return stamp
def close(self):
if self.journal is not None:
self.journal.close()
self.journal = None
@@ -0,0 +1,21 @@
"""Compatibility adapters for one-release calibration migrations."""
from .config_v1 import (
legacy_default_profile_key,
product_profile_key,
resolve_legacy_profile_alias,
)
from .defaults import (
default_product_config_path,
default_three_camera_config_path,
)
from .paths import resolve_renamed_package_path
__all__ = [
"default_product_config_path",
"default_three_camera_config_path",
"legacy_default_profile_key",
"product_profile_key",
"resolve_legacy_profile_alias",
"resolve_renamed_package_path",
]
@@ -0,0 +1,47 @@
"""Identity migration for deployed product configuration schemas."""
from __future__ import annotations
from typing import Any, Mapping
from ..core import ProfileKey
def product_profile_key(raw: Mapping[str, Any]) -> ProfileKey:
version = int(raw.get("schema_version", -1))
if version in {2, 3}:
key = ProfileKey.parse(str(raw.get("profile_id", "")))
for field, actual in (
("model", key.model),
("side", key.side),
("tag_layout", key.layout),
):
configured = str(raw.get(field, "")).strip()
if configured and configured.lower() != actual.lower():
raise ValueError(f"{field} differs from profile_id")
return key
if version != 1:
raise ValueError("product config schema_version must be 1, 2 or 3")
model = str(raw.get("model", "")).strip().upper()
side = str(raw.get("side", "")).strip().lower()
layout = str(raw.get("tag_layout", "")).strip().lower()
if not layout and (model, side) == ("G20", "right"):
layout = "g20_right_19"
return ProfileKey(model, side, layout, 1)
def legacy_default_profile_key() -> ProfileKey:
"""Preserve the former no-argument executable for one release."""
return ProfileKey("G20", "right", "g20_right_19", 1)
def resolve_legacy_profile_alias(key: ProfileKey) -> ProfileKey:
"""Map retired layout identifiers to their reviewed physical profile."""
if (
key.model == "G20"
and key.side == "right"
and key.layout == "g20_right_15"
and key.revision == 1
):
return ProfileKey("G20", "right", "g20_right_19", 1)
return key
@@ -0,0 +1,16 @@
"""One-release default selection for invocations without ``--config``."""
from pathlib import Path
from ament_index_python.packages import get_package_share_directory
def default_product_config_path() -> Path:
share = Path(get_package_share_directory("linkerhand_calibration"))
return share / "config/g20_right_product.yaml"
def default_three_camera_config_path() -> Path:
"""Resolve the installed calibration defaults through the ROS index."""
share = Path(get_package_share_directory("linkerhand_calibration"))
return share / "config/three_camera_calibration.yaml"
@@ -0,0 +1,4 @@
"""Legacy single-camera algorithms retained for one compatibility release."""
from .session_v1 import uses_coupled_full_hand_zero_solver
__all__ = ["uses_coupled_full_hand_zero_solver"]
@@ -0,0 +1,24 @@
"""Version selection for replaying durable pre-v3 hardware sessions."""
from __future__ import annotations
from typing import Any, Mapping
def uses_coupled_full_hand_zero_solver(
session_start: Mapping[str, Any],
) -> bool:
"""Return the solver contract recorded by the legacy session header.
Capabilities are not consulted by the live runtime. This adapter reads
the durable v1 header only so offline replay can reproduce an artifact
created before the independent thumb solver was introduced.
"""
capabilities = {
str(value) for value in session_start.get("capabilities", ())
}
return (
int(session_start.get("sample_schema_version", 1)) == 1
and "palm_axis_side_channel_v2" in capabilities
and "palm_axis_relative_motion_v3" not in capabilities
)
@@ -0,0 +1,638 @@
"""Pure calibration math and command helpers.
This module deliberately has no ROS imports so the geometry, fitting, and
output schema can be tested without a camera or a connected hand.
"""
from __future__ import annotations
from dataclasses import dataclass, field
import math
from typing import Any, Iterable, Mapping, Sequence
import numpy as np
from scipy.spatial.transform import Rotation
COMMAND_NAMES: tuple[str, ...] = (
"thumb_cmc_pitch",
"index_mcp_pitch",
"middle_mcp_pitch",
"ring_mcp_pitch",
"pinky_mcp_pitch",
"thumb_cmc_roll",
"index_mcp_roll",
"middle_mcp_roll",
"ring_mcp_roll",
"pinky_mcp_roll",
"thumb_cmc_yaw",
"reserved_11",
"reserved_12",
"reserved_13",
"reserved_14",
"thumb_mcp",
"index_pip",
"middle_pip",
"ring_pip",
"pinky_pip",
)
BASELINE_COMMAND: tuple[int, ...] = (
255,
255,
255,
255,
255,
255,
193,
148,
105,
42,
245,
255,
255,
255,
255,
255,
255,
255,
255,
255,
)
PAIR_ROOT = "t0_t3"
PAIR_MCP = "t3_t4"
PAIR_IP = "t4_t5"
PAIR_NAMES: tuple[str, ...] = (PAIR_ROOT, PAIR_MCP, PAIR_IP)
DIRECTION_DECREASING = "decreasing"
DIRECTION_INCREASING = "increasing"
DIRECTIONS: tuple[str, ...] = (
DIRECTION_DECREASING,
DIRECTION_INCREASING,
)
PHASE_ROOT = "root"
PHASE_TIP = "tip"
JOINT_SPECS: dict[str, tuple[str, str, int]] = {
"thumb_cmc_pitch": (PHASE_ROOT, PAIR_ROOT, 0),
"thumb_mcp": (PHASE_TIP, PAIR_MCP, 15),
"thumb_ip": (PHASE_TIP, PAIR_IP, 15),
}
def build_command(
motor_index: int,
command_u8: int,
baseline: Sequence[int] = BASELINE_COMMAND,
) -> list[int]:
"""Return one full G20 command with exactly one replaced motor slot."""
if len(baseline) != 20:
raise ValueError("baseline must contain exactly 20 values")
values = [int(value) for value in baseline]
if any(value < 0 or value > 255 for value in values):
raise ValueError("baseline values must be in [0, 255]")
if motor_index not in (0, 5, 15):
raise ValueError(
"front thumb calibration only permits motor 0, 5 or 15"
)
command_u8 = int(command_u8)
if command_u8 < 0 or command_u8 > 255:
raise ValueError("command_u8 must be in [0, 255]")
values[motor_index] = command_u8
return values
def scan_targets(
repetitions: int = 3,
command_step: int = 1,
) -> list[tuple[int, str, int]]:
"""Build repeated 255->0->255 scan targets on a bounded command grid."""
if repetitions < 1:
raise ValueError("repetitions must be positive")
if command_step < 1 or command_step > 255:
raise ValueError("command_step must be in [1, 255]")
increasing = list(range(0, 256, command_step))
if increasing[-1] != 255:
increasing.append(255)
decreasing = list(reversed(increasing))
targets: list[tuple[int, str, int]] = []
for cycle in range(repetitions):
targets.extend(
(cycle, DIRECTION_DECREASING, command)
for command in decreasing
)
targets.extend(
(cycle, DIRECTION_INCREASING, command)
for command in increasing
)
return targets
def normalize_quaternion_xyzw(values: Sequence[float]) -> np.ndarray:
quaternion = np.asarray(values, dtype=float)
if quaternion.shape != (4,) or not np.all(np.isfinite(quaternion)):
raise ValueError("quaternion must contain four finite xyzw values")
norm = float(np.linalg.norm(quaternion))
if norm < 1e-12:
raise ValueError("quaternion norm is zero")
return quaternion / norm
def relative_quaternion_xyzw(
parent_camera_quaternion: Sequence[float],
child_camera_quaternion: Sequence[float],
) -> tuple[float, float, float, float]:
"""Compute parent->child orientation from two camera->tag rotations."""
parent = Rotation.from_quat(normalize_quaternion_xyzw(parent_camera_quaternion))
child = Rotation.from_quat(normalize_quaternion_xyzw(child_camera_quaternion))
quaternion = (parent.inv() * child).as_quat()
return tuple(float(value) for value in quaternion)
def image_plane_tag_quaternion_xyzw(
corners_xy: Sequence[Sequence[float]],
) -> tuple[float, float, float, float]:
"""Estimate tag orientation about the optical axis from ordered corners."""
corners = np.asarray(corners_xy, dtype=float)
if corners.shape != (4, 2) or not np.all(np.isfinite(corners)):
raise ValueError("corners_xy must contain four finite xy points")
# AprilTag corners 0->1 and 3->2 both follow the tag-local x axis.
# Average the two edges to reduce sub-pixel corner noise and perspective
# asymmetry. Image y points down, hence the minus sign for a right-handed
# camera-frame z rotation.
x_axis = (corners[1] - corners[0]) + (corners[2] - corners[3])
if float(np.linalg.norm(x_axis)) < 1e-9:
raise ValueError("tag x-axis is degenerate")
angle = -math.atan2(float(x_axis[1]), float(x_axis[0]))
quaternion = Rotation.from_rotvec([0.0, 0.0, angle]).as_quat()
return tuple(float(value) for value in quaternion)
def robust_rotation_summary(
quaternions_xyzw: Sequence[Sequence[float]],
) -> tuple[tuple[float, float, float, float], float]:
"""Return a robust orientation and maximum angular residual in radians."""
if not quaternions_xyzw:
raise ValueError("at least one quaternion is required")
rotations = Rotation.from_quat(
np.asarray(
[normalize_quaternion_xyzw(value) for value in quaternions_xyzw],
dtype=float,
)
)
reference = rotations[0]
delta_vectors = (reference.inv() * rotations).as_rotvec()
median_delta = np.median(delta_vectors, axis=0)
robust = reference * Rotation.from_rotvec(median_delta)
residuals = (robust.inv() * rotations).magnitude()
maximum = float(np.max(residuals)) if residuals.size else 0.0
return (
tuple(float(value) for value in robust.as_quat()),
maximum,
)
def rotation_spread_rad(
quaternions_xyzw: Sequence[Sequence[float]],
) -> float:
"""Return the maximum geodesic residual around a robust orientation."""
_, spread = robust_rotation_summary(quaternions_xyzw)
return spread
def rotation_rms_rad(
quaternions_xyzw: Sequence[Sequence[float]],
*,
outlier_threshold_rad: float | None = None,
) -> float:
"""Return RMS geodesic noise around a robust orientation."""
robust, _ = robust_rotation_summary(quaternions_xyzw)
reference = Rotation.from_quat(robust)
rotations = Rotation.from_quat(
np.asarray(
[normalize_quaternion_xyzw(value) for value in quaternions_xyzw],
dtype=float,
)
)
residuals = (reference.inv() * rotations).magnitude()
if outlier_threshold_rad is not None:
threshold = float(outlier_threshold_rad)
if threshold <= 0.0:
raise ValueError("outlier_threshold_rad must be positive")
residuals = residuals[residuals <= threshold]
if residuals.size == 0:
return float("inf")
return float(np.sqrt(np.mean(np.square(residuals))))
def rotation_inlier_fraction(
quaternions_xyzw: Sequence[Sequence[float]],
*,
outlier_threshold_rad: float,
) -> float:
"""Return the fraction close to the robust orientation."""
threshold = float(outlier_threshold_rad)
if threshold <= 0.0:
raise ValueError("outlier_threshold_rad must be positive")
robust, _ = robust_rotation_summary(quaternions_xyzw)
reference = Rotation.from_quat(robust)
rotations = Rotation.from_quat(
np.asarray(
[normalize_quaternion_xyzw(value) for value in quaternions_xyzw],
dtype=float,
)
)
residuals = (reference.inv() * rotations).magnitude()
return float(np.mean(residuals <= threshold))
def delta_rotation_vector(
reference_xyzw: Sequence[float],
observed_xyzw: Sequence[float],
) -> np.ndarray:
reference = Rotation.from_quat(normalize_quaternion_xyzw(reference_xyzw))
observed = Rotation.from_quat(normalize_quaternion_xyzw(observed_xyzw))
return (reference.inv() * observed).as_rotvec()
def fit_rotation_axis(
vectors: Sequence[Sequence[float]],
commands: Sequence[int],
) -> np.ndarray:
"""Fit and orient the single rotational axis used by one motor sweep."""
matrix = np.asarray(vectors, dtype=float)
command_values = np.asarray(commands, dtype=int)
if matrix.ndim != 2 or matrix.shape[1] != 3:
raise ValueError("vectors must have shape (N, 3)")
if command_values.shape != (matrix.shape[0],):
raise ValueError("commands must match vectors")
useful = np.linalg.norm(matrix, axis=1) > 1e-6
if int(np.count_nonzero(useful)) < 3:
raise ValueError("insufficient non-zero rotations to fit an axis")
_, _, vh = np.linalg.svd(matrix[useful], full_matrices=False)
axis = vh[0]
projections = matrix @ axis
low = projections[command_values <= 16]
high = projections[command_values >= 239]
if low.size and high.size and float(np.median(low)) < float(np.median(high)):
axis = -axis
return axis / np.linalg.norm(axis)
def isotonic_nonincreasing(values: Sequence[float]) -> np.ndarray:
"""Unweighted PAVA projection onto non-increasing values."""
original = np.asarray(values, dtype=float)
if original.ndim != 1 or not np.all(np.isfinite(original)):
raise ValueError("values must be a finite vector")
negated = -original
levels: list[float] = []
weights: list[int] = []
starts: list[int] = []
for index, value in enumerate(negated):
levels.append(float(value))
weights.append(1)
starts.append(index)
while len(levels) >= 2 and levels[-2] > levels[-1]:
total_weight = weights[-2] + weights[-1]
merged = (
levels[-2] * weights[-2] + levels[-1] * weights[-1]
) / total_weight
levels[-2:] = [merged]
weights[-2:] = [total_weight]
starts.pop()
projected = np.empty_like(original)
for block_index, (level, start) in enumerate(zip(levels, starts)):
end = starts[block_index + 1] if block_index + 1 < len(starts) else len(original)
projected[start:end] = -level
return projected
def _record_rotation(record: Mapping[str, Any], pair: str) -> tuple[float, ...]:
rotations = record.get("relative_quaternion_xyzw", {})
value = rotations.get(pair)
if value is None:
raise ValueError(f"sample record is missing {pair}")
return tuple(float(component) for component in value)
@dataclass(frozen=True)
class FitResult:
joints: dict[str, dict[str, Any]]
axes: dict[str, tuple[float, float, float]]
references: dict[str, tuple[float, float, float, float]]
ip_coupling: dict[str, float]
max_monotonic_correction_rad: float
max_hysteresis_rad: float
measurement_mode: str = "rotation"
trajectory_models: dict[str, Any] = field(default_factory=dict)
trajectory_quality: dict[str, Any] = field(default_factory=dict)
def measure_from_reference(
self,
joint_name: str,
observed_quaternion_xyzw: Sequence[float],
reference_quaternion_xyzw: Sequence[float] | None = None,
) -> float:
reference = (
reference_quaternion_xyzw
if reference_quaternion_xyzw is not None
else self.references[joint_name]
)
vector = delta_rotation_vector(reference, observed_quaternion_xyzw)
axis = np.asarray(self.axes[joint_name], dtype=float)
return float(vector @ axis)
def fit_calibration_curves(records: Iterable[Mapping[str, Any]]) -> FitResult:
"""Fit six complete 256-entry curves from dense or sparse scan records."""
