AI RESEARCH

SafeAlign-VLA: A Negative-Enhanced Safe Alignment Framework for Risk-Aware Autonomous Driving

arXiv CS.CV

ArXi:2605.19524v1 Announce Type: cross End-to-end autonomous driving systems excel in common scenarios but struggle with safety-critical long-tail cases. Vision-Language-Action (VLA) models are promising due to their strong reasoning capabilities. However, most VLA-based approaches rely on positive expert nstrations, rarely exploiting negative samples, leading to insufficient understanding of risky behaviors and safety boundaries.