PointGoal2 on Safety-Gymnasium
67Normalized RewardBCQ-Lag
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BCQ-Lag2026.04 | 67 | 3.18 | |
| BEAR-Lag2026.04 | 67 | 3.11 | |
| DT + SASBackbone=DT, Alignment strategy (SAS)=Lyapunov-Guided2026.04 | 65 | 1.78 | |
| CDTBackbone=CDT2026.04 | 59 | 1.34 | |
| BC-All2026.04 | 54 | 1.97 | |
| CDT+SASAlignment=SAS, Base Model=CDT2026.04 | 52 | 94 | |
| CDT + SASBackbone=CDT, Alignment strategy (SAS)=Lyapunov-Guided2026.04 | 52 | 0.94 | |
| CPQ2026.04 | 40 | 1.31 | |
| COptiDICE2026.04 | 38 | 1.92 | |
| FISOR2026.04 | 30 | 35 | |
| BC-Safe2026.04 | 29 | 0.78 | |
| DCRL2026.04 | 28 | 0.26 | |
| BCQ-Lagcost thresholds=3, evaluation episodes=20, random seeds=32026.04 | 0.67 | 3.18 | |
| BEAR-Lagcost thresholds=3, evaluation episodes=20, random seeds=32026.04 | 0.67 | 3.11 | |
| DT + SAScost thresholds=3, evaluation episodes=20, random seeds=32026.04 | 0.65 | 1.78 | |
| CDTcost thresholds=3, evaluation episodes=20, random seeds=32026.04 | 0.59 | 1.34 | |
| BC-Allcost thresholds=3, evaluation episodes=20, random seeds=32026.04 | 0.54 | 1.97 | |
| CDT + SAScost thresholds=3, evaluation episodes=20, random seeds=32026.04 | 0.52 | 0.94 | |
| CPQcost thresholds=3, evaluation episodes=20, random seeds=32026.04 | 0.4 | 1.31 | |
| COptiDICEcost thresholds=3, evaluation episodes=20, random seeds=32026.04 | 0.38 | 1.92 | |
| BC-Safecost thresholds=3, evaluation episodes=20, random seeds=32026.04 | 0.29 | 0.78 |