samples = [
dict(record)
for record in records
if record.get("kind", "sample") == "sample"
]
if not samples:
raise ValueError("no scan records were provided")
joint_results: dict[str, dict[str, Any]] = {}
axes: dict[str, tuple[float, float, float]] = {}
references: dict[str, tuple[float, float, float, float]] = {}
maximum_correction = 0.0
maximum_hysteresis = 0.0
for joint_name, (phase, pair, motor_index) in JOINT_SPECS.items():
phase_records = [record for record in samples if record.get("phase") == phase]
if not phase_records:
raise ValueError(f"no records for phase {phase}")
cycle_references: dict[int, tuple[float, ...]] = {}
for record in phase_records:
if (
record.get("direction") == DIRECTION_DECREASING
and int(record.get("command_u8", -1)) == 255
):
cycle_references.setdefault(
int(record["cycle"]),
_record_rotation(record, pair),
)
cycles = sorted({int(record["cycle"]) for record in phase_records})
if any(cycle not in cycle_references for cycle in cycles):
raise ValueError(f"{joint_name} is missing a command-255 cycle reference")
vectors: list[np.ndarray] = []
commands: list[int] = []
indexed: list[tuple[Mapping[str, Any], np.ndarray]] = []
for record in phase_records:
cycle = int(record["cycle"])
vector = delta_rotation_vector(
cycle_references[cycle],
_record_rotation(record, pair),
)
vectors.append(vector)
commands.append(int(record["command_u8"]))
indexed.append((record, vector))
axis = fit_rotation_axis(vectors, commands)
axes[joint_name] = tuple(float(value) for value in axis)
references[joint_name] = robust_rotation_summary(
list(cycle_references.values())
)[0]
branch_values: dict[str, list[list[float]]] = {
direction: [[] for _ in range(256)] for direction in DIRECTIONS
}
for record, vector in indexed:
direction = str(record["direction"])
command = int(record["command_u8"])
branch_values[direction][command].append(float(vector @ axis))
fitted_branches: dict[str, list[float]] = {}
for direction in DIRECTIONS:
sample_commands = np.asarray(
[
command
for command, values in enumerate(branch_values[direction])
if values
],
dtype=int,
)
if (
sample_commands.size < 3
or int(sample_commands[0]) != 0
or int(sample_commands[-1]) != 255
):
raise ValueError(
f"{joint_name}.{direction} requires at least three samples "
"including commands 0 and 255"
)
raw = np.asarray(
[
float(np.median(branch_values[direction][command]))
for command in sample_commands
],
dtype=float,
)
raw -= raw[-1]
projected_samples = isotonic_nonincreasing(raw)
projected_samples -= projected_samples[-1]
correction = float(np.max(np.abs(projected_samples - raw)))
maximum_correction = max(maximum_correction, correction)
projected = np.interp(
np.arange(256, dtype=float),
sample_commands.astype(float),
projected_samples,
)
projected -= projected[255]
fitted_branches[direction] = [
round(float(value), 8) for value in projected
]
hysteresis = float(
np.max(
np.abs(
np.asarray(fitted_branches[DIRECTION_DECREASING])
- np.asarray(fitted_branches[DIRECTION_INCREASING])
)
)
)
maximum_hysteresis = max(maximum_hysteresis, hysteresis)
combined_curve = 0.5 * (
np.asarray(
fitted_branches[DIRECTION_DECREASING], dtype=float
)
+ np.asarray(
fitted_branches[DIRECTION_INCREASING], dtype=float
)
)
combined_curve -= combined_curve[255]
joint_result: dict[str, Any] = {
"motor_index": motor_index,
"angle_rad": [
round(float(value), 8) for value in combined_curve
],
"decreasing_rad": fitted_branches[DIRECTION_DECREASING],
"increasing_rad": fitted_branches[DIRECTION_INCREASING],
}
if joint_name == "thumb_ip":
joint_result["passive"] = True
joint_results[joint_name] = joint_result
mcp = joint_results["thumb_mcp"]
ip = joint_results["thumb_ip"]
x = np.asarray(mcp["angle_rad"], dtype=float)
y = np.asarray(ip["angle_rad"], dtype=float)
design = np.column_stack((x, np.ones_like(x)))
multiplier, offset = np.linalg.lstsq(design, y, rcond=None)[0]
predicted = multiplier * x + offset
residual_sum = float(np.sum((y - predicted) ** 2))
total_sum = float(np.sum((y - np.mean(y)) ** 2))
r_squared = 1.0 if total_sum < 1e-12 else 1.0 - residual_sum / total_sum
return FitResult(
joints=joint_results,
axes=axes,
references=references,
ip_coupling={
"multiplier": round(float(multiplier), 8),
"offset_rad": round(float(offset), 8),
"r_squared": round(float(r_squared), 8),
},
max_monotonic_correction_rad=maximum_correction,
max_hysteresis_rad=maximum_hysteresis,
)
def create_final_payload(
*,
serial_number: str,
fit: FitResult,
validation_errors_rad: Sequence[float],
passed: bool,
baseline: Sequence[int] = BASELINE_COMMAND,
) -> dict[str, Any]:
errors = np.abs(np.asarray(validation_errors_rad, dtype=float))
mae = float(np.mean(errors)) if errors.size else float("nan")
p95 = float(np.percentile(errors, 95)) if errors.size else float("nan")
runtime_joints: dict[str, dict[str, Any]] = {}
for joint_name, joint in fit.joints.items():
runtime_joint: dict[str, Any] = {
"motor_index": int(joint["motor_index"]),
"angle_rad": [
round(float(value), 8) for value in joint["angle_rad"]
],
}
if joint_name == "thumb_ip":
runtime_joint["passive"] = True
runtime_joints[joint_name] = runtime_joint
payload = {
"schema_version": 2,
"model": "G20",
"side": "left",
"serial_number": str(serial_number),
"angle_unit": "rad",
"command_range": [0, 255],
"zero_command_u8": 255,
"baseline_command_u8": [int(value) for value in baseline],
"joints": runtime_joints,
"ip_coupling": {
"multiplier": fit.ip_coupling["multiplier"],
"offset_rad": fit.ip_coupling["offset_rad"],
},
"quality": {
"passed": bool(passed),
"validation_mae_rad": None if not np.isfinite(mae) else round(mae, 8),
"validation_p95_rad": None if not np.isfinite(p95) else round(p95, 8),
},
}
validate_final_payload(payload)
return payload
def maximum_non_target_drift_rad(
records: Iterable[Mapping[str, Any]],
fit: FitResult,
) -> float:
"""Measure unintended active-joint motion during the two isolated scans."""
samples = [
dict(record)
for record in records
if record.get("kind", "sample") == "sample"
]
maximum = 0.0
checks = (
(PHASE_ROOT, "thumb_mcp", PAIR_MCP),
(PHASE_ROOT, "thumb_ip", PAIR_IP),
(PHASE_TIP, "thumb_cmc_pitch", PAIR_ROOT),
)
for phase, joint_name, pair in checks:
phase_records = [record for record in samples if record.get("phase") == phase]
for cycle in sorted({int(record["cycle"]) for record in phase_records}):
cycle_records = [
record for record in phase_records if int(record["cycle"]) == cycle
]
reference_record = next(
(
record
for record in cycle_records
if record.get("direction") == DIRECTION_DECREASING
and int(record.get("command_u8", -1)) == 255
),
None,
)
if reference_record is None:
continue
reference = _record_rotation(reference_record, pair)
axis = np.asarray(fit.axes[joint_name], dtype=float)
for record in cycle_records:
drift = abs(
float(
delta_rotation_vector(
reference,
_record_rotation(record, pair),
)
@ axis
)
)
maximum = max(maximum, drift)
return maximum
def validate_final_payload(payload: Mapping[str, Any]) -> None:
"""Validate the deliberately small runtime JSON schema."""
if payload.get("schema_version") != 2:
raise ValueError("schema_version must be 2")
if payload.get("model") != "G20" or payload.get("side") != "left":
raise ValueError("payload must describe a left G20")
if payload.get("angle_unit") != "rad":
raise ValueError("angle_unit must be rad")
baseline = payload.get("baseline_command_u8")
if not isinstance(baseline, list) or len(baseline) != 20:
raise ValueError("baseline_command_u8 must contain 20 values")
joints = payload.get("joints")
if not isinstance(joints, Mapping) or set(joints) != set(JOINT_SPECS):
raise ValueError("payload must contain exactly the three thumb joints")
for joint_name, joint in joints.items():
expected_motor = JOINT_SPECS[joint_name][2]
if int(joint.get("motor_index", -1)) != expected_motor:
raise ValueError(f"{joint_name} has the wrong motor index")
curve = joint.get("angle_rad")
if not isinstance(curve, list) or len(curve) != 256:
raise ValueError(
f"{joint_name}.angle_rad must contain 256 values"
)
values = np.asarray(curve, dtype=float)
if not np.all(np.isfinite(values)):
raise ValueError(
f"{joint_name}.angle_rad contains non-finite values"
)
if np.any(np.diff(values) > 1e-7):
raise ValueError(
f"{joint_name}.angle_rad must be non-increasing"
)
if abs(float(values[255])) > 1e-6:
raise ValueError(
f"{joint_name}.angle_rad[255] must be zero"
)
if joints["thumb_ip"].get("passive") is not True:
raise ValueError("thumb_ip must be marked passive")
@@ -0,0 +1,14 @@
# 历史兼容区
此目录保留旧格式读写、旧布局、历史数据诊断及必要的离线工具,不参与正式在线调度。
四型号的 runner/node/pipeline 已由 `runtime/runner.py``runtime/session.py`
`runtime/artifacts/finalization.py` 替代。只保留仍被历史工具调用的入口;
L6/O6 无调用的 pipeline 包装已删除。
旧独立断点实现、低速预检和在线节点已移除。旧发布函数已拒绝更新正式发布指针。
这里生成的离线诊断文件不能作为标准 URDF 已验收的证据;正式回放使用
`calibrate_hand --config <product.yaml> --offline-raw <raw_samples.jsonl>`
不要在此目录增加新型号。新型号提供 Profile、产品 YAML、原始 CAD/mesh;仅新 SDK 协议增加 Adapter。
需要恢复已删除的历史实现时,使用工作区
`calibration_output/refactor_backup.4WRWNn/` 中的归档,不要重新接入生产入口。
@@ -0,0 +1 @@
"""Read-only compatibility for archived sessions and diagnostic tools."""
@@ -0,0 +1,17 @@
"""Model- and side-specific calibration policies."""
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.registry import (
EngineBindings,
ProfileRegistry,
RegisteredProfile,
get_default_registry,
)
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.runtime_schema import validate_schema_v6_runtime_payload
__all__ = [
"EngineBindings",
"ProfileRegistry",
"RegisteredProfile",
"get_default_registry",
"validate_schema_v6_runtime_payload",
]
@@ -0,0 +1,15 @@
"""Registered profiles for this hand family."""
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.registry import ProfileRegistry
def register_profiles(registry: ProfileRegistry) -> None:
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.g20.legacy_11 import build_left_profile, build_right_profile
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.g20.right_19 import build_profile
registry.register(build_profile())
registry.register(build_left_profile())
registry.register(build_right_profile())
__all__ = ["register_profiles"]
@@ -0,0 +1,244 @@
"""Adapt reviewed family profiles to the shared typed contract."""
from __future__ import annotations
from linkerhand_calibration.core import (
CalibrationProfile,
CommandLayout,
MeasurementPolicy,
MeasurementSpec,
MotionPolicy,
ProfileKey,
QualityPolicy,
ScopePolicy,
TagSpec,
TaskSpec,
ViewSpec,
VisionRigSpec,
ZeroSolvePolicy,
)
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.g20.profile import MIMIC_DERIVED_FINGER_DIPS
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.registry import EngineBindings, RegisteredProfile
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.g20.artifacts import build_artifact_policy
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.g20.motion import (
build_calibration_motion_command,
build_calibration_preparation_waypoints,
build_calibration_return_waypoints,
)
_HARD_THRESHOLD_KEYS = frozenset(
{
"minimum_detection_rate",
"maximum_reprojection_error_px",
"maximum_axis_cycle_difference_rad",
"maximum_pose_line_rms_m",
"maximum_validation_error_rad",
}
)
def _run_cli(args: list[str] | None = None) -> None:
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.g20.runner import main
main(args)
def _run_node(args: list[str] | None = None) -> None:
from linkerhand_calibration.runtime.ros.entrypoint import main
main(args)
def _legacy_typed_profile(
*,
key: ProfileKey,
namespace: str,
hand_profile,
zero_profile,
mechanical_endpoint_joints: frozenset[str] = frozenset(),
post_solve_endpoint_joints: frozenset[str] = frozenset(),
) -> CalibrationProfile:
fixed_by_view = {
view: frozenset(roles)
for view, roles in hand_profile.preflight_view_roles.items()
}
views = tuple(
ViewSpec(
name=view,
tags=tuple(
TagSpec(
role=role,
tag_id=int(tag_id),
fixed_reference=role in fixed_by_view.get(view, frozenset()),
)
for role, tag_id in roles.items()
),
)
for view, roles in hand_profile.view_tags.items()
)
record_specs = hand_profile.record_specs
command_index_by_joint = {
name: int(spec.motor_index) for name, spec in record_specs.items()
}
command_index_by_joint.update(
{
name: int(spec.motor_index)
for name, spec in hand_profile.joint_specs.items()
}
)
measurements = {
name: MeasurementSpec(
joint=name,
kind=str(spec.zero_kind or "curve"),
view=spec.view,
parent_role=spec.parent_role,
child_role=spec.child_role,
validation_source=(hand_profile.axis_validation_sources or {}).get(
name
),
pose_axis_line_required=bool(
spec.pose_axis_line_required
),
)
for name, spec in record_specs.items()
}
tasks = tuple(
TaskSpec(
key=spec.key,
view=spec.view,
command_index=int(spec.motor_index),
joints=tuple(spec.joints),
auxiliary_commands=tuple(spec.auxiliary_commands),
validation_only=bool(spec.validation_only),
)
for spec in hand_profile.sweep_specs
)
active = frozenset(hand_profile.active_joints)
passive = frozenset(hand_profile.passive_joints)
thumb = frozenset(name for name in active if name.startswith("thumb_"))
fingers = active - thumb
typed = CalibrationProfile(
key=key,
namespace=namespace,
command=CommandLayout(
names=tuple(hand_profile.command_names),
baseline_u8=tuple(int(value) for value in hand_profile.baseline_command),
command_index_by_joint=command_index_by_joint,
disabled_indices=frozenset(
index
for index, name in enumerate(hand_profile.command_names)
if name.startswith("reserved_")
),
),
vision=VisionRigSpec(
views=views,
common_frame="calibration_common",
extrinsic_reference_view=views[0].name,
extrinsics_quality_limits={
"reprojection_rms_px": 1.2,
"maximum_rotation_repeatability_deg": 0.3,
"maximum_translation_repeatability_m": 0.0015,
},
minimum_capture_counts={
"front_side_captures": 15,
"front_top_captures": 15,
},
),
motion=MotionPolicy(
tasks=tasks,
precheck_sweeps=bool(hand_profile.precheck_sweeps),
steady_command_checkpoints=bool(
hand_profile.steady_command_checkpoints
),
),
measurement=MeasurementPolicy(
measurements=measurements,
cross_view_sources=dict(
hand_profile.axis_validation_sources or {}
),
image_curve_joints=frozenset(
hand_profile.image_trajectory_joints
),
directional_zero=bool(hand_profile.directional_zero),
cross_view_roll_curve=bool(hand_profile.cross_view_roll_curve),
stable_cross_view_cone_bias=bool(
hand_profile.stable_cross_view_cone_bias
),
),
zero=ZeroSolvePolicy(
active_joints=active,
passive_joints=passive,
direct_zero_joints=tuple(zero_profile.direct_zero_joints),
axis_joints=tuple(zero_profile.axis_joints),
mechanical_endpoint_joints=mechanical_endpoint_joints,
post_solve_endpoint_joints=post_solve_endpoint_joints,
mimic_source_by_joint={
target: source
for target, source in MIMIC_DERIVED_FINGER_DIPS.items()
if target in passive and source in active
},
cad_frozen_joints=frozenset(
passive - set(zero_profile.static_output_zero_offsets_rad)
),
),
quality=QualityPolicy(
training_cycles=(0, 1, 2),
holdout_cycle=3 if key.layout != "legacy_11" else None,
hard_threshold_keys=_HARD_THRESHOLD_KEYS,
isolated_holdout=bool(hand_profile.isolated_holdout),
),
scope=ScopePolicy(
calibrate_joints={
"full": active,
"thumb": thumb,
"fingers": fingers,
},
frozen_joints={
"full": frozenset(),
"thumb": fingers,
"fingers": thumb,
},
),
artifacts=build_artifact_policy(
frozenset(hand_profile.capabilities)
),
urdf_authorized_fields={
name: frozenset({"origin.rpy", "limit.upper"})
for name in active
} | {
name: frozenset({"mimic.offset"})
for name in passive
if name in zero_profile.static_output_zero_offsets_rad
},
)
return typed
def adapt_profile(
*, key: ProfileKey, namespace: str, hand_profile, zero_profile,
mechanical_endpoint_joints: frozenset[str] = frozenset(),
post_solve_endpoint_joints: frozenset[str] = frozenset(),
typed_profile: CalibrationProfile | None = None,
) -> RegisteredProfile:
# Current products supply their authoritative YAML; only the archived
# 11-Tag compatibility contract is still compiled from its old definition.
typed = typed_profile if typed_profile is not None else _legacy_typed_profile(
key=key, namespace=namespace, hand_profile=hand_profile,
zero_profile=zero_profile, mechanical_endpoint_joints=mechanical_endpoint_joints,
post_solve_endpoint_joints=post_solve_endpoint_joints,
)
if typed.key != key or typed.namespace != namespace:
raise ValueError("declarative profile identity differs from product binding")
return RegisteredProfile(
profile=typed,
engine=EngineBindings(
hand_profile=hand_profile,
zero_profile=zero_profile,
motion_command=build_calibration_motion_command,
preparation_waypoints=build_calibration_preparation_waypoints,
return_waypoints=build_calibration_return_waypoints,
cli_main=_run_cli,
node_main=_run_node,
),
)
@@ -0,0 +1,31 @@
"""Runtime JSON and corrected-URDF naming policy."""
from linkerhand_calibration.core import ArtifactPolicy
def build_artifact_policy(
compatibility_tokens: frozenset[str],
) -> ArtifactPolicy:
return ArtifactPolicy(
output_schema_version=4,
calibration_filename="g20_{side}_{serial_number}_calibration.json",
corrected_urdf_filename=(
"linkerhand_g20_{side}_{serial_number}_zero_calibrated.urdf"
),
protected_input_fields=frozenset(
{
"source_urdf_sha256",
"camera_extrinsics_sha256",
"calibration_config_sha256",
"tag_config_sha256",
"profile_config_sha256",
}
),
session_compatibility_tokens=frozenset(compatibility_tokens),
publish_corrected_urdf=(
"urdf_zero_publication" in compatibility_tokens
),
)
__all__ = ["build_artifact_policy"]
@@ -0,0 +1,26 @@
"""Reviewed command-channel layouts for this hand family."""
