Open-loop Adversarial Scenario Generation on WOMD and MetaDrive
100Attack Success Rate (ASR)Rule
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Rule2026.05 | 100 | 5.048 | 5.614 | 1.734 | 7.724 | 2.08 | 8.546 | 0.204 | |
| KG-ASG2026.05 | 92.6 | 3.457 | 3.018 | 0.02 | 0 | 2.223 | 0.886 | 0.233 | |
| CAT2026.05 | 91.35 | 3.961 | 3.143 | 2.466 | 9.078 | 1.556 | 7.233 | 0.225 | |
| SAGE2026.05 | 76.15 | 4.121 | 2.479 | 0.731 | 1.084 | 2.098 | 8.088 | 0.184 | |
| AdvTrajOpt2026.05 | 70.46 | 2.652 | 2.775 | 2.547 | 10.16 | 1.754 | 6.177 | 0.268 | |
| SEAL2026.05 | 63.93 | 1.269 | 2.423 | 2.732 | 11.612 | 1.544 | 6.959 | 0.202 | |
| KING2026.05 | 40.85 | 2.243 | 3.434 | 3.126 | 6.056 | 0.972 | 255.5 | 0.098 | |
| GOOSE2026.05 | 36.07 | 2.378 | 21.32 | 4.426 | 14.48 | 6.368 | 8.286 | 0.154 |