Stage alignment
Each stage-level environment uses the matching physical success predicate from the long task.
Industrial · 3 stages
Nut & Bolt
Stand a bolt upright, seat a nut on its axis, then rotate the nut counterclockwise through a full half turn.
Stage 01 / 03Long environment
Screwing-v1EmbodimentBimanual Sharpa
Observation162 state dims
Action58 targets
Horizon2400 steps
Stage definitions
Exact slices of the parent sequence.
Canonical 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 | Place boltScrewingBoltPickPlace-v1 | Left | 29 | 700 | Stand the bolt upright within the target region, release it, and hold stable for three steps. |
| 02 | Seat nutScrewingNutPickPlace-v1 | Right | 29 | 800 | Align the nut within 5 mm of the bolt axis and seat it within the height and tilt tolerances. |
| 03 | Turn nutScrewingTurn-v1 | Right | 29 | 900 | Accumulate at least 180° of net counterclockwise rotation while the nut remains seated. |
Paper results
Nut & Bolt in the benchmark.
Stage success in the training scene (%, 50 episodes). Best scores in each row are highlighted.
| Stage | PPO | RLPD | ACT | Diffusion Policy | GR00T | π0.5 |
|---|---|---|---|---|---|---|
| PickBolt | 0.0 | 28.0 | 8.0 | 22.0 | 44.0 | 12.0 |
| SeatNut | 0.0 | 34.0 | 12.0 | 6.0 | 26.0 | 2.0 |
| TurnNut | 76.0 | 100.0 | 96.0 | 100.0 | 100.0 | 100.0 |
Cross-scene results · mean ± scene standard deviation
| Stage | PPO | RLPD | ACT | Diffusion Policy | GR00T | π0.5 |
|---|---|---|---|---|---|---|
| PickBolt | 0.7 ± 1.2 | 0.0 ± 0.0 | 8.7 ± 5.0 | 2.0 ± 3.5 | 21.3 ± 13.3 | 18.0 ± 2.0 |
| SeatNut | 0.0 ± 0.0 | 0.0 ± 0.0 | 2.7 ± 2.3 | 0.7 ± 1.2 | 2.0 ± 3.5 | 3.3 ± 2.3 |
| TurnNut | 26.7 ± 11.0 | 11.3 ± 16.3 | 44.0 ± 19.7 | 11.3 ± 12.7 | 80.0 ± 19.7 | 87.3 ± 5.0 |
Complete sequence · end-to-end success (%)
| Task | PPO | RLPD | ACT | Diffusion Policy | GR00T | π0.5 |
|---|---|---|---|---|---|---|
| Nut and bolt | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
Source: manuscript experiment tables. Compare all methods and tasks →
Shared contractSame workstation.
Same workstation.
Same predicates.
Persistence
Hold counters or latches prevent unstable contact from being counted as task completion.
Replay
State and action traces can be restored and replayed for validation and offline rendering.