G20_COMMAND_NAMES: tuple[str, ...] = (
"thumb_cmc_pitch",
"index_mcp_pitch",
"middle_mcp_pitch",
"ring_mcp_pitch",
"pinky_mcp_pitch",
"thumb_cmc_roll",
"index_mcp_roll",
"middle_mcp_roll",
"ring_mcp_roll",
"pinky_mcp_roll",
"thumb_cmc_yaw",
"reserved_11",
"reserved_12",
"reserved_13",
"reserved_14",
"thumb_mcp",
"index_pip",
"middle_pip",
"ring_pip",
"pinky_pip",
)
__all__ = ["G20_COMMAND_NAMES"]
@@ -0,0 +1,144 @@
"""Read-only regression checks for the reviewed hardware sessions."""
from __future__ import annotations
import argparse
import hashlib
import json
from pathlib import Path
from typing import Any
_PASS_SESSIONS = {
"20260830_181154": "full",
"20260831_141123": "thumb",
"20260831_163843": "thumb",
}
_FAIL_SESSIONS = {
"20260831_111837": ("FIT-MODEL-401", "joint_fit_check_failed"),
"20260831_142322": ("VAL-QUALITY-501", "zero_model_validation_failed"),
}
def _sha256_file(path: Path) -> str:
digest = hashlib.sha256()
with path.open("rb") as stream:
for chunk in iter(lambda: stream.read(1024 * 1024), b""):
digest.update(chunk)
return digest.hexdigest()
def _summary(session: Path) -> dict[str, Any]:
value = json.loads(
(session / "calibration_summary_zh.json").read_text(encoding="utf-8")
)
if not isinstance(value, dict):
raise ValueError(f"invalid golden summary: {session}")
return value
def validate_golden_sessions(
serial_root: str | Path,
*,
replay_full_session: bool = True,
) -> dict[str, Any]:
"""Prove PASS/FAIL, hashes, partial freezes, and failure non-publication."""
root = Path(serial_root).expanduser().resolve()
results: dict[str, Any] = {}
for session_id, scope in _PASS_SESSIONS.items():
session = root / session_id
summary = _summary(session)
if summary.get("result") != "PASS":
raise ValueError(f"golden PASS changed: {session_id}")
if summary.get("calibration_scope") != scope:
raise ValueError(f"golden scope changed: {session_id}")
artifacts = summary["artifacts"]
hashes = summary["hashes"]
calibration_json = session / artifacts["json"]
corrected_urdf = session / artifacts["urdf"]
actual_json_hash = _sha256_file(calibration_json)
actual_urdf_hash = _sha256_file(corrected_urdf)
if actual_json_hash != hashes["calibration_json_sha256"]:
raise ValueError(f"golden JSON hash changed: {session_id}")
if actual_urdf_hash != hashes["corrected_urdf_sha256"]:
raise ValueError(f"golden URDF hash changed: {session_id}")
if session_id == "20260831_141123" and len(
summary.get("preserved_certified_zero_joints", ())
) != 12:
raise ValueError("merged thumb no longer freezes all finger zeros")
if session_id == "20260831_163843":
payload = json.loads(calibration_json.read_text(encoding="utf-8"))
if payload.get("artifact_type") != (
"g20_right_standalone_thumb_calibration"
):
raise ValueError("standalone thumb artifact type changed")
results[session_id] = {
"result": "PASS",
"json_sha256": actual_json_hash,
"urdf_sha256": actual_urdf_hash,
}
for session_id, (error_code, reason) in _FAIL_SESSIONS.items():
session = root / session_id
summary = _summary(session)
if (
summary.get("result") != "FAIL"
or summary.get("error_code") != error_code
or summary.get("reason") != reason
):
raise ValueError(f"golden failure decision changed: {session_id}")
results[session_id] = {
"result": "FAIL",
"error_code": error_code,
"reason": reason,
}
failure_ids = set(_FAIL_SESSIONS)
for pointer_name in ("latest_passed", "latest_thumb_passed"):
pointer = root / pointer_name
if pointer.exists() and pointer.resolve().name in failure_ids:
raise ValueError(f"failure session was published through {pointer_name}")
if replay_full_session:
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.g20.offline_replay import replay_session
session_id = "20260830_181154"
replay = replay_session(root / session_id, write_outputs=False)
expected = results[session_id]
if replay["computed_final_json_sha256"] != expected["json_sha256"]:
raise ValueError(
"full-session replay JSON is not byte-identical: "
f"computed={replay['computed_final_json_sha256']} "
f"expected={expected['json_sha256']}"
)
if replay["corrected_urdf_sha256"] != expected["urdf_sha256"]:
raise ValueError(
"full-session replay URDF is not byte-identical: "
f"computed={replay['corrected_urdf_sha256']} "
f"expected={expected['urdf_sha256']}"
)
results[session_id]["offline_replay"] = "byte_identical"
return results
def main(args: list[str] | None = None) -> None:
parser = argparse.ArgumentParser(
description="Validate the five reviewed calibration sessions"
)
parser.add_argument("serial_root")
parser.add_argument("--no-replay", action="store_true")
selected = parser.parse_args(args)
print(
json.dumps(
validate_golden_sessions(
selected.serial_root,
replay_full_session=not selected.no_replay,
),
ensure_ascii=False,
indent=2,
sort_keys=True,
)
)
if __name__ == "__main__":
main()
@@ -0,0 +1,35 @@
"""One-release typed wrappers for the legacy 11-Tag layouts."""
from linkerhand_calibration.core import ProfileKey
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.g20.profile import get_hand_calibration_profile
from linkerhand_calibration.urdf_zero import get_zero_calibration_profile
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.registry import RegisteredProfile
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.g20._adapter import adapt_profile
LEFT_KEY = ProfileKey("G20", "left", "legacy_11", 1)
RIGHT_KEY = ProfileKey("G20", "right", "legacy_11", 1)
def build_left_profile() -> RegisteredProfile:
hand = get_hand_calibration_profile(LEFT_KEY.side, LEFT_KEY.layout)
return adapt_profile(
key=LEFT_KEY,
namespace="/g20_calibration",
hand_profile=hand,
zero_profile=get_zero_calibration_profile(
LEFT_KEY.side, LEFT_KEY.layout
),
)
def build_right_profile() -> RegisteredProfile:
hand = get_hand_calibration_profile(RIGHT_KEY.side, RIGHT_KEY.layout)
return adapt_profile(
key=RIGHT_KEY,
namespace="/g20_calibration",
hand_profile=hand,
zero_profile=get_zero_calibration_profile(
RIGHT_KEY.side, RIGHT_KEY.layout
),
)
@@ -0,0 +1,15 @@
"""Reviewed motion and safe-waypoint strategy exports."""
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.g20.profile import (
build_calibration_motion_command,
build_calibration_preparation_waypoints,
build_calibration_return_waypoints,
build_calibration_speed_profile,
)
__all__ = [
"build_calibration_motion_command",
"build_calibration_preparation_waypoints",
"build_calibration_return_waypoints",
"build_calibration_speed_profile",
]
@@ -0,0 +1,687 @@
"""Validate and atomically publish one inseparable G20 JSON/URDF session."""
from __future__ import annotations
import copy
import json
import math
import os
from pathlib import Path
import re
from typing import Any, Mapping, Sequence
import xml.etree.ElementTree as ET
import numpy as np
from scipy.spatial.transform import Rotation
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.g20.profile import (
G20_COMBINATION_REQUIRED_TARGET_KEYS,
G20_RIGHT_19_LAYOUT,
get_hand_calibration_profile,
validate_compact_payload,
)
from linkerhand_calibration.product import ProductConfig, sha256_file
from linkerhand_calibration.storage import atomic_write_json
from linkerhand_calibration.core.urdf import UrdfCorrectionPlan, build_correction_plan
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.g20.zero_solver import (
RIGHT_19_MECHANICAL_ENDPOINT_JOINTS,
get_zero_calibration_profile,
)
PASSIVE_JOINTS = frozenset(
{"thumb_ip", "index_dip", "middle_dip", "ring_dip", "pinky_dip"}
)
VISUALLY_MEASURED_FINGER_DIPS = frozenset(
{"index_dip", "middle_dip", "ring_dip", "pinky_dip"}
)
ACTIVE_ZERO_JOINTS = frozenset(
get_zero_calibration_profile("right", G20_RIGHT_19_LAYOUT).direct_zero_joints
)
RETAINED_ACTIVE_ZERO_JOINTS = frozenset(
get_hand_calibration_profile("right", G20_RIGHT_19_LAYOUT).active_joints
) - ACTIVE_ZERO_JOINTS
SCHEMA_V4_OFFSET_QUANTIZATION_TOLERANCE_RAD = 5.1e-9
def _load_json(path: Path) -> dict[str, Any]:
with path.open("r", encoding="utf-8") as stream:
payload = json.load(stream)
if not isinstance(payload, dict):
raise ValueError(f"{path} must contain a JSON object")
return payload
def _joint_blocks(text: str) -> dict[str, str]:
pattern = re.compile(
r"<joint\b[^>]*\bname\s*=\s*([\"'])(?P<name>[^\"']+)\1[^>]*>.*?</joint>",
re.DOTALL,
)
return {match.group("name"): match.group(0) for match in pattern.finditer(text)}
def _mask_origin_rpy(block: str) -> str:
return re.sub(
r"(<origin\b[^>]*\brpy\s*=\s*)([\"'])[^\"']*\2",
r"\1\2__CALIBRATED_RPY__\2",
block,
count=1,
)
def _mask_active_origin_rpy_fields(text: str) -> str:
pattern = re.compile(
r"<joint\b[^>]*\bname\s*=\s*([\"'])(?P<name>[^\"']+)\1[^>]*>.*?</joint>",
re.DOTALL,
)
def replace(match: re.Match[str]) -> str:
block = match.group(0)
if match.group("name") not in ACTIVE_ZERO_JOINTS:
return block
masked = _mask_origin_rpy(block)
if masked == block:
raise ValueError(
f"active joint {match.group('name')} has no origin.rpy"
)
return masked
return pattern.sub(replace, text)
def _mask_endpoint_coordinate_fields(text: str) -> str:
"""Mask only limit/mimic fields induced by endpoint zero coordinates."""
pattern = re.compile(
r"<joint\b[^>]*\bname\s*=\s*([\"'])(?P<name>[^\"']+)\1[^>]*>.*?</joint>",
re.DOTALL,
)
def replace(match: re.Match[str]) -> str:
block = match.group(0)
name = match.group("name")
if name in RIGHT_19_MECHANICAL_ENDPOINT_JOINTS:
return re.sub(
r"(<limit\b[^>]*\bupper\s*=\s*)([\"'])[^\"']*\2",
r"\1\2__CALIBRATED_UPPER__\2",
block,
count=1,
)
mimic = re.search(
r"<mimic\b[^>]*\bjoint\s*=\s*([\"'])(?P<source>[^\"']+)\1[^>]*>",
block,
re.DOTALL,
)
if (
mimic is not None
and mimic.group("source") in RIGHT_19_MECHANICAL_ENDPOINT_JOINTS
):
return re.sub(
r"(<mimic\b[^>]*\boffset\s*=\s*)([\"'])[^\"']*\2",
r"\1\2__CALIBRATED_MIMIC_OFFSET__\2",
block,
count=1,
)
return block
return pattern.sub(replace, text)
def _triplet(value: str) -> np.ndarray:
result = np.asarray([float(item) for item in value.split()], dtype=float)
if result.shape != (3,) or not np.all(np.isfinite(result)):
raise ValueError(f"invalid URDF triplet: {value}")
return result
def _joint_elements(path: str | Path) -> dict[str, ET.Element]:
return {
str(joint.get("name")): joint
for joint in ET.parse(Path(path)).getroot().findall("joint")
}
def _verify_expected_origin_offsets(
source: str | Path,
corrected: str | Path,
expected_offsets_rad: Mapping[str, float],
) -> None:
offsets = {str(name): float(value) for name, value in expected_offsets_rad.items()}
if set(offsets) != ACTIVE_ZERO_JOINTS or any(
not math.isfinite(value) for value in offsets.values()
):
raise ValueError("expected offsets must contain all finite active static-zero values")
before = _joint_elements(source)
after = _joint_elements(corrected)
maximum_rotation_error = 0.0
for name, original_joint in before.items():
corrected_joint = after[name]
original_origin = original_joint.find("origin")
corrected_origin = corrected_joint.find("origin")
if original_origin is None or corrected_origin is None:
if original_origin is not corrected_origin:
raise ValueError(f"corrected URDF changed origin presence for {name}")
continue
original_xyz = _triplet(original_origin.get("xyz", "0 0 0"))
corrected_xyz = _triplet(corrected_origin.get("xyz", "0 0 0"))
if not np.allclose(original_xyz, corrected_xyz, atol=1.0e-12, rtol=0.0):
raise ValueError(f"corrected URDF changed origin.xyz for {name}")
original_rotation = Rotation.from_euler(
"xyz", _triplet(original_origin.get("rpy", "0 0 0"))
)
corrected_rotation = Rotation.from_euler(
"xyz", _triplet(corrected_origin.get("rpy", "0 0 0"))
)
expected_rotation = original_rotation
if name in offsets:
axis_node = original_joint.find("axis")
axis = _triplet(
"1 0 0" if axis_node is None else axis_node.get("xyz", "1 0 0")
)
norm = float(np.linalg.norm(axis))
if norm <= 1.0e-12:
raise ValueError(f"source URDF joint {name} has a degenerate axis")
expected_rotation = expected_rotation * Rotation.from_rotvec(
axis / norm * offsets[name]
)
error = float((expected_rotation.inv() * corrected_rotation).magnitude())
maximum_rotation_error = max(maximum_rotation_error, error)
# Schema v4 intentionally stores zero offsets at eight decimal places.
# Older sessions wrote the corrected URDF from the full-precision solve,
# so comparing that file with the published JSON necessarily permits half
# of one last-place unit. This is about 2.9e-7 degrees and is far below
# any calibration or URDF numerical significance.
if maximum_rotation_error > SCHEMA_V4_OFFSET_QUANTIZATION_TOLERANCE_RAD:
raise ValueError(
"corrected URDF origin.rpy does not match the published zero offsets"
)
def verify_corrected_urdf(
source: str | Path,
corrected: str | Path,
*,
expected_offsets_rad: Mapping[str, float] | None = None,
endpoint_anchored_offsets_rad: Mapping[str, float] | None = None,
correction_plan: UrdfCorrectionPlan | None = None,
) -> tuple[str, ...]:
"""Prove that only active-joint origin.rpy attributes changed.
Passive joint blocks, including their mimic elements, are compared as raw
UTF-8 text so formatting and numeric spelling are protected as well.
"""
source_text = Path(source).read_text(encoding="utf-8")
corrected_text = Path(corrected).read_text(encoding="utf-8")
if correction_plan is not None:
correction_plan.verify_source(source)
if expected_offsets_rad is None:
raise ValueError(
"correction-plan validation requires expected offsets"
)
correction_plan.authorize_offsets(expected_offsets_rad)
before = _joint_blocks(source_text)
after = _joint_blocks(corrected_text)
if set(before) != set(after):
raise ValueError("corrected URDF changed the joint set")
# Text patchers may normalize only the final newline. It has no URDF/XML
# semantics; every joint block and every non-rpy byte remains guarded
# below.
masked_source = _mask_endpoint_coordinate_fields(
_mask_active_origin_rpy_fields(source_text)
).rstrip("\r\n")
masked_corrected = _mask_endpoint_coordinate_fields(
_mask_active_origin_rpy_fields(corrected_text)
).rstrip("\r\n")
if masked_source != masked_corrected:
raise ValueError(
"corrected URDF changed content outside active origin.rpy or "
"endpoint-calibrated fields"
)
changed: list[str] = []
for name in before:
if before[name] == after[name]:
continue
if name not in ACTIVE_ZERO_JOINTS and name not in PASSIVE_JOINTS:
raise ValueError(f"corrected URDF changed protected joint {name}")
if name in ACTIVE_ZERO_JOINTS:
changed.append(name)
if expected_offsets_rad is not None:
_verify_expected_origin_offsets(source, corrected, expected_offsets_rad)
endpoint_offsets = {
str(name): float(value)
for name, value in dict(endpoint_anchored_offsets_rad or {}).items()
}
if endpoint_offsets:
if correction_plan is not None and not set(endpoint_offsets).issubset(
correction_plan.endpoint_limit_joints
):
raise ValueError(
"endpoint offsets are not authorized by the correction plan"
)
source_joints = _joint_elements(source)
corrected_joints = _joint_elements(corrected)
for name, offset in endpoint_offsets.items():
original_limit = source_joints[name].find("limit")
corrected_limit = corrected_joints[name].find("limit")
expected_upper = float(original_limit.get("upper")) - offset
# Schema v4 stores the corresponding zero at eight decimal
# places, while the URDF is written from the full-precision solve.
# Match the half-last-place tolerance used for origin rotations.
if (
abs(float(corrected_limit.get("upper")) - expected_upper)
> SCHEMA_V4_OFFSET_QUANTIZATION_TOLERANCE_RAD
):
raise ValueError(f"corrected URDF has invalid {name} upper limit")
for name in PASSIVE_JOINTS:
original_mimic = source_joints[name].find("mimic")
corrected_mimic = corrected_joints[name].find("mimic")
if original_mimic is None or corrected_mimic is None:
continue
source_name = str(original_mimic.get("joint"))
multiplier = float(original_mimic.get("multiplier", "1"))
expected = float(original_mimic.get("offset", "0"))
expected += multiplier * endpoint_offsets.get(source_name, 0.0)
# The active endpoint offset comes from schema-v4 JSON rounded to
# eight decimal places, while the URDF mimic was written from the
# full-precision solve. Propagate exactly the same accepted
# quantization through the mimic multiplier; retain a much
# smaller allowance for XML decimal formatting itself.
tolerance = (
abs(multiplier)
* SCHEMA_V4_OFFSET_QUANTIZATION_TOLERANCE_RAD
+ 1.0e-12
)
if abs(float(corrected_mimic.get("offset", "0")) - expected) > tolerance:
raise ValueError(f"corrected URDF has invalid {name} mimic offset")
return tuple(sorted(changed))
def verify_urdf_mesh_resources(urdf: str | Path) -> dict[str, Path]:
"""Return every session-local mesh after proving it resolves safely."""
path = Path(urdf).expanduser().resolve()
root = ET.parse(path).getroot()
resources: dict[str, Path] = {}
for mesh in root.findall(".//mesh"):
filename = str(mesh.get("filename", "")).strip()
if not filename or "://" in filename or filename.startswith("package:"):
continue
relative = Path(filename)
if relative.is_absolute() or ".." in relative.parts:
raise ValueError(f"URDF has an unsafe local mesh path: {filename}")
resolved = (path.parent / relative).resolve()
try:
resolved.relative_to(path.parent)
except ValueError as error:
raise ValueError(f"URDF mesh escapes the session: {filename}") from error
if not resolved.is_file():
raise ValueError(f"URDF mesh resource is missing: {filename}")
resources[relative.as_posix()] = resolved
return dict(sorted(resources.items()))
def validate_runtime_curves_against_urdf_limits(
payload: Mapping[str, Any], runtime_urdf: str | Path
) -> None:
"""Reject a curve whose q leaves its runtime URDF coordinate limits.
