Multi-agent driving behavior modeling on DeepScenario
0RMSE (m) (All)Ground Truth
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Ground Truth2025.12 | 0 | 46.9 | 1.865 | 0 | 83.1 | 21.9 | 0 | 3.995 | |
| OursObservation Type=IC, Num Parameter=429K, Training Time [h] (T4 16GB)=114.28, Training Time [h] (A100 80GB)=41.672025.12 | 10.71 | 32 | 1.1 | 13.38 | 57 | 53 | 6.93 | 1.83 | |
| Ours (small)Observation Type=IC, Num Parameter=59K, Training Time [h] (T4 16GB)=45.48, Training Time [h] (A100 80GB)=29.582025.12 | 11.36 | 33 | 1.23 | 14.26 | 58 | 83 | 7.13 | 1.74 | |
| GraphAIRLObservation Type=AC, Num Parameter=145K, Training Time [h] (T4 16GB)=133.43, Training Time [h] (A100 80GB)=56.422025.12 | 11.75 | 40 | 2.64 | 15.1 | 71 | 186 | 6.74 | 3.64 | |
| BCObservation Type=IC, Num Parameter=429K, Training Time [h] (T4 16GB)=8.33, Training Time [h] (A100 80GB)=1.082025.12 | 11.86 | 644 | 18.2 | 15.33 | 1,144 | 1,575 | 6.62 | 21.4 | |
| GraphAIRL†Observation Type=AC, Num Parameter=145K, Training Time [h] (T4 16GB)=135.16, Training Time [h] (A100 80GB)=57.56, Trained with proposed target reward=true2025.12 | 12.15 | 35 | 1.56 | 15.56 | 62 | 88 | 7.06 | 2.43 | |
| LateFusionMLP†Observation Type=AC, Num Parameter=104K, Training Time [h] (T4 16GB)=119.54, Training Time [h] (A100 80GB)=43.88, Trained with proposed target reward=true2025.12 | 15.14 | 58 | 3.85 | 19.68 | 104 | 452 | 8.37 | 2.99 | |
| CV2025.12 | 27.78 | 2,770 | 27.54 | 38.44 | 4,917 | 3,285 | 6.97 | 20.69 |