Beyond Imitation: Learning Safe End-to-End Autonomous Driving from Hard Negatives

ArXi:2605.19771v1 Announce Type: cross Existing imitation learning methods for end-to-end autonomous driving predominantly learn from successful nstrations by minimizing geometric deviations from expert trajectories. This paradigm implicitly assumes that spatial proximity implies behavioral safety, leading to a critical objective mismatch: trajectories with nearly identical imitation losses may exhibit drastically different safety outcomes, where one remains recoverable while the other results in collision.