Early Failure Detection on LIBERO Unseen
73.8ROC-AUC (q=0.25)ActProbe
Evaluation results
| Method | Links | |
|---|---|---|
| ActProbePolicy=OpenVLA2026.06 | 73.8 | |
| ActProbePolicy=π02026.06 | 71.4 | |
| SAFE-MLPPolicy=OpenVLA2026.06 | 67.9 | |
| STAC-SinglePolicy=π02026.06 | 65.2 | |
| SAFE-LSTM-TDQCPolicy=π02026.06 | 62.8 | |
| SAFE-MLP-TDQCPolicy=π02026.06 | 60.7 | |
| SAFE-LSTMPolicy=π02026.06 | 59.2 | |
| SAFE-MLPPolicy=π02026.06 | 58.9 | |
| LogpZOPolicy=π02026.06 | 58.3 | |
| SAFE-LSTM-TDQCPolicy=OpenVLA2026.06 | 57.6 | |
| Cosine k-NNPolicy=π02026.06 | 57.2 | |
| SAFE-LSTMPolicy=OpenVLA2026.06 | 55.9 | |
| SAFE-MLP-TDQCPolicy=OpenVLA2026.06 | 52.8 | |
| LogpZOPolicy=OpenVLA2026.06 | 52.6 | |
| Cosine k-NNPolicy=OpenVLA2026.06 | 42 |