Early Failure Detection on LIBERO (Seen)
0.809ROC-AUC (q=0.25)SAFE-MLP
Evaluation Results
| Method | Links | |
|---|---|---|
| SAFE-MLPPolicy=OpenVLA2026.06 | 0.809 | |
| ActProbePolicy=π02026.06 | 0.801 | |
| SAFE-LSTM-TDQCPolicy=π02026.06 | 0.796 | |
| SAFE-LSTMPolicy=π02026.06 | 0.782 | |
| SAFE-MLPPolicy=π02026.06 | 0.773 | |
| SAFE-MLP-TDQCPolicy=π02026.06 | 0.766 | |
| ActProbePolicy=OpenVLA2026.06 | 0.763 | |
| SAFE-LSTMPolicy=OpenVLA2026.06 | 0.757 | |
| LogpZOPolicy=π02026.06 | 0.731 | |
| Cosine k-NNPolicy=OpenVLA2026.06 | 0.721 | |
| Cosine k-NNPolicy=π02026.06 | 0.689 | |
| SAFE-LSTM-TDQCPolicy=OpenVLA2026.06 | 0.675 | |
| SAFE-MLP-TDQCPolicy=OpenVLA2026.06 | 0.657 | |
| STAC-SinglePolicy=π02026.06 | 0.653 | |
| LogpZOPolicy=OpenVLA2026.06 | 0.577 |