Curves in schema v4 are expressed in the corrected URDF joint coordinate,
not in the source-CAD coordinate. Endpoint zero calibration can therefore
move a corrected coordinate limit while preserving the same physical CAD
endpoint; callers publishing a calibrated pair must pass that corrected
URDF here.
"""
validate_compact_payload(payload)
runtime_joints = _joint_elements(runtime_urdf)
for name, calibration in payload["joints"].items():
joint = runtime_joints.get(str(name))
if joint is None:
raise ValueError(f"runtime URDF is missing joint {name}")
limit = joint.find("limit")
if limit is None or limit.get("lower") is None or limit.get("upper") is None:
raise ValueError(f"runtime URDF joint {name} has no finite position limit")
lower = float(limit.get("lower"))
upper = float(limit.get("upper"))
if not math.isfinite(lower) or not math.isfinite(upper) or lower > upper:
raise ValueError(f"runtime URDF joint {name} has invalid position limits")
curve = np.asarray(calibration["angle_rad"], dtype=float)
minimum = float(np.min(curve))
maximum = float(np.max(curve))
tolerance = 1.0e-7
if minimum < lower - tolerance or maximum > upper + tolerance:
raise ValueError(
f"runtime curve exceeds runtime URDF limit for {name}: "
f"[{minimum:.9g}, {maximum:.9g}] not within "
f"[{lower:.9g}, {upper:.9g}]"
)
def clamp_compact_payload_to_urdf_limits(
payload: Mapping[str, Any], runtime_urdf: str | Path
) -> tuple[dict[str, Any], dict[str, int]]:
"""Return a schema-preserving payload bounded in its runtime coordinates."""
result = copy.deepcopy(dict(payload))
validate_compact_payload(result)
runtime_joints = _joint_elements(runtime_urdf)
clipped_by_joint: dict[str, int] = {}
for name, calibration in result["joints"].items():
joint = runtime_joints.get(str(name))
limit = None if joint is None else joint.find("limit")
if (
limit is None
or limit.get("lower") is None
or limit.get("upper") is None
):
raise ValueError(f"runtime URDF joint {name} has no finite position limit")
lower = float(limit.get("lower"))
upper = float(limit.get("upper"))
if not math.isfinite(lower) or not math.isfinite(upper) or lower > upper:
raise ValueError(f"runtime URDF joint {name} has invalid position limits")
source = np.asarray(calibration["angle_rad"], dtype=float)
bounded = np.clip(source, lower, upper)
count = int(np.count_nonzero(bounded != source))
if count:
clipped_by_joint[str(name)] = count
calibration["angle_rad"] = [
round(float(value), 8) for value in bounded
]
validate_compact_payload(result)
return result, clipped_by_joint
def build_mujoco_validation_commands(
baseline: Sequence[int],
) -> dict[str, Any]:
if len(baseline) != 20:
raise ValueError("baseline must contain exactly 20 commands")
base = [int(value) for value in baseline]
if any(not 0 <= value <= 255 for value in base):
raise ValueError("baseline commands must be in [0, 255]")
def pose(name: str, label: str, changes: Mapping[int, int]) -> dict[str, Any]:
values = list(base)
for index, value in changes.items():
values[int(index)] = int(value)
return {"name": name, "label_zh": label, "command_u8": values}
return {
"schema_version": 1,
"topic": "/g20/cb_right_hand_control_cmd",
"serial_number": "G20_RIGHT_001",
"poses": [
pose("all_open", "全开", {}),
pose("thumb_middle", "拇指中位", {0: 160, 5: 160, 10: 160, 15: 160}),
pose("index_middle", "食指中位", {1: 160, 6: 127, 16: 160}),
pose("middle_middle", "中指中位", {2: 160, 7: 127, 17: 160}),
pose("ring_middle", "无名指中位", {3: 160, 8: 127, 18: 160}),
pose("pinky_middle", "小指中位", {4: 160, 9: 127, 19: 160}),
pose(
"half_grip",
"四指半握",
{1: 160, 2: 160, 3: 160, 4: 160, 16: 160, 17: 160, 18: 160, 19: 160},
),
pose(
"light_pinch",
"轻捏",
{0: 176, 5: 176, 10: 176, 15: 176, 1: 176, 6: 127, 16: 176},
),
],
}
def atomic_session_pointer(root: str | Path, name: str, session: str | Path) -> Path:
if "passed" in name:
raise ValueError("release pointers require the common ArtifactPublisher")
parent = Path(root).resolve()
target = Path(session).resolve()
if target.parent != parent:
raise ValueError("session pointer target must be a direct child of the serial root")
if not target.is_dir():
raise ValueError(f"session directory does not exist: {target}")
if name not in {"latest_attempt", "latest_passed", "latest_thumb_passed"}:
raise ValueError("unsupported session pointer name")
parent.mkdir(parents=True, exist_ok=True)
destination = parent / name
temporary = parent / f".{name}.{os.getpid()}.tmp"
if temporary.is_symlink():
temporary.unlink()
elif temporary.exists():
raise ValueError(f"temporary pointer path is occupied: {temporary}")
os.symlink(target.name, temporary, target_is_directory=True)
os.replace(temporary, destination)
return destination
def active_offsets(payload: Mapping[str, Any]) -> dict[str, float]:
validate_compact_payload(payload)
result = {
name: float(payload["joints"][name]["zero_angles"]["urdf_zero_offset_rad"])
for name in ACTIVE_ZERO_JOINTS
}
if set(result) != ACTIVE_ZERO_JOINTS or any(
not math.isfinite(value) for value in result.values()
):
raise ValueError(
"payload does not contain all finite observable active zero offsets"
)
return result
def standalone_thumb_offsets(payload: Mapping[str, Any]) -> dict[str, float]:
"""Validate a standalone thumb artifact and expand CAD finger zeros."""
if (
payload.get("schema_version") != 1
or payload.get("artifact_type")
!= "g20_right_standalone_thumb_calibration"
or payload.get("side") != "right"
or payload.get("non_thumb_zero_policy") != "source_cad_unchanged"
or not bool(payload.get("quality", {}).get("passed"))
):
raise ValueError("standalone thumb JSON contract is invalid")
joints = payload.get("joints")
expected_joint_names = {
"thumb_cmc_roll",
"thumb_cmc_yaw",
"thumb_cmc_pitch",
"thumb_mcp",
"thumb_ip",
}
if not isinstance(joints, Mapping) or set(joints) != expected_joint_names:
raise ValueError("standalone thumb JSON has the wrong joint set")
thumb_names = expected_joint_names - {"thumb_ip"}
result = {name: 0.0 for name in ACTIVE_ZERO_JOINTS}
for name in thumb_names:
try:
value = float(joints[name]["zero_angles"]["urdf_zero_offset_rad"])
except (KeyError, TypeError, ValueError) as error:
raise ValueError(
f"standalone thumb JSON is missing zero for {name}"
) from error
curve = np.asarray(joints[name].get("angle_rad", ()), dtype=float)
if curve.shape != (256,) or not np.all(np.isfinite(curve)):
raise ValueError(f"standalone thumb JSON curve is invalid: {name}")
result[name] = value
passive_curve = np.asarray(
joints["thumb_ip"].get("angle_rad", ()), dtype=float
)
if (
passive_curve.shape != (256,)
or not np.all(np.isfinite(passive_curve))
or not bool(joints["thumb_ip"].get("passive"))
or any(not math.isfinite(value) for value in result.values())
):
raise ValueError("standalone thumb JSON passive curve is invalid")
return result
def verify_partial_scope_preserves_certified_zeros(
*,
scope: str,
source_session: str | Path,
serial_root: str | Path | None = None,
serial_number: str,
current_offsets: Mapping[str, float],
) -> None:
"""Reject a partial artifact that changes any certified non-target zero."""
selected = str(scope).strip().lower()
if selected not in {"thumb", "fingers"}:
return
source = Path(source_session).expanduser()
if not source.is_absolute():
if serial_root is None:
raise ValueError(
"relative partial base session requires its serial root"
)
source = Path(serial_root).expanduser() / source
source = source.resolve()
base_payload = _load_json(
source / f"g20_right_{serial_number}_calibration.json"
)
base_offsets = (
standalone_thumb_offsets(base_payload)
if base_payload.get("artifact_type")
== "g20_right_standalone_thumb_calibration"
else active_offsets(base_payload)
)
preserve_thumb = selected == "fingers"
changed = {
name: abs(float(current_offsets[name]) - float(base_offsets[name]))
for name in ACTIVE_ZERO_JOINTS
if name.startswith("thumb_") == preserve_thumb
and abs(float(current_offsets[name]) - float(base_offsets[name]))
> SCHEMA_V4_OFFSET_QUANTIZATION_TOLERANCE_RAD
}
if changed:
raise ValueError(
f"partial {selected} calibration changed certified non-target "
"zeros: "
+ ",".join(sorted(changed))
)
def compare_session_offsets(
first: Mapping[str, Any],
second: Mapping[str, Any],
*,
maximum_difference_rad: float,
) -> dict[str, float]:
first_standalone = (
first.get("artifact_type")
== "g20_right_standalone_thumb_calibration"
)
second_standalone = (
second.get("artifact_type")
== "g20_right_standalone_thumb_calibration"
)
if first_standalone != second_standalone:
raise ValueError("cannot compare standalone thumb and full-hand artifacts")
left = (
standalone_thumb_offsets(first)
if first_standalone
else active_offsets(first)
)
right = (
standalone_thumb_offsets(second)
if second_standalone
else active_offsets(second)
)
differences = {
name: abs(left[name] - right[name])
for name in sorted(left)
}
failed = {name: value for name, value in differences.items() if value > maximum_difference_rad}
if failed:
details = ", ".join(
f"{name}={math.degrees(value):.3f}deg" for name, value in failed.items()
)
raise ValueError(f"independent-session static-zero mismatch: {details}")
return differences
def session_artifact_paths(session: str | Path, serial_number: str) -> dict[str, Path]:
directory = Path(session).resolve()
urdfs = sorted(directory.glob(f"*zero_calibrated_{serial_number}_*.urdf"))
if len(urdfs) != 1:
raise ValueError("session must contain exactly one corrected URDF")
return {
"json": directory / f"g20_right_{serial_number}_calibration.json",
"urdf": urdfs[0],
"summary": directory / "calibration_summary_zh.json",
"commands": directory / "mujoco_validation_commands.json",
"raw": directory / "raw_samples.jsonl",
"log": directory / "calibration.log",
}
def find_compatible_prior_session(
config: ProductConfig,
current: str | Path,
current_payload: Mapping[str, Any],
) -> tuple[Path | None, dict[str, float]]:
current_path = Path(current).resolve()
for candidate in sorted(config.session_root.glob("20??????_??????"), reverse=True):
if candidate.resolve() == current_path:
continue
json_path = candidate / f"g20_right_{config.serial_number}_calibration.json"
summary_path = candidate / "calibration_summary_zh.json"
if not json_path.is_file() or not summary_path.is_file():
continue
try:
summary = _load_json(summary_path)
if not bool(summary.get("quality", {}).get("passed")):
continue
payload = _load_json(json_path)
differences = compare_session_offsets(
payload,
current_payload,
maximum_difference_rad=config.static_repeatability_rad,
)
except (OSError, ValueError, KeyError, TypeError, json.JSONDecodeError):
continue
return candidate, differences
return None, {}
def _verify_combination_validation(combination: Mapping[str, Any]) -> None:
# The formal product uses the independently held-out fourth sweep cycle.
# The optional eight-pose check remains a developer diagnostic because the
# axis-line zero solve does not establish an absolute Cartesian hand-base
# transform. When explicitly enabled, retain its strict coverage and
# error checks.
if not bool(combination.get("enabled")):
return
position_p95 = float(combination.get("position_p95_m", float("inf")))
orientation_p95 = float(
combination.get("orientation_p95_rad", float("inf"))
)
required = {str(value) for value in combination.get("required_targets", ())}
expected = set(G20_COMBINATION_REQUIRED_TARGET_KEYS)
observations = combination.get("observation_counts")
validations = combination.get("validation_counts")
if required != expected:
raise ValueError("combination validation has the wrong required-target set")
if not isinstance(observations, Mapping) or not isinstance(validations, Mapping):
raise ValueError("combination validation is missing per-target coverage")
missing_observations = sorted(
key for key in expected if int(observations.get(key, 0)) < 2
)
missing_validations = sorted(
key for key in expected if int(validations.get(key, 0)) < 1
)
if missing_observations or missing_validations:
raise ValueError(
"combination validation target coverage is incomplete: "
f"observations={','.join(missing_observations) or '-'}; "
f"validations={','.join(missing_validations) or '-'}"
)
if (
not bool(combination.get("completed"))
or int(combination.get("completed_poses", 0)) != 8
or position_p95 > 0.003
or orientation_p95 > math.radians(2.0)
):
raise ValueError(
"eight-pose four-finger combination validation is incomplete or failed"
)
def finalize_session_artifacts(*args, **kwargs):
raise ValueError("legacy publication is retired; use calibrate_hand --offline-raw with the common finalizer")
@@ -0,0 +1,893 @@
"""Chinese, operator-facing diagnostics for three-camera calibration."""
from __future__ import annotations
import re
from typing import Any, Mapping, Sequence
STATE_NAMES_ZH = {
"PREFLIGHT": "设备和标签预检",
"WAIT_START": "等待开始标定",
"IMPORTING_BASE": "正在读取基础标定会话",
"REVALIDATING_INHERITED": "正在复核继承的四指数据",
"RETURN_BASELINE": "正在恢复目标姿态",
"PREPARE_SWEEP": "正在到达扫描起点",
"SWEEP": "正在采集轨迹",
"FITTING": "正在拟合轨迹和零位",
"VALIDATION_MOVE": "正在移动到随机复测位置",
"VALIDATION_CAPTURE": "正在采集随机复测数据",
"PAUSED": "标定已暂停",
"ABORTED": "标定已终止",
"COMPLETE": "标定已完成",
}
VIEW_NAMES_ZH = {
"front": "正面",
"side": "侧面",
"top": "上面",
}
JOINT_NAMES_ZH = {
"thumb_cmc_pitch": "拇指CMC俯仰",
"thumb_cmc_roll": "拇指CMC滚转",
"thumb_mcp": "拇指MCP",
"thumb_ip": "拇指IP(被动)",
"index_mcp_roll": "食指MCP侧摆",
"index_mcp_pitch": "食指MCP屈伸",
"index_pip": "食指PIP",
"index_dip": "食指DIP(被动)",
"middle_mcp_roll": "中指MCP侧摆",
"middle_mcp_pitch": "中指MCP屈伸",
"middle_pip": "中指PIP",
"middle_dip": "中指DIP(被动)",
"ring_mcp_roll": "无名指MCP侧摆",
"ring_mcp_pitch": "无名指MCP屈伸",
"ring_pip": "无名指PIP",
"ring_dip": "无名指DIP(被动)",
"pinky_mcp_roll": "小指MCP侧摆",
"pinky_mcp_pitch": "小指MCP屈伸",
"pinky_pip": "小指PIP",
"pinky_dip": "小指DIP(被动)",
"thumb_cmc_yaw": "拇指CMC侧摆",
"index_mcp_roll_side": "食指MCP侧摆(侧面校验)",
"middle_mcp_roll_side": "中指MCP侧摆(侧面校验)",
"ring_mcp_roll_side": "无名指MCP侧摆(侧面校验)",
"pinky_mcp_roll_side": "小指MCP侧摆(侧面校验)",
}
def _format_u8(value: Any) -> str:
if value is None:
return "尚无反馈"
return f"{float(value):.1f}"
def _task_text(active: Mapping[str, Any]) -> str:
if not active:
return "尚无活动任务"
view = VIEW_NAMES_ZH.get(str(active.get("view", "")), str(active.get("view", "")))
if active.get("kind") == "fit_failure":
joints = active.get("joints", [])
joint_text = "/".join(
JOINT_NAMES_ZH.get(str(joint), str(joint)) for joint in joints
)
return (
f"{view}机位,{joint_text}拟合检查失败,"
f"电机{active.get('motor_index')}"
f"{active.get('attempt', 1)}次尝试"
)
if active.get("kind") == "zero_model_failure":
joints = active.get("joints", [])
joint_text = "/".join(
JOINT_NAMES_ZH.get(str(joint), str(joint)) for joint in joints
)
return (
f"{view}机位,{joint_text}零位/URDF验证失败,"
f"电机{active.get('motor_index')},不会自动重扫"
)
if active.get("kind") == "motion_stall":
return (
f"电机{active.get('motor_index', '?')}运动停滞,目标"
f"{_format_u8(active.get('target_u8'))}、实际"
f"{_format_u8(active.get('actual_u8'))}"
)
if active.get("kind") == "cross_view_roll_diagnostic":
return (
f"{active.get('finger', '?')}侧摆跨机位诊断完成:"
f"正面最大{float(active.get('front_maximum_deg', 0.0)):.2f}°,"
f"侧面最大{float(active.get('side_maximum_deg', 0.0)):.2f}°"
)
if active.get("kind") == "validation":
return (
f"{view}机位,随机复测,电机{active.get('motor_index')}"
f"目标命令{active.get('command_u8')}"
)
joints = active.get("joints", [])
joint_text = "/".join(
JOINT_NAMES_ZH.get(str(joint), str(joint)) for joint in joints
)
start = active.get("start_u8")
target = active.get("target_u8")
cycle = active.get("cycle", "?")
repetitions = active.get("repetitions", "?")
task = (
f"{view}机位,{joint_text},电机{active.get('motor_index')}"
f"{cycle}/{repetitions}轮,{start}{target}"
)
fit_attempt = int(active.get("fit_attempt", 1))
if fit_attempt > 1:
retry_cycles = active.get("fit_retry_cycles", [])
if retry_cycles:
task += "(补采异常轮" + "/".join(
str(cycle) for cycle in retry_cycles
) + ""
else:
task += f"(拟合补采第{fit_attempt}次)"
return task
def three_camera_reason_zh(
state: str,
reason: str,
active: Mapping[str, Any],
) -> tuple[str, str]:
"""Translate a reason code and provide one concrete operator action."""
