Rollout-level failure prediction on LIBERO GOAL (test)
0.841AUROCSW+ATR
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SW+ATRw=60, alpha=0.92026.03 | 0.841 | 84.3 | |
| Sliding Window Poolingw=702026.03 | 0.837 | 84.6 | |
| SW+ATR+BOparameters=w*, alpha*, beta*2026.03 | 0.811 | 80.7 | |
| Action Transfer Reweightingalpha=0.92026.03 | 0.67 | 81.9 | |
| Baselineparams=mean_{t,d}2026.03 | 0.641 | 75.5 |