# Configuration for DexMachina (Genesis) dynamics retargeting pipeline
defaults:
  - _self_

# Dataset configuration (not used by DexMachina, but kept for compatibility)
dataset_dir: "example_datasets"
dataset_name: "arctic"
robot_type: "inspire_hand"  # "inspire_hand", "xhand", "dex3_hand"
embodiment_type: "bimanual"
task: "box-30-230"  # format: "object_name-start_frame-end_frame"
data_id: 0
viewer: "gs"
rerun_spawn: false
device: "cuda:0"

# Simulation and control parameters
simulator: dexmachina
sim_dt: 0.02  # Genesis simulation timestep
ctrl_dt: 0.20  # Control timestep (how often MPC updates)
knot_dt: 0.20  # Knot point spacing
ref_dt: 0.02  # Reference trajectory timestep
horizon: 0.40  # MPC horizon in seconds
show_viewer: true
save_video: true
save_info: true
max_sim_steps: -1  # -1 means use full trajectory length
nu: 36

# MPC optimization parameters
max_num_iterations: 32  # Number of iterations per MPC step
num_samples: 1024  # Number of parallel samples for CEM
last_ctrl_noise_scale: 1.0
first_ctrl_noise_scale: 0.5
final_noise_scale: 0.01
pos_noise_scale: 1.0
rot_noise_scale: 1.0
joint_noise_scale: 1.0
improvement_threshold: 0.01
improvement_check_steps: 4
terminal_rew_scale: 10.0
temperature: 1.0

# Reward mode
use_rl_reward: true  # true to use dexmachina RL training reward formulation

# Domain randomization (curriculum learning)
num_dr: 1  # Number of domain randomization settings per iteration

# Output configuration
output_dir: "outputs/dexmachina"