Stage alignment
Each stage-level environment uses the matching physical success predicate from the long task.
Open a refrigerator, move an apple into a blender, close the door, place the lid, and press the power button.
Stage 01 / 05DualSharpaPrepareJuice-v1Canonical instructions, handedness, action width, horizon, and success conditions are drawn from the current task registry and success-contract documentation.
| # | Stage / environment | Hand | Action | Max steps | Success summary |
|---|---|---|---|---|---|
| 01 | Open refrigeratorSharpaOpenFridge-v1 | Left | 29 | 650 | Open the door to at least 75° at low angular speed and hold for six steps. |
| 02 | Move appleSharpaTakeAppleToBlender-v1 | Right | 29 | 700 | Lift the apple out of the refrigerator and settle it inside the blender for 12 steps. |
| 03 | Close refrigeratorSharpaCloseFridge-v1 | Left | 29 | 650 | Close the door to within 1° at low angular speed and hold for six steps. |
| 04 | Place blender lidSharpaMobilePlaceBlenderLid-v1 | Right | 29 | 400 | Lift and seat the lid within 5 mm xy, 8 mm z and 15° tilt for 15 steps. |
| 05 | Press buttonSharpaMobilePressBlenderButton-v1 | Right | 29 | 400 | Depress the 2 mm power button by at least 80% for four steps. |
Stage success in the training scene (%, 50 episodes). Best scores in each row are highlighted.
| Stage | PPO | RLPD | ACT | Diffusion Policy | GR00T | π0.5 |
|---|---|---|---|---|---|---|
| OpenFridge | 90.0 | 98.0 | 100.0 | 88.0 | 94.0 | 96.0 |
| AppleToBlender | 0.0 | 0.0 | 92.0 | 42.0 | 54.0 | 38.0 |
| CloseFridge | 100.0 | 100.0 | 100.0 | 100.0 | 98.0 | 100.0 |
| PlaceLid | 0.0 | 78.0 | 26.0 | 26.0 | 42.0 | 16.0 |
| PressButton | 26.0 | 100.0 | 66.0 | 72.0 | 64.0 | 68.0 |
| Stage | PPO | RLPD | ACT | Diffusion Policy | GR00T | π0.5 |
|---|---|---|---|---|---|---|
| OpenFridge | 65.3 ± 22.0 | 78.7 ± 35.2 | 92.7 ± 9.5 | 68.0 ± 33.3 | 70.0 ± 45.1 | 96.0 ± 3.5 |
| AppleToBlender | 0.7 ± 1.2 | 0.0 ± 0.0 | 71.3 ± 14.2 | 7.3 ± 12.7 | 18.0 ± 2.0 | 38.7 ± 4.2 |
| CloseFridge | 100.0 ± 0.0 | 100.0 ± 0.0 | 99.3 ± 1.2 | 66.0 ± 47.3 | 92.0 ± 8.7 | 100.0 ± 0.0 |
| PlaceLid | 0.0 ± 0.0 | 0.0 ± 0.0 | 18.0 ± 11.1 | 1.3 ± 2.3 | 28.7 ± 12.9 | 16.0 ± 2.0 |
| PressButton | 18.0 ± 5.3 | 18.7 ± 32.3 | 54.0 ± 19.3 | 52.7 ± 40.6 | 42.7 ± 10.1 | 62.0 ± 7.2 |
| Task | PPO | RLPD | ACT | Diffusion Policy | GR00T | π0.5 |
|---|---|---|---|---|---|---|
| Juice preparation | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
Source: manuscript experiment tables. Compare all methods and tasks →
Each stage-level environment uses the matching physical success predicate from the long task.
Hold counters or latches prevent unstable contact from being counted as task completion.
State and action traces can be restored and replayed for validation and offline rendering.