Robotic Manipulation on RoboCasa365
42Average Success Rateπ0.5
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| π0.5Pre-training Type=Action-labeled pretraining (VLA)2026.06 | 42 | 34 | — | — | |
| QWEN-ROBOTMANIP2026.06 | 35.9 | — | — | — | |
| QWEN-ROBOTMANIP-ContextContext-Aware=true2026.06 | 33.8 | — | — | — | |
| RLDX-12026.06 | 33.2 | — | — | — | |
| RLDX-1mode=fine-tuned, training steps=250K, global batch size=1962026.05 | 32.1 | 67.3 | 19 | 5.6 | |
| µ0 + action expertPre-training Type=Video-only pretraining2026.06 | 30.25 | 54 | — | — | |
| GR00T N1.6mode=fine-tuned, training steps=250K, global batch size=1962026.05 | 26.9 | 61.1 | 12.6 | 2.6 | |
| π0Pre-training Type=Action-labeled pretraining (VLA)2026.06 | 25.25 | 44 | — | — | |
| GR00T-N1.52026.06 | 23.9 | — | — | — | |
| TraceGen + action expertPre-training Type=Video-only pretraining2026.06 | 23 | 38 | — | — | |
| Diffusion PolicyPre-training Type=No pretraining2026.06 | 22.75 | 34 | — | — | |
| GR00T-N1.62026.06 | 21.9 | — | — | — | |
| π0-FASTmode=fine-tuned, training steps=250K, global batch size=1962026.05 | 21.7 | 51.7 | 8 | 1.8 | |
| GR00T N1.5mode=fine-tuned, training steps=250K, global batch size=1962026.05 | 20 | 43 | 9.6 | 4.4 | |
| π0.5mode=fine-tuned, training steps=250K, global batch size=1962026.05 | 16.9 | 39.6 | 7.1 | 1.2 | |
| π0.52026.06 | 16.9 | — | — | — | |
| π02026.06 | 15 | — | — | — | |
| π0mode=fine-tuned, training steps=250K, global batch size=1962026.05 | 14.8 | 34.6 | 6.1 | 1.1 |