CARVE-Q: Quantum-Proposed, Classically Certified Interactive Driving Repair

Researchers introduce CARVE-Q, a quantum-classical hybrid system that certifies safe repairs for vetoed autonomous driving maneuvers while maintaining classical safety authority. The approach uses quantum minimum-finding algorithms to reduce computational complexity from linear to square-root time in multi-agent repair scenarios, validated on real-world driving datasets with perfect rule compliance.

CARVE-Q addresses a fundamental challenge in autonomous vehicle safety: when a driving system blocks a maneuver, determining whether an alternative repair exists that respects traffic rules, right-of-way, and liability boundaries. Traditional prediction and game-theoretic planners suggest alternatives but cannot guarantee compliance with hard safety constraints—a critical gap in certifiable autonomy systems.

The research builds on decades of autonomous vehicle development where safety certification has remained the primary bottleneck. As vehicles operate in multi-agent environments, the computational space for finding valid repairs grows exponentially with the number of actors involved. This creates a product lattice that becomes intractable for classical algorithms alone.

CARVE-Q's innovation lies in its strategic hybrid approach: quantum computing handles only the lattice search problem while classical systems retain all safety authority and rule verification. This architecture sidesteps concerns about quantum reliability in safety-critical domains by compartmentalizing quantum speedup to a specific black-box optimization task. The theoretical guarantees—proven square-root query reduction via Dürr-Hoyer and Grover algorithms—translate to measurable acceleration on problem sizes up to 65,536 assignments.

The validation on INTERACTION dataset replay demonstrates practical viability: 100% right-of-way respect, zero false positives in priority determination, and consistent blame attribution across repairs. For autonomous vehicle developers, this establishes a concrete pattern for quantum-AI integration that maintains auditability and legal defensibility. The work signals that near-term quantum utility may emerge first in certified systems engineering rather than general ML acceleration, potentially reshaping how industry approaches safety-critical autonomous systems over the next 3-5 years.

• →CARVE-Q uses quantum minimum-finding to reduce multi-agent repair search from O(M) to O(√M) classical oracle queries while maintaining classical safety verification authority.
• →The system produces auditable certificates recording binding rules, cost allocation, right-of-way compliance, and fallback strategies for each repair decision.
• →Validation on real driving data (INTERACTION dataset) achieved 100% right-of-way respect and zero priority false positives across 100% of repairs tested.
• →The hybrid quantum-classical architecture compartmentalizes quantum speedup to lattice search only, preserving classical safety authority for regulatory and legal compliance.
• →This work demonstrates a near-term quantum utility pattern applicable to certified autonomous systems beyond vehicles, relevant to safety-critical AI generally.