Robot Manipulation Trajectory Refinement on 7 Simulated Objects Average
91.4Success RateVideo2Sim2Real
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Video2Sim2Real2026.06 | 91.4 | 100 | 3.7 | 5.7 | |
| Residual RLReward type=Flow-tracking plus auxiliary rewards2026.06 | 52.9 | 29 | 58.6 | 55.4 | |
| Residual RLReward type=Flow-tracking reward2026.06 | 49.5 | 42.9 | 39.9 | 38.2 | |
| DeepMimic RLReward type=Flow-tracking plus auxiliary rewards2026.06 | 24.3 | 56.7 | 29.8 | 63.8 | |
| DeepMimic RLReward type=Flow-tracking reward2026.06 | 19.5 | 54.8 | 24.6 | 87.4 | |
| Pre-Contact InitReward type=Flow-tracking plus auxiliary rewards2026.06 | 19.5 | 14.3 | 169 | 42 | |
| Object-onlyReward type=Flow-tracking plus auxiliary rewards2026.06 | 15.7 | 17.6 | 523 | 19.4 | |
| Opt-Pre-Contact InitReward type=Flow-tracking plus auxiliary rewards2026.06 | 14.3 | 16.7 | 300.6 | 28.6 | |
| Object-onlyReward type=Flow-tracking reward2026.06 | 9.5 | 13.8 | 1,153 | 13.7 | |
| Pre-Contact InitReward type=Flow-tracking reward2026.06 | 6.2 | 37.1 | 143.6 | 25.4 | |
| Opt-Pre-Contact InitReward type=Flow-tracking reward2026.06 | 1.9 | 33.3 | 321.9 | 22.5 |