Sparse-reward manipulation on Hammer Cleanup simulated environment
100Success RateCSIL++ Ens.
Evaluation Results
| Method | Links | |
|---|---|---|
| CSIL++ Ens.Policy architecture=Ensemble, Learning step=50k2026.06 | 100 | |
| CSIL++ Res.Policy architecture=Residual, Learning step=0k2026.06 | 100 | |
| XQC+OD Ens.Policy architecture=Ensemble, Learning step=0k2026.06 | 100 | |
| XQC+OD Res.Policy architecture=Residual, Learning step=0k2026.06 | 100 | |
| VLA2026.06 | 98 | |
| PLD RL (Res.)Policy architecture=Residual, Learning step=300k2026.06 | 96 |