Aggregate performance across all suite tasks on Safety Gymnasium
0.416Aggregate RewardCPQ
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CPQ2026.04 | 0.416 | 3.63 | |
| DT + SASBackbone=DT, Alignment strategy (SAS)=Lyapunov-Guided2026.04 | 0.402 | 1.03 | |
| CDT + SASBackbone=CDT, Alignment strategy (SAS)=Lyapunov-Guided2026.04 | 0.392 | 0.895 | |
| CDTBackbone=CDT2026.04 | 0.389 | 1.18 | |
| BCQ-Lag2026.04 | 0.319 | 1.52 | |
| BEAR-Lag2026.04 | 0.316 | 1.58 | |
| BC-All2026.04 | 0.234 | 1.85 | |
| COptiDICE2026.04 | 0.172 | 1.23 | |
| BC-Safe2026.04 | 0.151 | 0.602 | |
| DCRL2026.04 | 0.098 | 0.737 |