List your company

Autonomous Driving Ego Robustness Evaluation on SafeBench held-out (test)

5CR (Straight Obstacle)ChatScene
1.425.75074.3May 20, 2026

Evaluation results

MethodLinks
ChatSceneAlgo.=ChatScene2026.0552.55505400013.489.189.468.393.892.675.791.894.886.9
ScenePilotAlgo.=ScenePilot2026.052017.512.5007.5007.283.28189.394.194.38794.794.789.8
PPDescription=Surrogate ego before fine-tuning2026.05958577.5947585858685.459.75655.847.658.553.745.442.752.5