Closed-loop Driving Policy Evaluation on WOMD adversarial environments, w_adv=1.0 (test)
45.14RewardSAGE
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SAGE2025.09 | 45.14 | 0.61 | 0.69 | 0.31 | 8.98 | 28.85 | |
| Replay (No Adv)2025.09 | 41.32 | 0.68 | 0.62 | 0.44 | 8.77 | 30.42 | |
| CAT2025.09 | 37.7 | 0.7 | 0.58 | 0.37 | 6.85 | 31.83 | |
| Rule-based Adv2025.09 | 32.99 | 0.72 | 0.5 | 0.33 | 5.99 | 30.51 |