reason = str(reason)
sample = active.get("sample", {}) if active else {}
missing = [int(value) for value in sample.get("missing_endpoint_u8", [])]
sample_range = (
f"{_format_u8(sample.get('minimum_u8'))}"
f"{_format_u8(sample.get('maximum_u8'))}"
)
tolerance = sample.get("endpoint_tolerance_u8", "?")
if reason.startswith("motor_state_stalled:"):
fields = reason.split(":")
context = fields[1] if len(fields) > 1 else "unknown"
error_match = re.search(r"error_u8=([0-9.]+)", reason)
error = error_match.group(1) if error_match else "未知"
timeout_match = re.search(r"timeout_seconds=([0-9.]+)", reason)
timeout_value = active.get("timeout_seconds")
if timeout_value is None and timeout_match is not None:
timeout_value = float(timeout_match.group(1))
duration = (
f"连续{float(timeout_value):g}"
if timeout_value is not None
else "在规定时间内"
)
motor = active.get("motor_index")
if motor is not None:
return (
f"电机{motor}反馈{duration}没有向目标推进;目标"
f"{_format_u8(active.get('target_u8'))}、实际"
f"{_format_u8(active.get('actual_u8'))}、误差{error} u8"
f"允许容差±{_format_u8(active.get('tolerance_u8'))} u8"
f"(阶段={context})。程序已保持当前位置。",
"若实际反馈是稳定的固件端点,应只配置该电机该端点的专用容差后"
"重启;若仍在变化或有摩擦,则先排查机械问题,不要反复resume强推。",
)
return (
f"电机反馈{duration}没有向目标推进;停止位置距目标{error}个u8"
f"(阶段={context})。程序已保持当前位置,防止机械碰撞或摩擦加重。",
"检查该电机是否在机械端点稳定饱和或存在碰撞。若实际反馈已是该型号的"
"正常端点,应配置该电机专用端点容差后重启标定;不要反复调用resume强推。",
)
base_reason, separator, reason_detail = reason.partition(":")
if base_reason in {
"sweep_missing_endpoint_bin",
"sweep_bins_too_few",
"sweep_bin_gap_too_large",
"task_precheck_missing_command_127",
"task_precheck_detection_rate_too_low",
"synchronised_tag_state_timeout",
}:
reason = base_reason
detail_label = (
JOINT_NAMES_ZH.get(reason_detail, reason_detail)
if separator and reason_detail
else ""
)
detail_prefix = f"{detail_label}" if detail_label else ""
if "URDF zero offset reached the configured" in reason:
bound_match = re.search(
r"configured\s+([0-9.]+)\s+degree bound", reason
)
bound = bound_match.group(1) if bound_match else "配置的"
hit_text = ""
if "bound:" in reason:
hit_text = reason.split("bound:", 1)[1].split(
"; all_offsets:", 1
)[0]
for name, label in JOINT_NAMES_ZH.items():
hit_text = hit_text.replace(name, label)
hit_suffix = f";触边关节:{hit_text}" if hit_text else ""
return (
f"联合URDF零位求解触及±{bound}°安全边界{hit_suffix}。这不是可靠的"
"零位结果,而是三机位米制位姿或固定关节轴链无法由纯零位旋转共同解释。",
"不要调用resume,也不要增大零位边界。先确认Tag有效黑框边长、三相机"
"内外参和原始CAD URDF;Tag尺寸修正后必须调用start重新采集,旧尺度"
"产生的轨迹不能直接生成修正URDF。",
)
if reason == "sweep_missing_endpoint_bin":
missing_text = "".join(str(value) for value in missing) or "0或255"
return (
f"{detail_prefix}本方向已有{active.get('valid_frames', 0)}帧同步有效数据,但缺少"
f"电机端点{missing_text}附近的有效分箱;采样到的实际电机范围为"
f"{sample_range},端点容差为±{tolerance}。这通常表示电机虽然运动到"
"端点,但该时刻没有同时取得有效Tag图像和电机状态。",
"确认当前机位所需Tag在整个行程(尤其缺失端点)均可见,然后调用"
"/g20_calibration/resume;程序会重新扫描当前方向,不要调用start。",
)
if reason == "sweep_bins_too_few":
return (
f"{detail_prefix}有效电机分箱只有{sample.get('bin_count', 0)}个,要求至少"
f"{sample.get('minimum_bin_count', '?')}个;当前采样范围{sample_range}",
"检查Tag连续识别和电机状态频率,修正后调用resume重新扫描当前方向。",
)
if reason == "sweep_bin_gap_too_large":
gap_start = sample.get("maximum_bin_gap_start_u8")
gap_end = sample.get("maximum_bin_gap_end_u8")
gap_range = (
""
if gap_start is None or gap_end is None
else f"{gap_start}{gap_end}"
)
return (
f"{detail_prefix}轨迹相邻有效电机分箱的最大空缺为{sample.get('maximum_bin_gap', '?')}"
f"{gap_range}允许值不超过"
f"{sample.get('allowed_maximum_bin_gap', '?')}",
"检查运动中Tag是否间歇丢失;修正遮挡、反光或对焦后调用resume。",
)
if reason == "synchronised_tag_state_timeout":
group_reasons = active.get("group_pnp_reasons", {})
if isinstance(group_reasons, Mapping) and group_reasons:
tag_rejections = active.get("pnp_rejection_counts", {})
group_rejections = active.get(
"group_pnp_rejection_counts", {}
)
missing_roles = active.get(
"group_missing_candidate_roles", {}
)
candidate_diagnostics = active.get(
"pnp_candidate_diagnostics", {}
)
view_details: list[str] = []
for view, value in group_reasons.items():
view_name = str(view)
parts = [str(value)]
missing = (
missing_roles.get(view_name, ())
if isinstance(missing_roles, Mapping)
else ()
)
if isinstance(missing, Sequence) and not isinstance(
missing, (str, bytes)
) and missing:
parts.append(
"缺候选=" + ",".join(str(role) for role in missing)
)
counts: dict[str, int] = {}
for source in (tag_rejections, group_rejections):
values = (
source.get(view_name)
if isinstance(source, Mapping)
else None
)
if isinstance(values, Mapping):
for name, count in values.items():
counts[str(name)] = counts.get(str(name), 0) + int(
count
)
if counts:
common = sorted(
counts.items(), key=lambda pair: (-pair[1], pair[0])
)[:3]
parts.append(
"累计拒绝="
+ ",".join(
f"{name}×{count}" for name, count in common
)
)
view_candidates = (
candidate_diagnostics.get(view_name, {})
if isinstance(candidate_diagnostics, Mapping)
else {}
)
if isinstance(view_candidates, Mapping) and missing:
summaries: list[str] = []
for role in missing:
diagnostic = view_candidates.get(str(role), {})
if not isinstance(diagnostic, Mapping):
continue
summaries.append(
f"{role}(solve="
f"{int(diagnostic.get('solved_candidate_count', 0))},"
"reproj="
f"{int(diagnostic.get('reprojection_candidate_count', 0))},"
"tilt="
f"{int(diagnostic.get('independent_tilt_candidate_count', 0))})"
)
if summaries:
parts.append("候选统计=" + ",".join(summaries))
view_details.append(
f"{VIEW_NAMES_ZH.get(view_name, view_name)}="
+ "".join(parts)
)
reason_text = "".join(view_details)
return (
f"{detail_prefix}已经取得部分有效轨迹,但Tag仍可见且反馈正常时,"
"后续连续图像帧"
"被整组PnP几何检查拒绝"
f"{reason_text}),因此无法与电机状态形成有效轨迹帧。",
"不要调整或反复粘贴Tag;保留当前会话中的"
"group_pnp_candidate_event"
"按缺失角色的候选统计检查PnP分支逻辑。",
)
return (
f"{detail_prefix}运动过程中连续超过允许时间没有取得“所需Tag全部有效且能与电机状态"
"按时间戳配对”的图像帧。",
"查看下面活动机位的缺失Tag,确认状态话题仍在更新;修正后调用resume,"
"程序会重扫当前方向。",
)
if reason == "task_precheck_missing_command_127":
return (
f"{detail_prefix}低速预检没有取得反馈127附近的同步Tag样本。",
"检查中位姿态的Tag遮挡和反光;程序只会重扫当前物理任务。",
)
if reason == "task_precheck_detection_rate_too_low":
return (
f"{detail_prefix}低速预检的有效Tag识别率低于门限。",
"检查该机位当前任务Tag的遮挡、反光和对焦;程序只会重扫当前物理任务。",
)
if reason == "sweep_start_position_timeout":
return (
f"电机{active.get('motor_index')}未在规定时间到达扫描起点"
f"{active.get('start_u8')},当前实际值{_format_u8(active.get('actual_u8'))}",
"检查CAN、机械手使能和是否存在机械卡阻,确认安全后调用resume。",
)
if reason == "sweep_start_tag_timeout":
group_reasons = active.get("group_pnp_reasons", {})
progress_by_view = active.get("pnp_initialization_progress", {})
tag_rejections = active.get("pnp_rejection_counts", {})
group_rejections = active.get("group_pnp_rejection_counts", {})
if any(
isinstance(value, Mapping) and bool(value)
for value in (
group_reasons,
progress_by_view,
tag_rejections,
group_rejections,
)
):
details: list[str] = []
views = set()
for value in (
group_reasons,
progress_by_view,
tag_rejections,
group_rejections,
):
if isinstance(value, Mapping):
views.update(str(view) for view in value)
for view in sorted(views):
parts: list[str] = []
progress = (
progress_by_view.get(view)
if isinstance(progress_by_view, Mapping)
else None
)
if isinstance(progress, Mapping):
parts.append(
"初始化"
f"{int(progress.get('accepted', 0))}/"
f"{int(progress.get('required', 0))}"
)
counts: dict[str, int] = {}
for source in (tag_rejections, group_rejections):
values = (
source.get(view)
if isinstance(source, Mapping)
else None
)
if isinstance(values, Mapping):
for name, count in values.items():
counts[str(name)] = (
counts.get(str(name), 0) + int(count)
)
if counts:
common = sorted(
counts.items(), key=lambda pair: (-pair[1], pair[0])
)[:3]
parts.append(
"累计拒绝="
+ ",".join(
f"{name}×{count}" for name, count in common
)
)
latest = (
group_reasons.get(view)
if isinstance(group_reasons, Mapping)
else None
)
if latest and not str(latest).startswith(
"group_initializing:"
):
parts.append(f"最后状态={latest}")
if parts:
details.append(
f"{VIEW_NAMES_ZH.get(view, view)}=" + "".join(parts)
)
reason_text = "".join(details) or "未形成完整初始化窗口"
return (
"被测电机已经到达扫描起点,所需Tag也可见,但三维PnP位姿初始化"
f"没有完成({reason_text}),因此没有生成同步端点帧。",
"不要根据可见性重复粘贴Tag;保留累计拒绝原因并检查PnP候选选择。",
)
return (
"被测电机已经到达扫描起点,但当前任务所需的实时运动Tag没有形成足够的"
"同步有效帧。允许遮挡的固定掌部Tag会显示为“锁”,不会触发此错误。",
"只检查标记为✗的实时运动Tag、反光和外部遮挡;不要移动相机或手掌底座。",
)
if reason == "sweep_timeout":
return (
"当前方向在规定时间内未完成端点到达、有效帧数和行程覆盖要求。",
"检查电机实际值、Tag连续识别和标定速度,修正后调用resume。",
)
if reason == "return_baseline_timeout":
return (
"一个或多个标定电机未在规定时间返回基准命令。",
"检查机械手状态、CAN和机械卡阻,确认安全后调用resume。",
)
if reason == "validation_move_timeout":
return (
"随机复测时电机未在规定时间到达目标命令。",
"检查机械手状态和机械卡阻,确认安全后调用resume。",
)
if reason == "validation_capture_timeout":
return (
"随机复测位置没有采集到足够的同步有效Tag帧。",
"检查当前机位Tag可见性后调用resume。",
)
if reason == "palm_orientation_quality_failed":
failures = active.get("failures", [])
detail = (
str(failures[0].get("reason", "方向观测不足"))
if failures
else "方向观测不足"
)
if "thumb_cmc_" in detail:
return (
"拇指CMC yaw无法由顶部Tag 8/9的零位邻近短轨迹稳定确定:"
+ detail,
"保持Tag安装不变;确保顶部Tag 8/9在拇指CMC pitch和roll"
"从零位开始的前1/4行程持续可见后重新标定。",
)
return (
"掌部公共方向无法由至少三根手指的短时正面轨迹稳定确定:"
+ detail,
"保持Tag安装不变;让正面Tag 1013在对应MCP-pitch起始段"
"至少可见15°行程后重新标定。",
)
if reason in {"joint_fit_check_failed", "joint_fit_systematic_failure"}:
metric_names = {
"plane_rms_mm": "平面拟合RMS",
"radial_rms_mm": "圆半径拟合RMS",
"radius_mm": "拟合半径",
"image_radial_rms_px": "二维圆半径拟合RMS",
"image_radial_p95_px": "二维圆半径误差P95",
"image_radius_px": "二维拟合半径",
"arc_deg": "实测圆弧",
"monotonic_correction_deg": "最大单调修正",
"hysteresis_deg": "最大正反程差",
"baseline_hysteresis_deg": "baseline正反程关节角差",
"baseline_directional_gap_deg": "baseline方向分支间隙",
"baseline_directional_gap_range_deg": "baseline分支间隙跨轮极差",
"cycle_travel_range_deg": "三轮行程差",
"rotation_orthogonal_rms_deg": "三维旋转轴外残差RMS",
"axis_plane_rms_mm": "三维圆轴向RMS",
"axis_radial_rms_mm": "三维圆半径RMS",
"axis_pose_line_rms_mm": "姿态轨迹轴线RMS",
"axis_line_cycle_rms_mm": "四轮轴线位置RMS",
"rotation_circle_axis_difference_deg": "姿态轴与圆轨迹轴夹角",
"axis_cycle_difference_deg": "各轮转轴方向极差",
"cross_view_roll_curve": "正面/侧面关节角曲线差异RMS",
"third_cycle_axis_holdout_deg": "第三轮留出零位可观测轴向误差",
"third_cycle_axis_line_rms_mm": "第三轮留出轴线RMS",
"third_cycle_trajectory_p95_deg": "第三轮留出轨迹误差P95",
"zero_cycle_offset_range_deg": "训练轮零位极差",
"zero_confidence_95_half_width_deg": "零位95%置信半宽",
"state_image_sync_p95_ms": "图像与电机状态同步误差P95",
"tag_valid_rate_percent": "所需Tag同时有效率",
}
metric_units = {
"plane_rms_mm": "mm",
"radial_rms_mm": "mm",
"radius_mm": "mm",
"image_radial_rms_px": "px",
"image_radial_p95_px": "px",
"image_radius_px": "px",
"arc_deg": "°",
"monotonic_correction_deg": "°",
"hysteresis_deg": "°",
"baseline_hysteresis_deg": "°",
"baseline_directional_gap_deg": "°",
"baseline_directional_gap_range_deg": "°",
"cycle_travel_range_deg": "°",
"rotation_orthogonal_rms_deg": "°",
"axis_plane_rms_mm": "mm",
"axis_radial_rms_mm": "mm",
"axis_pose_line_rms_mm": "mm",
"axis_line_cycle_rms_mm": "mm",
"rotation_circle_axis_difference_deg": "°",
"axis_cycle_difference_deg": "°",
"third_cycle_axis_holdout_deg": "°",
"third_cycle_axis_line_rms_mm": "mm",
"third_cycle_trajectory_p95_deg": "°",
"zero_cycle_offset_range_deg": "°",
"zero_confidence_95_half_width_deg": "°",
"state_image_sync_p95_ms": "ms",
"tag_valid_rate_percent": "%",
"cross_view_roll_curve": "°",
}
details: list[str] = []
for failure in active.get("failures", []):
joint = JOINT_NAMES_ZH.get(
str(failure.get("joint")), str(failure.get("joint"))
)
metric = str(failure.get("metric", ""))
if metric in metric_names:
comparison = str(failure.get("comparison", ""))
requirement = "不超过" if comparison == "maximum" else "至少"
unit = metric_units[metric]
detail = (
f"{joint}{metric_names[metric]}"
f"{float(failure.get('actual', 0.0)):.2f}{unit}"
f"要求{requirement}{float(failure.get('limit', 0.0)):.2f}{unit}"
)
cycle_travel = failure.get("cycle_travel_deg", [])
if cycle_travel:
detail += "(各轮=" + "/".join(
f"{float(value):.2f}°" for value in cycle_travel
) + ""
cycle_values = failure.get("cycle_values_deg", [])
if not cycle_values:
cycle_values = failure.get("cycle_offset_deg", [])
if cycle_values:
detail += "(各轮=" + "/".join(
f"{float(value):.2f}°" for value in cycle_values
) + ""
details.append(detail)
else:
cycle = failure.get("cycle")
cycle_text = "" if cycle is None else f"{cycle}"
details.append(
f"{joint}{cycle_text}{metric or '轨迹'}拟合失败:"
f"{failure.get('reason', '未知原因')}"
)
detail_text = "".join(details) or "当前关节的轨迹拟合未通过"
directional_gap_failure = any(
str(failure.get("metric", "")).startswith(
"baseline_directional_gap"
)
for failure in active.get("failures", [])
)
if reason == "joint_fit_systematic_failure":
cross_view_systematic = any(
failure.get("classification")
in {
"stable_cross_view_installation_or_model_bias",
"stable_cross_view_direction_conflict",
}
for failure in active.get("failures", [])
)
suggestion = (
"四轮都出现稳定的正面/侧面差异,属于Tag安装外参或跨视角模型偏差,"
"继续重扫不会改善;检查Tag刚性安装与跨视角安装变换,不要放宽门限。"
if cross_view_systematic
else "各轮重复出现同一模型冲突,继续运动不会改善;程序已禁止自动重扫。"
"请直接复制诊断块给开发者,不要放宽门限。"
)
else:
source_task_names = set(active.get("source_task_names", []))
thumb_yaw_source_retry = source_task_names == {
"thumb_cmc_pitch_front",
"thumb_cmc_roll_front",
}
suggestion = (
"方向分支已由软件保留,不要放宽门限;请检查传动回差或高支架刚度,"
"处理后重新执行一键标定命令,程序会从最近可靠断点继续。"
if directional_gap_failure
else (
"保持顶部Tag 8/9无遮挡;程序只替换决定yaw零位的"
"CMC pitch/roll顶部轴观测并重扫"
f"{active.get('directions_to_rescan', 16)}个方向,"
"不会无效重扫yaw侧摆。"
if thumb_yaw_source_retry
else "修正Tag位置、遮挡或机械行程后重新执行一键标定命令;"
"程序只清除当前失败关节的数据并重扫"
f"{active.get('directions_to_rescan', 6)}个方向,"
"不需要手工调用ROS服务。"
)
)
return (
detail_text + "。程序已在当前关节结束后立即停止后续步骤。",
suggestion,
)
if reason == "zero_model_validation_failed":
reason_names = {
"zero_offset_reached_configured_bound": "零位解触及安全边界",
"zero_offset_exceeds_configured_limit": "零位估计超过安全范围",
"zero_offset_reached_diagnostic_bound": "零位估计仍触及诊断搜索边界",
"zero_offset_cycle_difference_too_large": "三轮零位离散过大",
"zero_offset_not_statistically_significant": "零位偏移未达到统计显著性",
"zero_axis_cone_mismatch_too_large": (
"父子轴夹角与原始URDF不一致,零位旋转无法解释"
),
"zero_phase_axis_line_residual_too_large": (
"整段SE(3)运动无法稳定确定平行轴线相位"
),
"zero_offset_did_not_improve_with_95pct_confidence": (
"第三轮留出验证未以95%置信度改善"
),
}
details: list[str] = []
