Robot Manipulation on Realworld Manipulation (test)
76.3Average Success RateHARP-VLA
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| HARP-VLAVision Encoder Freezing=True, Alignment Loss=SRPD, Trials per task=602026.05 | 76.3 | 76.7 | 81.7 | 85 | 61.7 | |
| HARP-VLAVision Encoder Freezing=False, Alignment Loss=SRPD, Trials per task=602026.05 | 73.3 | 76.7 | 80 | 78.3 | 58.3 | |
| HARP-VLAAlignment Loss=L2, Trials per task=602026.05 | 70 | 70 | 71.7 | 81.7 | 56.7 | |
| π0.5Trials per task=602026.05 | 69.2 | 71.7 | 83.3 | 68.3 | 53.3 | |
| OpenVLA-OFTTrials per task=602026.05 | 60.9 | 51.7 | 71.7 | 76.7 | 43.3 | |
| π0Trials per task=602026.05 | 56.3 | 58.3 | 75 | 56.7 | 35 | |
| NAC2026.06 | 50 | — | — | — | — | |
| FAST2026.06 | 40 | — | — | — | — | |
| OAT2026.06 | 40 | — | — | — | — | |
| UniVLATrials per task=602026.05 | 38.4 | 38.3 | 61.7 | 31.7 | 21.7 | |
| VQ-VLA2026.06 | 31.25 | — | — | — | — | |
| Diffusion Policy2026.06 | 22.5 | — | — | — | — | |
| OpenVLATrials per task=602026.05 | 10.4 | 0 | 23.3 | 18.3 | 0 | |
| Bin2026.06 | 6.25 | — | — | — | — |