Failure Detection on WidowX (unseen)
0.153Brier ScoreSAFE-RNN-TDQC (Ours)
Evaluation Results
| Method | Links | |
|---|---|---|
| SAFE-RNN-TDQC (Ours)VLA Model=OpenVLA2026.04 | 0.153 | |
| SAFE-MLP BCEVLA Model=OpenVLA2026.04 | 0.164 | |
| SAFE-MLP-TDQC (Ours)VLA Model=OpenVLA2026.04 | 0.169 | |
| RNN-TDQC (Ours)VLA Model=OpenVLA2026.04 | 0.192 | |
| SAFE-RNNVLA Model=OpenVLA2026.04 | 0.213 | |
| Running Avg prob.VLA Model=OpenVLA2026.04 | 0.257 | |
| Avg prob.VLA Model=OpenVLA2026.04 | 0.282 | |
| RNN-BCEVLA Model=OpenVLA2026.04 | 0.344 | |
| Avg entropyVLA Model=OpenVLA2026.04 | 0.426 | |
| Running Avg entropyVLA Model=OpenVLA2026.04 | 0.432 | |
| Max prob.VLA Model=OpenVLA2026.04 | 0.579 |