Action Generation on Environment OOD
36.6Rule-based ScoreOrca
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| OrcaBackbone=frozen, Action Expert=trainable from scratch2026.06 | 36.6 | 64 | 16 | 4 | 33.9 | 19.3 | 32.9 | 1.8 | |
| π0.5Backbone=frozen2026.06 | 27.6 | 54 | 16 | 2 | 27.9 | 17.7 | 31.5 | 1.5 | |
| V-JEPA 2.1Backbone=frozen, Action Expert=trainable from scratch2026.06 | 15.2 | 40 | 12 | 0 | 23 | 13.9 | 25.8 | 0 | |
| Qwen3.5Backbone=frozen, Action Expert=trainable from scratch2026.06 | 12.4 | 26 | 10 | 0 | 18.3 | 11.2 | 19.2 | 0 |