CARVE algorithm resolves 98% of vetoed autonomous driving maneuvers

Autonomous driving stacks often rely on hard-rule margins that can be unnecessarily restrictive. A common failure mode occurs when a maneuver is vetoed because the ego vehicle's rule margin is slightly negative, even though a small lawful accommodation by a non-priority agent would make the maneuver feasible. Existing rulebooks, shields, and reachability filters are strong at vetoing unsafe actions but do not provide a runtime proof object that states which bounded multi-agent edit repairs the maneuver, who owns the edit, whether it is right-of-way affordable, and what fallback remains if the request is not observed.

Enter CARVE (Certified Affordable Repair of Vetoed Maneuvers via Envelopes), a prediction-free certificate layer developed by Yifan Wang. CARVE operates over a finite lattice of ego-owned and agent-owned tactical operators. Agent-owned requests are only admissible inside a cooperation envelope that separates kinematic reachability from normative priority. The resulting certificate records the binding rule, repair category, repair set, responsibility-weighted cost split, and fallback. In testing on 589 Lanelet2-geometry-grounded INTERACTION replay episodes, CARVE-Greedy accepted 98.64% of initially vetoed maneuvers and recovered 370/378 human-resolved false vetoes. It preserved 589/589 right-of-way respect, achieved zero priority-agent false positives, and maintained 400/400 negative-stress vetoes. The system is formally proven for soundness, structural right-of-way respect, exact finite-lattice minimality, fallback contingency, and blame-consistency. Crucially, CARVE does not predict or require another driver's compliance—it certifies whether a proposed interaction is bounded, attributable, and normatively admissible under declared assumptions.