for failure in active.get("failures", []):
joint = JOINT_NAMES_ZH.get(
str(failure.get("joint")), str(failure.get("joint"))
)
if failure.get("metric") == "zero_guard":
reason_text = reason_names.get(
str(failure.get("reason")), str(failure.get("reason"))
)
if "actual_deg" in failure and "limit_deg" in failure:
reason_text += (
f"(估计{float(failure['actual_deg']):+.2f}°,"
f"允许±{float(failure['limit_deg']):.2f}°)"
)
details.append(f"{joint}{reason_text}")
return (
"轨迹采集已完成,但零位/URDF几何验证失败"
+ ("" + "".join(details) + "" if details else "")
+ "。程序没有生成正式JSON或修正URDF。",
"该类稳定模型失败不能靠重复运动修复,程序不会自动重扫;"
"请检查Tag固定、相机外参和原始URDF后重新启动新标定。",
)
if reason in {"waiting_for_devices_and_sdk", "device_preflight_lost"}:
return (
"正在等待三台相机数据、内外参身份以及机械手SDK反馈就绪;此阶段不以Tag可见性阻止基准恢复。",
"保持机械手运动范围无障碍;设备就绪后系统会先安全恢复基准形态,再检查掌部Tag。",
)
if reason in {"waiting_for_three_cameras_tags_and_sdk", "preflight_lost"}:
return (
"正在等待三台相机内参、外参身份匹配、帧率、全部必需Tag以及机械手SDK同时就绪。",
"根据下面每个机位的缺失Tag和有效率排查;全部就绪后程序会进入等待开始状态。",
)
if reason == "waiting_for_baseline_tags_after_recovery":
return (
"机械手已经稳定恢复到基准形态,正在用新采集的画面确认三台相机各自的固定掌部Tag。",
"若某个掌部Tag持续缺失,只调整遮挡手指或检查Tag固定情况,不要移动相机和手掌底座。",
)
if reason == "locking_fixed_base_references":
return (
"基准形态Tag预检已通过,正在把三台相机的固定掌部Tag稳健锁定为本会话参考。",
"无需操作;锁定完成后允许任务姿态遮挡固定掌部Tag。",
)
if reason == "fixed_base_reference_moved":
return (
"顶部Tag 8在本会话基准锁定后连续多帧发生角点位移;程序已立即保持机械手当前位置,"
"本会话中已采集数据不再用于发布。",
"Tag 8允许在下一次标定预检前重新摆放,但本次不能继续;固定Tag 8和顶部相机后"
"重新启动新会话。",
)
if reason == "waiting_for_task_tags_at_sweep_start":
return (
"电机已到扫描起点,正在等待当前任务的实时运动Tag;显示为“锁”的固定掌部Tag"
"允许被手指遮挡。",
"只检查标记为✗的实时运动Tag;若均为✓或锁,程序会自动开始运动。",
)
if reason == "call_start_for_baseline_recovery":
return (
"相机数据和机械手反馈已就绪,等待一键程序触发安全基准恢复。",
"保持机械手运动范围无障碍;程序会自动开始,无需手工调用ROS服务。",
)
if reason == "call_start":
return (
"三机位预检已经通过,等待操作员确认开始。",
"清空机械手运动范围后调用/g20_calibration/start。",
)
if reason == "operator_pause":
return "操作员主动暂停了标定。", "确认安全后调用/g20_calibration/resume。"
if reason == "operator_abort":
return "操作员终止了本次标定,程序保持终止时的当前姿态。", "需要重新启动一次新标定。"
if reason == "collecting_timestamp_synchronised_tag_centres":
return "正在按时间戳配对Tag图像和电机状态并采集当前轨迹。", "无需操作,保持相机、标签和底座不动。"
if reason == "collecting_dedicated_baseline_hold":
return (
"正在从当前方向到达关节baseline并静止采集Tag与电机反馈;这批数据单独用于回差验收。",
"无需操作,保持相机、标签和底座不动。",
)
if reason == "steady checkpoint target is missing":
return (
"首轮稳态检查点已经到达最终端点,但采集状态没有及时切换到端点完成阶段。",
"程序已停止发布并保留已采样数据;这是软件状态切换问题,不需要调整相机、Tag或机械手。",
)
if reason == "cross_view_roll_front_failure_deferred":
return (
"正面侧摆回差不合格已保留,诊断模式将继续采集同一手指的侧面数据。",
"无需操作;该诊断会锁定URDF发布。",
)
if reason == "cross_view_roll_diagnostic_complete":
interpretation = str(active.get("interpretation", ""))
explanations = {
"both_views_confirm_direction_dependent_pose": (
"正面和侧面都确认了方向相关姿态,优先判断为roll输出机构或共同下游链的真实回差。"
),
"front_only_difference_check_roll_tag_bracket_or_front_pnp": (
"只有正面差异超限,优先检查roll Tag高支架刚度和正面PnP。"
),
"side_only_difference_check_side_tag_chain_or_side_pnp": (
"只有侧面差异超限,优先检查侧面Tag链和侧面PnP。"
),
"both_views_within_formal_hysteresis_limit": (
"两个机位的静止回差均满足正式门限。"
),
}
return (
explanations.get(interpretation, "四指侧摆跨机位诊断已经完成。")
+ " 本次为诊断会话,不会生成或发布URDF。",
"保存当前状态和raw_samples.jsonl;根据两机位结论处理后重新启动正式标定。",
)
if reason == "capturing_random_validation_pose":
return "正在当前随机命令位置采集复测数据。", "无需操作,保持设备不动。"
if reason in {"calibration_passed", "calibration_complete"}:
return "三维轨迹、关节轴零位和第三轮留出验证已经完成。", "检查JSON、修正URDF路径和quality.passed。"
if reason == "quality_failed":
return "标定流程完成,但拟合或随机复测质量没有达到验收阈值。", "检查最终JSON的quality以及启动终端中的拟合日志。"
if reason.startswith("validated_endpoint_zero_state"):
return (
"轨迹和URDF零位验证已经通过,但发布前检测到端点零位状态缺失或与"
"已验证模型不一致;这是程序内部状态生命周期错误,结果未发布。",
"不要移动相机、Tag或机械手底座;保留当前会话并把原因码交给开发者。",
)
if reason.startswith("PUB-ARTIFACT-601:"):
return (
"标定节点已经生成通过质量门限的JSON和候选URDF,但一键程序在正式发布前"
"发现这对产物的坐标、限位、哈希或资源一致性检查失败;原始URDF未被覆盖。",
"不要重新标定相机或调整Tag;保留本会话产物和启动日志供开发者检查发布契约。",
)
if reason == "combination_pose_prediction_failed":
return (
"单关节、零位和URDF几何验证已通过,但当前多关节组合姿态的Tag实测位姿与模型预测超过门限。",
"程序会在原姿态重新初始化PnP并自动复测;若最终仍失败,请把raw_samples.jsonl中的"
"combination_validation_failure记录交给开发者,不要重新采集16个单关节任务。",
)
if reason.startswith("prepare_") or state == "PREPARE_SWEEP":
return "正在把当前电机移动到本方向的扫描起点并等待稳定。", "无需操作。"
if reason == "holding_same_finger_clearance_before_next_task":
return (
"同一根手指的上一项已经完成;相邻手指继续保持当前避让姿态,只调整"
"被测关节以衔接下一项。",
"无需操作,不要手动展开正在避让的手指。",
)
if state == "RETURN_BASELINE":
return "正在把已使用的标定电机恢复到目标姿态。", "无需操作。"
if state == "FITTING":
return "所有扫描已经完成,正在联合拟合三维机械轴和URDF零位偏移。", "无需操作。"
return f"未分类原因码:{reason}", "保留该原因码和启动终端日志用于进一步定位。"
def render_three_camera_status_text_zh(payload: Mapping[str, Any]) -> str:
"""Render the complete operator status; the JSON topic remains unchanged."""
state = str(payload.get("state", ""))
active = payload.get("active", {})
reason_zh, action_zh = three_camera_reason_zh(
state, str(payload.get("reason", "")), active
)
progress = float(payload.get("progress", 0.0))
completed = payload.get("completed_sweeps", 0)
total = payload.get("total_sweeps", 0)
scan_progress = float(
payload.get(
"scan_progress",
0.0 if not total else float(completed) / float(total),
)
)
lines = [
f"状态:{STATE_NAMES_ZH.get(state, state)}{state}",
f"原因:{reason_zh}",
f"建议:{action_zh}",
f"总体进度:{progress:.1%}(计划扫描{completed}/{total}个方向,"
f"扫描进度{scan_progress:.1%}",
f"当前任务:{_task_text(active)}",
]
if state == "RETURN_BASELINE":
baseline_command = payload.get("baseline_command_u8", [])
return_command = payload.get("return_command_u8", baseline_command)
label = "恢复姿态" if return_command != baseline_command else "基准姿态"
lines.append(f"正在确认{label}{return_command}")
if active and active.get("kind") not in {
"fit_failure",
"zero_model_failure",
"motion_stall",
}:
retry_count = int(active.get("automatic_retry_count", 0))
if retry_count:
lines.append(
"自动重试:当前方向已自动重扫"
f"{retry_count}/{active.get('automatic_retry_limit', '?')}次,"
f"速度比例{float(active.get('retry_speed_scale', 1.0)):.0%}"
f"端点保持{float(active.get('endpoint_hold_seconds', 0.0)):.2f}s"
)
sample = active.get("sample", {})
motion_progress = active.get("motion_progress")
motion_text = (
"未知" if motion_progress is None else f"{float(motion_progress):.1%}"
)
lines.append(
"运动采样:"
f"目标{active.get('target_u8', active.get('command_u8', '?'))}"
f"实际{_format_u8(active.get('actual_u8'))}"
f"本方向{motion_text},有效帧{active.get('valid_frames', 0)}"
f"实际采样范围{_format_u8(sample.get('minimum_u8'))}"
f"{_format_u8(sample.get('maximum_u8'))}"
)
detection_frames = int(active.get("detection_frames", 0))
if detection_frames:
lines.append(
"本方向Tag检出:"
f"{float(active.get('detection_rate', 0.0)):.1%}"
f"{active.get('detection_valid_frames', 0)}/"
f"{detection_frames}帧)"
)
auxiliary = active.get("auxiliary_motors", [])
if auxiliary:
lines.append(
"避挡姿态:"
+ "".join(
f"电机{item.get('motor_index')}目标"
f"{item.get('command_u8')}、实际"
f"{_format_u8(item.get('actual_u8'))}"
for item in auxiliary
)
)
speed = active.get("speed", {})
if speed:
lines.append(
"阶段速度:五指目标"
f"{speed.get('commanded_finger_speed')}SDK报告"
f"{speed.get('reported_finger_speed')}"
)
if active.get("sweep_timeout_seconds") is not None:
lines.append(
"运动保护:扫描超时"
f"{float(active['sweep_timeout_seconds']):.1f}s"
"连续"
f"{float(active.get('motor_stall_timeout_seconds', 0.0)):.1f}s"
"进展不足"
f"{float(active.get('motor_stall_minimum_progress_u8', 0.0)):.1f}"
"则立即暂停"
)
lines.append("机位:")
for name, view in payload.get("views", {}).items():
missing = view.get("missing_tag_ids", [])
missing_text = "" if not missing else ",".join(map(str, missing))
lines.append(
f"- {VIEW_NAMES_ZH.get(str(name), str(name))}"
f"{'就绪' if view.get('ready') else '等待'}"
f"外参{'匹配' if view.get('camera_extrinsics_valid') else '不匹配'}"
f"{float(view.get('detection_hz', 0.0)):.1f}Hz"
f"全部必需Tag同时有效率{float(view.get('valid_rate', 0.0)):.1%}"
f"当前缺失Tag={missing_text}"
)
extrinsics_error = payload.get("camera_extrinsics_error")
if extrinsics_error:
lines.append(f"外参文件:{extrinsics_error}")
lines.append(f"JSON结果:{payload.get('result_path') or '尚未生成'}")
lines.append(
f"修正URDF{payload.get('corrected_urdf_path') or '尚未生成'}"
)
return "\n".join(lines)
@@ -0,0 +1,29 @@
"""Independent reviewed profile for the right 19-Tag product layout."""
from linkerhand_calibration.core import ProfileKey
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.g20.profile import G20_RIGHT_19_LAYOUT, get_hand_calibration_profile
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.g20.zero_policy import (
RIGHT_19_MECHANICAL_ENDPOINT_JOINTS,
RIGHT_19_POST_SOLVE_ENDPOINT_JOINTS,
get_zero_calibration_profile,
)
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.registry import RegisteredProfile
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.g20._adapter import adapt_profile
from linkerhand_calibration.profiles.loader import load_bundled_hand_profile
KEY = ProfileKey("G20", "right", G20_RIGHT_19_LAYOUT, 1)
def build_profile() -> RegisteredProfile:
hand = get_hand_calibration_profile(KEY.side, KEY.layout)
zero = get_zero_calibration_profile(KEY.side, KEY.layout)
return adapt_profile(
key=KEY,
namespace="/g20_calibration",
hand_profile=hand,
zero_profile=zero,
mechanical_endpoint_joints=RIGHT_19_MECHANICAL_ENDPOINT_JOINTS,
post_solve_endpoint_joints=RIGHT_19_POST_SOLVE_ENDPOINT_JOINTS,
typed_profile=load_bundled_hand_profile(KEY.layout),
)
@@ -0,0 +1,10 @@
"""Legacy import path; all online work uses the common product runner."""
from linkerhand_calibration.runtime.runner import main
from linkerhand_calibration.compat.legacy_diagnostic_tools.runner_helpers import (
_automatic_resume_candidate, _calibration_node_exited_before_status,
_launch_command, _resolve_partial_base_session, _status_timeout_seconds,
)
if __name__ == "__main__":
main()
@@ -0,0 +1,133 @@
"""Exact JSON handoff between G20 fitting and the existing URDF writer."""
from __future__ import annotations
import hashlib
import json
import math
from pathlib import Path
from typing import Any, Mapping
SCHEMA_VERSION = 1
ARTIFACT_TYPE = "linkerhand_g20_urdf_correction_input"
def _sha256_file(path: str | Path) -> str:
digest = hashlib.sha256()
with Path(path).expanduser().resolve().open("rb") as stream:
for chunk in iter(lambda: stream.read(1024 * 1024), b""):
digest.update(chunk)
return digest.hexdigest()
def build_g20_urdf_input_payload(
*,
side: str,
layout_id: str,
serial_number: str,
source_urdf: str | Path,
offsets_rad: Mapping[str, float],
endpoint_anchored_offsets_rad: Mapping[str, float],
) -> dict[str, Any]:
"""Build the exact correction parameters that will be read from disk."""
payload: dict[str, Any] = {
"schema_version": SCHEMA_VERSION,
"artifact_type": ARTIFACT_TYPE,
"profile_id": f"G20/{str(side).lower()}/{layout_id}",
"model": "G20",
"side": str(side).lower(),
"layout_id": str(layout_id),
"serial_number": str(serial_number),
"source_urdf_sha256": _sha256_file(source_urdf),
"offsets_rad": {
str(name): float(value) for name, value in sorted(offsets_rad.items())
},
"endpoint_anchored_offsets_rad": {
str(name): float(value)
for name, value in sorted(endpoint_anchored_offsets_rad.items())
},
}
validate_g20_urdf_input_payload(payload)
return payload
def validate_g20_urdf_input_payload(payload: Mapping[str, Any]) -> None:
required = {
"schema_version", "artifact_type", "profile_id", "model", "side",
"layout_id", "serial_number", "source_urdf_sha256", "offsets_rad",
"endpoint_anchored_offsets_rad",
}
if set(payload) != required:
raise ValueError("G20 URDF correction input has unexpected fields")
if (
payload.get("schema_version") != SCHEMA_VERSION
or payload.get("artifact_type") != ARTIFACT_TYPE
or payload.get("model") != "G20"
or payload.get("side") not in {"left", "right"}
):
raise ValueError("G20 URDF correction input identity is invalid")
expected_profile = (
f"G20/{payload['side']}/{payload['layout_id']}"
)
if payload.get("profile_id") != expected_profile:
raise ValueError("G20 URDF correction input profile is invalid")
source_hash = str(payload.get("source_urdf_sha256", ""))
if len(source_hash) != 64 or any(
character not in "0123456789abcdef" for character in source_hash
):
raise ValueError("G20 URDF correction input source hash is invalid")
offsets = payload.get("offsets_rad")
endpoints = payload.get("endpoint_anchored_offsets_rad")
if not isinstance(offsets, Mapping) or not offsets:
raise ValueError("G20 URDF correction input offsets are missing")
if not isinstance(endpoints, Mapping) or not set(endpoints) <= set(offsets):
raise ValueError("G20 URDF correction input endpoint offsets are invalid")
if any(not math.isfinite(float(value)) for value in offsets.values()):
raise ValueError("G20 URDF correction input contains a non-finite offset")
if any(not math.isfinite(float(value)) for value in endpoints.values()):
raise ValueError("G20 URDF correction input has a non-finite endpoint")
def load_g20_urdf_input(
path: str | Path,
*,
source_urdf: str | Path,
side: str,
layout_id: str,
serial_number: str,
) -> tuple[dict[str, float], dict[str, float]]:
"""Read, validate and authenticate correction parameters from JSON."""
source = Path(path).expanduser().resolve()
try:
payload = json.loads(source.read_text(encoding="utf-8"))
except (OSError, json.JSONDecodeError) as error:
raise ValueError(f"could not read G20 URDF correction input: {source}") from error
if not isinstance(payload, Mapping):
raise ValueError("G20 URDF correction input must be a JSON object")
validate_g20_urdf_input_payload(payload)
if (
payload["side"] != str(side).lower()
or payload["layout_id"] != str(layout_id)
or payload["serial_number"] != str(serial_number)
):
raise ValueError("G20 URDF correction input does not match the session")
if payload["source_urdf_sha256"] != _sha256_file(source_urdf):
raise ValueError("G20 source URDF changed after calibration JSON was written")
return (
{
str(name): float(value)
for name, value in payload["offsets_rad"].items()
},
{
str(name): float(value)
for name, value in payload["endpoint_anchored_offsets_rad"].items()
},
)
__all__ = [
"build_g20_urdf_input_payload",
"load_g20_urdf_input",
"validate_g20_urdf_input_payload",
]
@@ -0,0 +1,17 @@
"""Reviewed static-zero, endpoint, and mimic topology exports."""
