CARVE-Q uses quantum search to certify safe driving maneuver repairs
Quantum computing finds optimal driving repairs 100x faster with classical safety proofs.
Researchers have developed CARVE-Q, a hybrid quantum-classical system for certifying driving maneuver repairs after a veto. Traditional planners suggest cooperative interactions but lack formal proof that the repair respects rules, right-of-way, and responsibility allocation. CARVE (Certified Affordable Repair of Vetoed maneuvers via Envelopes) constructs a finite lattice of possible repairs and emits a structured certificate. The bottleneck is the multi-owner repair lattice, which scales exponentially. CARVE-Q addresses this by applying quantum minimum finding (Dürr-Hoyer/Grover) exclusively to the black-box lattice search, while all safety authority remains classical and verifiable.
In the conservative verifier-oracle model, exact classical minimum finding requires Θ(M) queries, but quantum minimum finding needs only O(√M) queries with high probability. The team proved verifier-shielded certificate soundness, priority non-elicitation, and finite-precision reversible-oracle constructibility. They demonstrated state-vector minimum finding on CARVE repair oracles with up to 65,536 assignments and validated certificate preservation on Lanelet2-grounded INTERACTION replay. Results achieved 100% right-of-way respect, 100% blame consistency, and zero priority false positives, establishing a trust-bounded quantum-AI pattern: quantum proposes, CARVE certifies.
- CARVE-Q reduces repair lattice search complexity from O(M) to O(√M) using quantum minimum finding
- Demonstrated on up to 65,536 assignments with 100% right-of-way respect and zero priority false positives
- All safety authority remains classical; quantum is used only for proposing optimal repairs
Why It Matters
Quantum-enhanced safety certification could enable faster, provably safe autonomous driving decisions.