Autonomous Driving on Waymo Open Motion Dataset (WOMD) Self-play 10k held-out scenarios (test)
0Collision Rate (%)spiced self-play
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| spiced self-playHuman demos used=30 hours, regularization=true2026.06 | 0 | 0.3 | 99.7 | |
| spiced self-playHuman demos used=30 min, regularization=true2026.06 | 0.2 | 0.5 | 99.3 | |
| spiced self-playHuman demos used=3 hours, regularization=true2026.06 | 0.2 | 0.6 | 99.6 | |
| unreg. self-playHuman demos used=—2026.06 | 1 | 0.2 | 99.9 | |
| spiced self-playHuman demos used=10 min, regularization=true2026.06 | 1 | 0.3 | 99 | |
| SMARTHuman demos used=52 days2026.06 | 6.1 | 53.5 | 91.7 | |
| SMARTHuman demos used=30 hours2026.06 | 7.7 | 53.3 | 86.5 | |
| SMARTHuman demos used=3 hours2026.06 | 8 | 53.6 | 86.2 | |
| SMARTHuman demos used=30 min2026.06 | 9.5 | 55.4 | 85.8 | |
| SMARTHuman demos used=10 min2026.06 | 11.9 | 55.8 | 84.5 |