Rollout-level failure detection on ReactorX arm ACT policy rollouts
0.93ROC-AUCForesight-Transformer
Evaluation Results
| Method | Links | |
|---|---|---|
| Foresight-TransformerArchitecture=Transformer2026.06 | 0.93 | |
| SAFE-MLP2026.06 | 0.89 | |
| RND2026.06 | 0.86 | |
| FAIL-Detect2026.06 | 0.85 | |
| Foresight-LSTMArchitecture=LSTM2026.06 | 0.85 | |
| SAFE-LSTM2026.06 | 0.7 | |
| Foresight-MLPArchitecture=MLP2026.06 | 0.5 |