Rollout-level failure detection on ReactorX arm SmolVLA policy rollouts
0.82ROC-AUCRND
Evaluation Results
| Method | Links | |
|---|---|---|
| RND2026.06 | 0.82 | |
| Foresight-TransformerArchitecture=Transformer2026.06 | 0.79 | |
| FAIL-Detect2026.06 | 0.71 | |
| SAFE-MLP2026.06 | 0.64 | |
| Foresight-LSTMArchitecture=LSTM2026.06 | 0.64 | |
| Foresight-MLPArchitecture=MLP2026.06 | 0.53 | |
| SAFE-LSTM2026.06 | 0.43 |