DriveSafer: End-to-End Autonomous Driving with Safety Guidance

TL;DR

DriveSafer is a safety framework for end-to-end autonomous driving that significantly reduces catastrophic failures by guiding generative models towards safer behaviors using safety constraints and guidance.

Contribution

It introduces DriveSafer, a novel safety-aware framework that improves safety outcomes in generative autonomous driving models by explicitly steering towards safe behaviors.

Findings

DriveSafer reduces catastrophic failures by 48% on NAVSIM benchmark.

It achieves over 65% reduction in drivable-area compliance failures.

Outperforms state-of-the-art DiffusionDrive model in safety metrics.

Abstract

End-to-End (E2E) autonomous driving models have shown growing capability in recent years, with performance improving on increasingly challenging benchmarks. However, modern generative E2E planners still suffer from a substantial number of catastrophic failures in safety-critical scenarios. We find that many such failures arise from violations of physical constraints and safety requirements, leading to unsafe behavior. Motivated by this finding, in this paper, we focus on improving safety outcomes in generative end-to-end driving with a targeted reduction of catastrophic planning failures, instead of enhancing average planning quality. Towards this end, we propose DriveSafer, a failure-aware safety framework for end-to-end planners. DriveSafer explicitly steers generative planners towards safe behaviors leveraging both training-time safety constraints and inference-time safety guidance. Compared to the state-of-the-art DiffusionDrive model, on the NAVSIM benchmark, DriveSafer reduces the number of catastrophic failures (PDMS=0) by 48%, with over 65% reduction in drivable-area compliance failures.