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.g20.profile import MIMIC_DERIVED_FINGER_DIPS
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.g20.zero_solver import (
RIGHT_19_ENDPOINT_MEASUREMENT_JOINTS,
RIGHT_19_MECHANICAL_ENDPOINT_JOINTS,
RIGHT_19_POST_SOLVE_ENDPOINT_JOINTS,
get_zero_calibration_profile,
)
__all__ = [
"MIMIC_DERIVED_FINGER_DIPS",
"RIGHT_19_ENDPOINT_MEASUREMENT_JOINTS",
"RIGHT_19_MECHANICAL_ENDPOINT_JOINTS",
"RIGHT_19_POST_SOLVE_ENDPOINT_JOINTS",
"get_zero_calibration_profile",
]
@@ -0,0 +1,931 @@
"""Three-dimensional joint-axis fitting and URDF zero correction."""
from __future__ import annotations
from dataclasses import replace
from datetime import datetime
import math
from pathlib import Path
import re
from typing import Any, Mapping, Sequence
import xml.etree.ElementTree as ET
import numpy as np
from scipy.spatial.transform import Rotation
from linkerhand_calibration.core import fit_rotation_axis, robust_rotation_summary
from linkerhand_calibration.core.urdf import (
UrdfCorrectionPlan,
UrdfJointPatch,
UrdfPatchSet,
materialize_relative_mesh_assets as _materialize_relative_mesh_assets,
write_urdf_patches,
)
from linkerhand_calibration.compat.legacy_diagnostic_tools.models.g20.profile import (
G20_RIGHT_19_LAYOUT,
IMAGE_TRAJECTORY_JOINTS,
LEFT_HAND_PROFILE,
RIGHT_19_END_ON_IMAGE_CURVE_JOINTS,
RIGHT_19_VISUALLY_MEASURED_PASSIVE_DIPS,
HandCalibrationProfile,
JointCurveFit,
get_hand_calibration_profile,
)
from linkerhand_calibration.sample_schema import explicit_domain_value
from linkerhand_calibration.trajectory import (
_fit_circle_with_axis,
_fit_joint_curve,
_fit_plane_axis,
_plane_basis,
)
from linkerhand_calibration.core.fitting import spatial as _spatial
from linkerhand_calibration.core.urdf.kinematics import UrdfKinematicModel, _parse_triplet, _axis_rotation
from linkerhand_calibration.core.fitting.spatial import (
ZeroCalibrationProfile,
circle_direction_is_constrained,
select_cross_view_roll_direction_source,
_zero_sensitive_axis_error_rad,
_axis_cone_mismatch_rad,
_vector,
_pose_matrix,
_relative_rotation,
_reference_group_key,
_canonical_reference_records,
_interpolate_reference_rotation,
_near_zero_records,
_baseline_reference,
baseline_hysteresis_by_cycle_rad,
fit_rotation_joint_curve,
measure_rotation_joint_observation,
measure_joint_curve_observation,
rotation_curve_holdout_errors,
joint_curve_holdout_errors,
JointAxisMeasurement,
PalmOrientationMeasurement,
_incremental_common_rotation_axis,
fit_partial_palm_orientation_measurement,
fit_partial_palm_orientation_measurements,
with_depth_free_axis_projection,
cross_view_side_line_source,
axis_line_uses_depth_free_interpretation_plane,
refit_axis_line_group_with_shared_radius,
maximum_axis_line_cycle_spread_m,
axis_line_cycle_rms_m,
_fit_axis_point_from_pose_trajectory,
ZeroSolveResult,
_angles_from_state
)
def solve_urdf_zero_offsets(**kwargs) -> ZeroSolveResult:
"""Compatibility input binding; all numerical work lives in core."""
options = dict(kwargs)
profile = options.get("zero_profile")
if profile is None:
profile = get_zero_calibration_profile(options.get("hand_type", "left"), options.get("tag_layout", "legacy_11"))
pattern = profile.hand.layout_id == G20_RIGHT_19_LAYOUT
profile = replace(profile,
parallel_root_pattern=pattern,
common_mode_zero_joints=tuple(name for name in profile.direct_zero_joints if name.endswith("_mcp_roll") and not name.startswith("thumb_")) if pattern else (),
small_offset_joints=frozenset(name for name in profile.direct_zero_joints if name.startswith(("index_", "middle_", "ring_", "pinky_"))),
orientation_anchor_joint=profile.orientation_anchor_joint or ("thumb_cmc_pitch" if profile.base_pose_strategy == "thumb_serial" else f"{profile.reference_finger}_mcp_pitch"),
base_pose_strategy="serial_chain" if profile.base_pose_strategy == "thumb_serial" else profile.base_pose_strategy,
)
options["zero_profile"] = profile
return _spatial.solve_urdf_zero_offsets(**options)
def fit_joint_axis_measurement(*args, **kwargs) -> JointAxisMeasurement:
kwargs.setdefault("constrained_circle_joints", CONSTRAINED_CIRCLE_JOINTS)
return _spatial.fit_joint_axis_measurement(*args, **kwargs)
def _build_zero_profile(hand: HandCalibrationProfile) -> ZeroCalibrationProfile:
if hand.layout_id == G20_RIGHT_19_LAYOUT:
return _build_right_19_zero_profile(hand)
reference = hand.reference_finger
reference_roll = f"{reference}_mcp_roll"
reference_pitch = f"{reference}_mcp_pitch"
reference_pip = f"{reference}_pip"
reference_dip = f"{reference}_dip"
direct = (
"thumb_cmc_roll",
"thumb_cmc_yaw",
"thumb_cmc_pitch",
"thumb_mcp",
reference_roll,
reference_pitch,
reference_pip,
)
inherited = {
f"{finger}_{suffix}": f"{reference}_{suffix}"
for finger in ("index", "middle", "ring", "pinky")
if finger != reference
for suffix in ("mcp_roll", "mcp_pitch", "pip")
}
axis_parent = {
"thumb_cmc_yaw": "thumb_cmc_roll",
"thumb_cmc_pitch": "thumb_cmc_yaw",
reference_pitch: reference_roll,
}
phase_parent = {
"thumb_mcp": "thumb_cmc_pitch",
"thumb_ip": "thumb_mcp",
reference_pip: reference_pitch,
reference_dip: reference_pip,
}
observer = {
"thumb_cmc_roll": "thumb_cmc_yaw",
"thumb_cmc_yaw": "thumb_cmc_pitch",
"thumb_cmc_pitch": "thumb_mcp",
"thumb_mcp": "thumb_ip",
reference_roll: reference_pitch,
reference_pitch: reference_pip,
reference_pip: reference_dip,
}
fixed_direct_zero_offsets = {
"thumb_mcp": 0.0,
reference_roll: 0.0,
reference_pitch: 0.0,
reference_pip: 0.0,
}
return ZeroCalibrationProfile(
hand=hand,
direct_zero_joints=direct,
axis_joints=(
"thumb_cmc_roll",
"thumb_cmc_yaw",
"thumb_cmc_pitch",
"thumb_mcp",
"thumb_ip",
reference_roll,
reference_pitch,
reference_pip,
reference_dip,
),
inherited_zero_joints=inherited,
inherited_static_zero_joints={},
constrained_circle_joints=(
hand.image_trajectory_joints | {reference_dip, "thumb_cmc_yaw"}
),
root_anchor_joints=frozenset({"thumb_cmc_roll", reference_roll}),
axis_parent_joint=axis_parent,
phase_parent_joint=phase_parent,
offset_observer_joint=observer,
same_view_axis_pair_by_offset={},
fixed_direct_zero_offsets_rad=fixed_direct_zero_offsets,
static_output_zero_offsets_rad={},
)
def _build_right_19_zero_profile(
hand: HandCalibrationProfile,
) -> ZeroCalibrationProfile:
"""Return the 16-active-zero graph for the 19-Tag layout.
Distal Tags recover all four PIP/DIP dynamic curves and axis quality, but
repeated motion with one unchanged camera/Tag installation cannot by
itself distinguish a fixed parallel-axis phase bias from an encoder zero.
The three coupled thumb-CMC offsets must therefore remain a visual
multi-axis solve; their electrical endpoints are not assumed to coincide
with source-CAD limits. Independently repeatable contact endpoints still
anchor thumb MCP, finger MCP-pitch and PIP from each session. DIP stays
passive and has no independently identifiable static zero.
"""
fingers = ("index", "middle", "ring", "pinky")
direct = (
"thumb_cmc_roll",
"thumb_cmc_yaw",
"thumb_cmc_pitch",
"thumb_mcp",
*(
f"{finger}_{suffix}"
for finger in fingers
for suffix in ("mcp_roll", "mcp_pitch", "pip")
),
)
axis_joints = (
"thumb_cmc_roll",
"thumb_cmc_yaw",
"thumb_cmc_pitch",
"thumb_mcp",
"thumb_ip",
*(
f"{finger}_{suffix}"
for finger in fingers
for suffix in ("mcp_roll", "mcp_pitch", "pip", "dip")
),
)
axis_parent = {
"thumb_cmc_yaw": "thumb_cmc_roll",
"thumb_cmc_pitch": "thumb_cmc_yaw",
**{
f"{finger}_mcp_pitch": f"{finger}_mcp_roll"
for finger in fingers
},
}
phase_parent = {
"thumb_mcp": "thumb_cmc_pitch",
"thumb_ip": "thumb_mcp",
**{
f"{finger}_{child}": f"{finger}_{parent}"
for finger in fingers
for child, parent in (
("pip", "mcp_pitch"),
("dip", "pip"),
)
},
}
observer = {
"thumb_cmc_roll": "thumb_cmc_yaw",
"thumb_cmc_yaw": "thumb_cmc_pitch",
"thumb_cmc_pitch": "thumb_mcp",
"thumb_mcp": "thumb_ip",
**{
f"{finger}_{target}": f"{finger}_{observed}"
for finger in fingers
for target, observed in (
("mcp_roll", "mcp_pitch"),
("mcp_pitch", "pip"),
("pip", "dip"),
)
},
}
root_anchors = frozenset(
{"thumb_cmc_roll", *(f"{finger}_mcp_roll" for finger in fingers)}
)
constrained = frozenset(
set(hand.image_trajectory_joints)
| {"thumb_cmc_yaw"}
# The side camera observes all four flexion axes nearly end-on. A
# 16 mm monocular planar Tag gives a very repeatable orientation axis,
# while its command-correlated depth bias can tilt the otherwise clean
# centre-trajectory plane by several degrees. Use the orientation
# axis to constrain that circle; PIP retains its independently fitted
# axis line and all final SE(3)/holdout quality gates.
| set(RIGHT_19_END_ON_IMAGE_CURVE_JOINTS)
# _fit_axis_measurement_raw constrains every passive DIP direction to
# its source-URDF-parallel upstream PIP. Declare the same fact here
# so validation does not compare that trusted direction with the
# depth-biased free plane of a distal Tag trajectory.
| set(RIGHT_19_VISUALLY_MEASURED_PASSIVE_DIPS)
)
return ZeroCalibrationProfile(
hand=hand,
direct_zero_joints=tuple(direct),
axis_joints=tuple(axis_joints),
inherited_zero_joints={},
inherited_static_zero_joints={},
constrained_circle_joints=constrained,
root_anchor_joints=root_anchors,
axis_parent_joint=axis_parent,
phase_parent_joint=phase_parent,
offset_observer_joint=observer,
same_view_axis_pair_by_offset={
"thumb_cmc_yaw": ("thumb_cmc_roll", "thumb_cmc_pitch")
},
# Endpoint-observable offsets are supplied by the caller. CMC remains
# visually solved and no serial-specific zero is hidden in the shared
# profile.
fixed_direct_zero_offsets_rad={},
static_output_zero_offsets_rad={},
)
def get_right_19_thumb_zero_profile() -> ZeroCalibrationProfile:
"""Return the independent four-zero thumb observer graph.
Roll and MCP are supplied as mechanical endpoint datums by the caller;
yaw is observed by the same-view roll/pitch axis pair and pitch by the
pitch/MCP serial phase. No finger axis is part of this profile.
"""
full = get_zero_calibration_profile("right", G20_RIGHT_19_LAYOUT)
thumb_direct = tuple(
name for name in full.direct_zero_joints if name.startswith("thumb_")
)
thumb_axes = tuple(
name for name in full.axis_joints if name.startswith("thumb_")
)
return replace(
full,
direct_zero_joints=thumb_direct,
axis_joints=thumb_axes,
constrained_circle_joints=frozenset(
name
for name in full.constrained_circle_joints
if name.startswith("thumb_")
),
root_anchor_joints=frozenset({"thumb_cmc_roll"}),
axis_parent_joint={
child: parent
for child, parent in full.axis_parent_joint.items()
if child.startswith("thumb_") and parent.startswith("thumb_")
},
phase_parent_joint={
child: parent
for child, parent in full.phase_parent_joint.items()
if child.startswith("thumb_") and parent.startswith("thumb_")
},
offset_observer_joint={
target: observer
for target, observer in full.offset_observer_joint.items()
if target.startswith("thumb_") and observer.startswith("thumb_")
},
same_view_axis_pair_by_offset={
target: pair
for target, pair in full.same_view_axis_pair_by_offset.items()
if target.startswith("thumb_")
},
fixed_direct_zero_offsets_rad={},
static_output_zero_offsets_rad={},
base_pose_strategy="thumb_serial",
)
RIGHT_19_MECHANICAL_ENDPOINT_JOINTS = frozenset(
{
"thumb_mcp",
*(
f"{finger}_{suffix}"
for finger in ("index", "middle", "ring", "pinky")
for suffix in ("mcp_pitch", "pip")
),
}
)
# Thumb CMC roll has a directly observed actuator-to-actuator travel, but its
# electrical stop is not used as a runtime URDF limit. Keep it out of
# RIGHT_19_MECHANICAL_ENDPOINT_JOINTS (whose offsets also move upper limits),
# and use its endpoint estimate only as the final static roll origin. This
# makes roll independent of the fitted palm-frame phase and of all yaw logic.
RIGHT_19_ROLL_ENDPOINT_ZERO_JOINTS = frozenset({"thumb_cmc_roll"})
RIGHT_19_ENDPOINT_MEASUREMENT_JOINTS = (
RIGHT_19_MECHANICAL_ENDPOINT_JOINTS
| RIGHT_19_ROLL_ENDPOINT_ZERO_JOINTS
)
RIGHT_19_POST_SOLVE_ENDPOINT_JOINTS = RIGHT_19_ROLL_ENDPOINT_ZERO_JOINTS
def anchor_right_19_mechanical_endpoint_curves(
curves: Mapping[str, JointCurveFit],
records_by_joint: Mapping[str, Sequence[Mapping[str, Any]]],
*,
maximum_direction_difference_rad: float = math.radians(1.0),
maximum_curve_correction_rad: float = math.radians(3.0),
feedback_endpoint_joints: frozenset[str] = frozenset(),
endpoint_joints: frozenset[str] | None = None,
) -> dict[str, JointCurveFit]:
"""Set mechanical-endpoint curve scale from direct SO(3) travel.
A fitted-axis projection is useful for a dense, signed command curve, but
a small view-dependent orthogonal PnP component can change that
projection's full-scale value. The rotation magnitude between the two
settled endpoint poses is the revolute-joint travel itself: conjugating
both poses by a rigid hand/camera transform or either fixed Tag mounting
rotation cannot change it.
