| { |
| "version": 2, |
| "source": "observation.state; actual recorded state, not action", |
| "right_joint_indices_zero_based": [ |
| 15, |
| 16, |
| 17, |
| 18, |
| 19, |
| 20, |
| 21 |
| ], |
| "right_gripper_index_zero_based": 29, |
| "joint_units": "rad", |
| "joint_weights": "equal; no coordinate normalization", |
| "episode_scope": "each atomic skill episode separately; chronological frame_index", |
| "path_formula": "L_t=sum(k=1..t, ||q_k-q_(k-1)||_2); progress_arm_path=L_t/L_end", |
| "event_skills": [ |
| "grasp", |
| "place", |
| "stack" |
| ], |
| "event_rule": "exactly one binary switch; event=0 before switch, 1 from switch frame", |
| "combination": "grasp/place/stack: progress_method=0.8*progress_arm_method+0.2*event; other skills: progress_method=progress_arm_method", |
| "no_event_rule": "constant gripper; gripper_event_complete=0; arm weight=1", |
| "degenerate_threshold_rad": 0.01, |
| "degenerate_policy": "reject total cumulative path below threshold; no silent time fallback", |
| "smoothing": "none; cumulative progress is nondecreasing by construction", |
| "computation_dtype": "float64", |
| "frame_fields": { |
| "progress_arm_path": "float32", |
| "gripper_event_complete": "int64", |
| "progress_path": "float32" |
| }, |
| "observation_state": "original 30-dimensional state preserved; labels are independent columns", |
| "training_target": "progress_path" |
| } |
|
|
