Robot Manipulation on RoboLab 120 (Overall)
20.6Vague Success RateCosmos3-Nano-Policy-DROID
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Cosmos3-Nano-Policy-DROIDCheckpoint type=off-the-shelf, Training mode=fine-tuned on DROID dataset2026.06 | 20.6 | 36.8 | 39.7 | |
| Cosmos3-Nano (PT-init)Checkpoint type=off-the-shelf, Training mode=fine-tuned on DROID dataset2026.06 | 16.7 | 28.1 | 30.2 | |
| π0.5Checkpoint type=off-the-shelf, Training mode=fine-tuned on DROID dataset2026.06 | 15.2 | 28 | 28.1 | |
| DreamZeroCheckpoint type=off-the-shelf, Training mode=fine-tuned on DROID dataset2026.06 | 14.9 | 25.7 | 23.9 | |
| π0-FASTCheckpoint type=off-the-shelf, Training mode=fine-tuned on DROID dataset2026.06 | 9.2 | 15.5 | 14.9 | |
| GR00T N1.6Checkpoint type=off-the-shelf, Training mode=fine-tuned on DROID dataset2026.06 | 5.4 | 7.2 | 5.3 | |
| paligemma-binningCheckpoint type=off-the-shelf, Training mode=fine-tuned on DROID dataset2026.06 | 3.1 | 3.4 | 5.5 | |
| π0Checkpoint type=off-the-shelf, Training mode=fine-tuned on DROID dataset2026.06 | 2.8 | 5 | 3.5 |