AI RESEARCH

CARVE: Certified Affordable Repair of Vetoed Maneuvers via Envelopes for Interactive Driving

arXiv CS.AI

ArXi:2606.02641v1 Announce Type: cross Interactive driving exposes a failure mode that is easy to miss in rule-aware autonomous-driving stacks: a hard-rule margin can be negative for an ego candidate even though a small lawful accommodation by a non-priority agent would re feasibility. Existing rulebooks, shields, and reachability filters are strong at vetoing unsafe actions, while prediction-based planners model likely responses.