Robot Policy Mode Editing (2 to 1) on Push-Wall
82Avg Success RateFiltered-retrain ref.
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Filtered-retrain ref.Backbone=π0.5 VLA, Mode Change=2 → 1, Evaluation Seeds=502026.06 | 82 | 88 | 82 | 88 | |
| MoREBackbone=π0.5 VLA, Mode Change=2 → 1, Evaluation Seeds=502026.06 | 72 | 80 | 79 | 84 | |
| Original policyBackbone=π0.5 VLA, Mode Change=2 → 1, Evaluation Seeds=502026.06 | 41 | 52 | 82 | 82 |