Use every supplied cycle and both sweep directions as independent endpoint
measurements, reject disagreement, and apply their robust median as one
scale correction to the already validated curve. Dense CMC-roll records
are selected in feedback coordinates; settled mechanical-contact records
remain selected in requested-command coordinates.
"""
if (
not math.isfinite(maximum_direction_difference_rad)
or maximum_direction_difference_rad <= 0.0
):
raise ValueError("endpoint direction difference limit must be positive")
if (
not math.isfinite(maximum_curve_correction_rad)
or maximum_curve_correction_rad <= 0.0
):
raise ValueError("endpoint curve correction limit must be positive")
result = dict(curves)
selected_endpoint_joints = (
RIGHT_19_ENDPOINT_MEASUREMENT_JOINTS
if endpoint_joints is None
else frozenset(str(name) for name in endpoint_joints)
)
unknown_endpoint_joints = sorted(
set(selected_endpoint_joints) - set(RIGHT_19_ENDPOINT_MEASUREMENT_JOINTS)
)
if unknown_endpoint_joints:
raise ValueError(
"endpoint joints are not endpoint measurements: "
+ ",".join(unknown_endpoint_joints)
)
unknown_feedback_joints = sorted(
set(feedback_endpoint_joints) - set(RIGHT_19_ENDPOINT_MEASUREMENT_JOINTS)
)
if unknown_feedback_joints:
raise ValueError(
"feedback endpoint joints are not endpoint measurements: "
+ ",".join(unknown_feedback_joints)
)
for name in sorted(selected_endpoint_joints):
fit = curves.get(name)
records = list(records_by_joint.get(name, ()))
if fit is None:
raise ValueError(f"missing measured endpoint curve for {name}")
direction_travel: dict[str, float] = {}
all_travel: list[float] = []
for direction in ("decreasing", "increasing"):
cycles = sorted(
{
int(record.get("cycle", 0))
for record in records
if str(record.get("direction")) == direction
}
)
branch_travel: list[float] = []
for cycle in cycles:
endpoint_rotations: dict[int, Rotation] = {}
for command in (0, 255):
quaternions = []
for record in records:
if (
str(record.get("direction")) != direction
or int(record.get("cycle", 0)) != cycle
):
continue
endpoint_value = (
explicit_domain_value(record, "feedback")
if name in feedback_endpoint_joints
else record.get(
"requested_command_u8",
record.get("command_u8", -1),
)
)
if endpoint_value is None:
continue
if int(round(float(endpoint_value))) == command:
quaternions.append(_relative_rotation(record))
if not quaternions:
raise ValueError(
f"{name} {direction} cycle {cycle} is missing "
f"settled endpoint {command}"
)
endpoint_rotations[command] = Rotation.from_quat(
robust_rotation_summary(quaternions)[0]
)
branch_travel.append(
float(
(
endpoint_rotations[255].inv()
* endpoint_rotations[0]
).magnitude()
)
)
if not branch_travel:
raise ValueError(f"{name} is missing {direction} endpoint travel")
direction_travel[direction] = float(np.median(branch_travel))
all_travel.extend(branch_travel)
travel_values = np.asarray(list(direction_travel.values()), dtype=float)
direction_difference = float(np.ptp(travel_values))
if direction_difference > maximum_direction_difference_rad:
raise ValueError(
f"{name} settled endpoint directions disagree: "
f"{math.degrees(direction_difference):.3f}deg"
)
cycle_range = float(np.ptp(np.asarray(all_travel, dtype=float)))
if cycle_range > maximum_direction_difference_rad:
raise ValueError(
f"{name} settled endpoint cycles disagree: "
f"{math.degrees(cycle_range):.3f}deg"
)
direct_travel = float(np.median(np.asarray(all_travel, dtype=float)))
curve = np.asarray(fit.angle_rad, dtype=float)
if curve.shape != (256,) or not np.all(np.isfinite(curve)):
raise ValueError(f"{name} endpoint curve must contain 256 finite bins")
projected_travel = float(curve[0] - curve[255])
if direct_travel <= 0.0 or projected_travel <= 0.0:
raise ValueError(f"{name} endpoint travel must be positive")
correction = abs(projected_travel - direct_travel)
if correction > maximum_curve_correction_rad:
raise ValueError(
f"{name} fitted/direct endpoint travel differs by "
f"{math.degrees(correction):.3f}deg"
)
scale = direct_travel / projected_travel
def scaled(values: Sequence[float]) -> tuple[float, ...]:
return tuple(float(value) * scale for value in values)
circle = dict(fit.circle)
circle.update(
{
"mechanical_endpoint_direct_travel_rad": direct_travel,
"mechanical_endpoint_direction_difference_rad": (
direction_difference
),
"mechanical_endpoint_cycle_range_rad": cycle_range,
"mechanical_endpoint_sample_count": len(all_travel),
"mechanical_endpoint_raw_curve_travel_rad": projected_travel,
"mechanical_endpoint_curve_scale": scale,
}
)
quality = dict(fit.quality)
if "arc_rad" in quality:
quality["arc_rad"] = float(quality["arc_rad"]) * scale
result[name] = replace(
fit,
angle_rad=scaled(fit.angle_rad),
decreasing_rad=scaled(fit.decreasing_rad),
increasing_rad=scaled(fit.increasing_rad),
circle=circle,
maximum_monotonic_correction_rad=(
float(fit.maximum_monotonic_correction_rad) * scale
),
maximum_hysteresis_rad=float(fit.maximum_hysteresis_rad) * scale,
quality=quality,
zero_offset_rad=float(fit.zero_offset_rad) * scale,
)
return result
def derive_right_19_mechanical_endpoint_offsets(
source_urdf: str | Path,
curves: Mapping[str, JointCurveFit],
*,
maximum_offset_rad: float = math.radians(5.0),
endpoint_joints: frozenset[str] | None = None,
) -> dict[str, float]:
"""Estimate encoder origins from repeatable measured mechanical endpoints.
For the four fingers feedback 0 is the independently verified palm-contact
endpoint; thumb MCP uses its independently verified actuator endpoint.
The source URDF upper limit describes those same physical endpoints.
Therefore ``origin_offset + measured_travel == CAD_upper``.
Thumb CMC pitch/yaw remain excluded because their endpoint-to-CAD contract
is not independently established. CMC roll is included as a post-solve
origin only: its measured travel determines roll without allowing the
palm common-direction fit to write that joint.
The returned corrections are measured, generally non-zero encoder zeros;
they do not retain the CAD origin. Using full relative rotation travel
makes them invariant to rigid hand movement and arbitrary fixed Tag
mounting rotation, unlike a cross-camera absolute planar-PnP phase.
"""
if not math.isfinite(maximum_offset_rad) or maximum_offset_rad <= 0.0:
raise ValueError("mechanical endpoint maximum offset must be positive")
root = ET.parse(Path(source_urdf).expanduser().resolve()).getroot()
joints = {str(node.get("name")): node for node in root.findall("joint")}
result: dict[str, float] = {}
selected_endpoint_joints = (
RIGHT_19_ENDPOINT_MEASUREMENT_JOINTS
if endpoint_joints is None
else frozenset(str(name) for name in endpoint_joints)
)
unknown_endpoint_joints = sorted(
set(selected_endpoint_joints) - set(RIGHT_19_ENDPOINT_MEASUREMENT_JOINTS)
)
if unknown_endpoint_joints:
raise ValueError(
"endpoint joints are not endpoint measurements: "
+ ",".join(unknown_endpoint_joints)
)
for name in sorted(selected_endpoint_joints):
fit = curves.get(name)
joint = joints.get(name)
limit = None if joint is None else joint.find("limit")
if fit is None:
raise ValueError(f"missing measured endpoint curve for {name}")
if limit is None or limit.get("upper") is None:
raise ValueError(f"source URDF joint {name} has no upper limit")
upper = float(limit.get("upper"))
curve = np.asarray(fit.angle_rad, dtype=float)
if curve.shape != (256,) or not np.all(np.isfinite(curve)):
raise ValueError(f"{name} endpoint curve must contain 256 finite bins")
measured_endpoint = float(curve[0])
if measured_endpoint <= 0.0 or abs(float(curve[255])) > math.radians(0.05):
raise ValueError(
f"{name} endpoint curve does not use command 255 as its zero reference"
)
if abs(measured_endpoint - float(np.max(curve))) > math.radians(0.05):
raise ValueError(
f"{name} command 0 is not the measured upper mechanical endpoint"
)
offset = upper - measured_endpoint
if not math.isfinite(offset) or abs(offset) > maximum_offset_rad:
raise ValueError(
f"{name} endpoint-derived zero offset is outside the safe range: "
f"{math.degrees(offset):.3f}deg"
)
result[name] = offset
return result
def get_zero_calibration_profile(
side: str, layout_id: str = "legacy_11"
) -> ZeroCalibrationProfile:
return _build_zero_profile(get_hand_calibration_profile(side, layout_id))
LEFT_ZERO_PROFILE = _build_zero_profile(LEFT_HAND_PROFILE)
# Backwards-compatible left-hand aliases.
CONSTRAINED_CIRCLE_JOINTS = LEFT_ZERO_PROFILE.constrained_circle_joints
DIRECT_ZERO_JOINTS = LEFT_ZERO_PROFILE.direct_zero_joints
FIXED_DIRECT_ZERO_OFFSETS_RAD = dict(
LEFT_ZERO_PROFILE.fixed_direct_zero_offsets_rad
)
OPTIMIZED_ZERO_JOINTS = tuple(
name for name in DIRECT_ZERO_JOINTS
if name not in FIXED_DIRECT_ZERO_OFFSETS_RAD
)
AXIS_JOINTS = LEFT_ZERO_PROFILE.axis_joints
INHERITED_ZERO_JOINTS = dict(LEFT_ZERO_PROFILE.inherited_zero_joints)
def merge_right_19_thumb_zero_result(
*,
thumb_result: ZeroSolveResult,
preserved_offsets_rad: Mapping[str, float] | None = None,
companion_result: ZeroSolveResult | None = None,
) -> ZeroSolveResult:
"""Merge an independent thumb solve without allowing reverse coupling.
``companion_result`` is the finger/full solver result whose thumb values
were fixed to ``thumb_result``. For a thumb-only run, the 12 certified
finger values are supplied through ``preserved_offsets_rad`` instead.
"""
thumb_names = {
"thumb_cmc_roll",
"thumb_cmc_yaw",
"thumb_cmc_pitch",
"thumb_mcp",
}
if set(thumb_result.direct_offsets_rad) != thumb_names:
raise ValueError("independent thumb result must contain exactly four zeros")
full_profile = get_zero_calibration_profile("right", G20_RIGHT_19_LAYOUT)
direct_names = set(full_profile.direct_zero_joints)
preserved = {
str(name): float(value)
for name, value in (preserved_offsets_rad or {}).items()
}
if set(preserved) - (direct_names - thumb_names):
raise ValueError("preserved offsets contain a thumb or unknown joint")
if any(not math.isfinite(value) for value in preserved.values()):
raise ValueError("preserved offsets must be finite")
if companion_result is None:
if set(preserved) != direct_names - thumb_names:
raise ValueError("thumb-only merge requires all 12 certified finger zeros")
direct = dict(preserved)
all_active = {name: 0.0 for name in full_profile.hand.active_joints}
all_active.update(preserved)
base_result = thumb_result
else:
if set(companion_result.direct_offsets_rad) != direct_names:
raise ValueError("full companion result must contain all 16 zeros")
direct = dict(companion_result.direct_offsets_rad)
all_active = dict(companion_result.all_active_offsets_rad)
base_result = companion_result
direct.update(thumb_result.direct_offsets_rad)
all_active.update(thumb_result.direct_offsets_rad)
def merged_map(name: str) -> dict[str, Any]:
values = (
{}
if companion_result is None
else dict(getattr(companion_result, name))
)
values.update(dict(getattr(thumb_result, name)))
return values
failure_reasons = merged_map("failure_reasons")
passed = bool(
thumb_result.passed
and (companion_result is None or companion_result.passed)
)
return replace(
base_result,
direct_offsets_rad=direct,
all_active_offsets_rad=all_active,
validation_errors_rad=(
tuple(thumb_result.validation_errors_rad)
if companion_result is None
else (
*companion_result.validation_errors_rad,
*thumb_result.validation_errors_rad,
)
),
validation_error_by_joint_rad=merged_map(
"validation_error_by_joint_rad"
),
validation_line_error_by_joint_m=merged_map(
"validation_line_error_by_joint_m"
),
axis_line_rms_m=(
thumb_result.axis_line_rms_m
if companion_result is None
else max(
companion_result.axis_line_rms_m,
thumb_result.axis_line_rms_m,
)
),
passed=passed,
cycle_offsets_rad=merged_map("cycle_offsets_rad"),
offset_uncertainty_rad=merged_map("offset_uncertainty_rad"),
offset_confidence_half_width_rad=merged_map(
"offset_confidence_half_width_rad"
),
validation_original_error_by_joint_rad=merged_map(
"validation_original_error_by_joint_rad"
),
validation_improvement_by_joint_rad=merged_map(
"validation_improvement_by_joint_rad"
),
validation_improvement_confidence_lower_rad=merged_map(
"validation_improvement_confidence_lower_rad"
),
offset_covariance_rad2=merged_map("offset_covariance_rad2"),
axis_cone_mismatch_by_joint_rad=merged_map(
"axis_cone_mismatch_by_joint_rad"
),
axis_cone_bias_classification_by_joint=merged_map(
"axis_cone_bias_classification_by_joint"
),
failure_reasons=failure_reasons,
)
def expand_right_19_thumb_zero_result_with_cad_fingers(
thumb_result: ZeroSolveResult,
) -> ZeroSolveResult:
"""Build a standalone thumb result with every finger left at source CAD.
This is intentionally different from inheriting a prior finger
calibration: the twelve non-thumb values are explicit zeros and no finger
observation or artifact is required.
"""
full_profile = get_zero_calibration_profile("right", G20_RIGHT_19_LAYOUT)
thumb_names = {
"thumb_cmc_roll",
"thumb_cmc_yaw",
"thumb_cmc_pitch",
"thumb_mcp",
}
if set(thumb_result.direct_offsets_rad) != thumb_names:
raise ValueError("independent thumb result must contain exactly four zeros")
direct = {name: 0.0 for name in full_profile.direct_zero_joints}
direct.update(thumb_result.direct_offsets_rad)
all_active = {name: 0.0 for name in full_profile.hand.active_joints}
all_active.update(thumb_result.direct_offsets_rad)
return replace(
thumb_result,
direct_offsets_rad=direct,
all_active_offsets_rad=all_active,
)
def write_zero_corrected_urdf(
*,
source_urdf: str | Path,
output_directory: str | Path,
serial_number: str,
offsets_rad: Mapping[str, float],
endpoint_anchored_offsets_rad: Mapping[str, float] | None = None,
timestamp: str | None = None,
correction_plan: UrdfCorrectionPlan | None = None,
) -> Path:
source = Path(source_urdf).expanduser().resolve()
output = Path(output_directory).expanduser().resolve()
if not source.is_file():
raise ValueError(f"source URDF does not exist: {source}")
if not offsets_rad:
raise ValueError("offsets_rad must contain at least one joint")
offsets = {str(name): float(value) for name, value in offsets_rad.items()}
endpoint_offsets = {
str(name): float(value)
for name, value in dict(endpoint_anchored_offsets_rad or {}).items()
}
if not set(endpoint_offsets) <= set(offsets):
raise ValueError("endpoint-anchored offsets must be URDF zero targets")
if correction_plan is not None:
correction_plan.verify_source(source)
correction_plan.authorize_offsets(offsets)
if not set(endpoint_offsets).issubset(
correction_plan.endpoint_limit_joints
):
raise ValueError(
"endpoint offsets are not authorized by the correction plan"
)
if any(
not math.isfinite(value) or abs(value) > math.radians(90.0)
for value in offsets.values()
):
raise ValueError("URDF zero offsets must be finite and within +/-90deg")
stamp = timestamp or datetime.now().strftime("%Y%m%d_%H%M%S")
if re.fullmatch(r"\d{8}_\d{6}", stamp) is None:
raise ValueError("URDF zero timestamp must use YYYYMMDD_HHMMSS")
safe_serial = "".join(
character if character.isalnum() or character in "_.-" else "_"
for character in str(serial_number)
)
if not safe_serial:
raise ValueError("serial_number must not be empty")
from linkerhand_calibration.core.urdf.plan import build_standard_correction_plan
import hashlib
if "calibrated" in source.stem.lower():
raise ValueError("source must be the original CAD URDF")
# The deployed JSON keeps the compatibility shape; zero coordinates must
# nevertheless transform both limits and every dependent mimic together.
# Endpoint metadata cannot select a different coordinate rule.
if any(not math.isclose(value, offsets[name], rel_tol=0, abs_tol=1e-12)
for name, value in endpoint_offsets.items()):
raise ValueError("endpoint metadata disagrees with the actual zero correction")
if correction_plan is not None:
rights = correction_plan.authorized_fields
if not rights:
raise ValueError("production correction plan lacks independent Profile field authorization")
source_hash = correction_plan.source_sha256
else:
# Compatibility for standalone offline correction tools. The online
# product path always supplies its protected, scope-authorized plan.
model = UrdfKinematicModel(source)
rights = {name: ("origin.rpy", "limit.lower", "limit.upper") for name in offsets}
for name, joint in model.joints.items():
if joint.mimic_joint is not None:
rights[name] = (*rights.get(name, ()), "mimic.offset")
source_hash = hashlib.sha256(source.read_bytes()).hexdigest()
plan = build_standard_correction_plan(source_urdf=source,
source_sha256=source_hash, zero_offsets_rad=offsets, authorized_fields=rights)
destination = output / f"{source.stem}_zero_calibrated_{safe_serial}_{stamp}.urdf"
return plan.write(source, destination)

Some files were not shown because too many files have changed in this diff Show More