Online Physical Risk Defense on LIBERO Long (Unseen Tasks)
98.3ASR w/o GuardSimpleVLA-Guard
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| SimpleVLA-Guardmode=online intervention, threshold calibration=Functional Conformal Prediction, alpha=0.052026.04 | 98.3 | 46.7 | 10 |