Robotic Manipulation on Pour manipulation task
100Success Rate (SR)pi_0
Evaluation results
| Method | Links | ||
|---|---|---|---|
| pi_0Checkpoint=20K, Base Policy=pi_02026.06 | 100 | — | |
| G3VLA (pi_0 + GT)Checkpoint=20K, Base Policy=pi_02026.06 | 100 | — | |
| pi_0Checkpoint=25K, Base Policy=pi_02026.06 | 100 | — | |
| G3VLA (pi_0 + GT)Checkpoint=25K, Base Policy=pi_02026.06 | 100 | — | |
| pi_0.5Checkpoint=25K, Base Policy=pi_0.52026.06 | 100 | — | |
| pi_0.5Checkpoint=20K, Base Policy=pi_0.52026.06 | 93.75 | — | |
| G3VLA (pi_0.5 + GT)Checkpoint=20K, Base Policy=pi_0.52026.06 | 93.75 | — | |
| G3VLA (pi_0.5 + GT)Checkpoint=25K, Base Policy=pi_0.52026.06 | 87.5 | — | |
| HATS (Hybrid) + DPPolicy=DP [38], Setting=Ours (Hyb.)2026.06 | 70 | 94.6 | |
| HATS (Hybrid) + DP3Policy=DP3 [39], Setting=Ours (Hyb.)2026.06 | 70 | 94.6 | |
| HATS (Centralized) + DPPolicy=DP [38], Setting=Ours (Cen.)2026.06 | 5 | 76.3 | |
| HATS (Centralized) + DP3Policy=DP3 [39], Setting=Ours (Cen.)2026.06 | 5 | 76.3 | |
| Diffusion Policy (DP)Setting=Full-Agent2026.06 | 0 | 0 | |
| Diffusion Policy (DP)Setting=Dual-Human2026.06 | 0 | 73.1 | |
| 3D Diffusion Policy (DP3)Setting=Full-Agent2026.06 | 0 | 0 | |
| 3D Diffusion Policy (DP3)Setting=Dual-Human2026.06 | 0 | 73.1 |