Files
SkillFactory/tests/test_standing_force_visualizer.py
T
chenlin 9fe9d27865 feat: checkpoint contact-cycle evaluation and RootXY PPO progress
Consolidate next-stage docs, archive superseded narratives, and preserve frozen document inputs. Retain invalid P2 pairs, partial RootXY training, failed attempts and nonselectable checkpoints as explicit limitations.

Include supervised evaluation/training, desktop GPU admission, bounded evidence and atomic-publication monitor repairs with CPU regressions. Ignore local dotenv configuration.

Validation: 1403 passed, 6 skipped for contact-cycle and related contracts/CLI tests. Full-suite legacy ABI and registration tests require Python 3.12/Torch 2.10/pxr; no runtime checks relaxed. No new physical run or accepted skill.
2026-10-15 14:18:17 +08:00

118 lines
4.4 KiB
Python

"""No-Kit geometry and read-only applied-force visualization regressions."""
import sys
from pathlib import Path
from types import SimpleNamespace
import pytest
import torch
sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "scripts"))
import standing_force_visualizer as viz
def rotated_z(q):
z = torch.zeros_like(q[:, :3])
z[:, 2] = 1
t = 2 * torch.cross(q[:, :3], z, dim=-1)
return z + q[:, 3:] * t + torch.cross(q[:, :3], t, dim=-1)
@pytest.mark.parametrize("dtype", [torch.float32, torch.float64])
def test_arrow_world_tail_tip_and_xyzw_all_directions(dtype):
force = torch.tensor(
[
[5, 0, 0],
[-5, 0, 0],
[0, 5, 0],
[0, -5, 0],
[0, 0, 10],
[0, 0, -10],
[2, -3, 4],
[0.001, -0.002, -10],
[1e-5, 0, 0],
],
dtype=dtype,
)
origin = torch.arange(len(force) * 3, dtype=dtype).reshape(-1, 3) / 10
before = origin.clone(), force.clone()
g = viz.arrow_geometry(origin, force, 0.03)
n = len(force)
direction = force / force.norm(dim=-1, keepdim=True)
assert g["translations"].shape == (2 * n, 3)
assert torch.allclose(rotated_z(g["orientations"][:n]), direction, atol=2e-7)
assert torch.allclose(g["orientations"].norm(dim=-1), torch.ones(2 * n, dtype=dtype))
assert torch.allclose(g["translations"][:n] - direction * g["scales"][:n, 2:] / 2, origin, atol=1e-6)
assert torch.allclose(g["translations"][n:] + direction * g["scales"][n:, 2:] / 2, origin + 0.03 * force, atol=1e-6)
assert g["marker_indices"].tolist() == [0] * n + [1] * n
assert torch.equal(origin, before[0]) and torch.equal(force, before[1])
def test_zero_force_and_zero_batch_have_no_visible_geometry():
for count in (0, 5):
g = viz.arrow_geometry(torch.zeros(count, 3), torch.zeros(count, 3))
assert g["translations"].shape == (0, 3)
assert g["orientations"].shape == (0, 4)
assert all(torch.isfinite(v).all() for v in g.values())
@pytest.mark.parametrize("scale", [0, -0.01, float("nan"), float("inf")])
def test_bad_scale_rejected_before_kit_import(scale):
with pytest.raises(ValueError, match="scale"):
viz.DisturbanceForceArrows(scale)
with pytest.raises(ValueError, match="scale"):
viz.arrow_geometry(torch.zeros(1, 3), torch.ones(1, 3), scale)
@pytest.mark.parametrize(
"force", [torch.ones(1, 6), torch.ones(2, 3), torch.full((1, 3), float("nan")), torch.ones(1, 3).long()]
)
def test_bad_force_rejected(force):
with pytest.raises(ValueError):
viz.arrow_geometry(torch.zeros(1, 3), force)
def test_applied_disturbance_only_and_visibility_lifecycle(monkeypatch):
class Markers:
def __init__(self):
self.visible = True
self.frames = []
def set_visibility(self, value):
self.visible = value
def visualize(self, **kwargs):
self.frames.append(kwargs)
markers = Markers()
monkeypatch.setattr(viz, "_make_markers", lambda: markers)
arrows = viz.DisturbanceForceArrows()
assert not markers.visible
position = torch.tensor([[12.0, -3.0, 0.4], [-8.0, 4.0, 0.5]])
# First environment is pure torque: must not be interpreted as a force.
wrench = torch.tensor([[0.0, 0.0, 0.0, 1.0, 2.0, 3.0], [0.0, 0.0, -10.0, 0.0, 0.0, 0.0]])
state = SimpleNamespace(
root=lambda: (position,),
last_disturbance=wrench,
last_guide=torch.ones(2, 6) * 500,
mount_wrench=lambda: (_ for _ in ()).throw(AssertionError("Not sensor data")),
)
before = position.clone(), wrench.clone(), state.last_guide.clone(), torch.get_rng_state().clone()
arrows.update(state)
assert markers.visible and len(markers.frames) == 1
g = markers.frames[0]
assert len(g["translations"]) == 2 # one shaft + one head, not two arrows
assert torch.allclose(
g["translations"][1] + rotated_z(g["orientations"])[1] * g["scales"][1, 2] / 2,
position[1] + torch.tensor([0.0, 0.0, -0.3]),
)
assert all(torch.equal(a, b) for a, b in zip(before, (position, wrench, state.last_guide, torch.get_rng_state())))
wrench.zero_()
arrows.update(state)
assert not markers.visible and len(markers.frames) == 1 # no stale arrow
wrench[0, 0] = 5
arrows.update(state)
assert markers.visible and len(markers.frames) == 2
arrows.close()
assert not markers.visible