Adaptive Neuro‑Symbolic Planning for Autonomous Urban Air Mobility Routing with Zero‑Trust Governance Guarantees

Описание и применение

Нейро‑символьный планировщик с нулевой‑доверительной политикой обеспечивает гарантированную безопасность маршрутов автономных летательных аппаратов в городском воздушном пространстве, сочетая нейросетевое предложение траекторий с SMT‑верификацией и аудитом.

Задача

Neural planners lack provable safety guarantees while symbolic planners cannot handle dynamic, noisy environments and real‑time latency, making autonomous UAM routing difficult.

Результат

The prototype achieved higher acceptance rates by incorporating verification‑aware loss and produced provably safe routes while respecting real‑time latency.

Рабочий процесс

• 1. Perception layer fuses sensor data and predicts traffic.
• 2. Neural proposal network (transformer‑based) generates multiple candidate trajectories.
• 3. Symbolic verification layer (SMT solver + temporal logic) filters candidates for constraint compliance.
• 4. Zero‑trust governance layer attests inputs, monitors runtime, and audits the pipeline.
• 5. Accepted trajectory is executed by the aircraft.

Доказательства

Neural networks are brilliant at pattern recognition and approximate reasoning, but they're notoriously terrible at guaranteeing anything.

Источник

Symbolic planners can prove properties about their outputs, but they crumble when the environment is noisy, dynamic, and full of uncertainty.

Источник

The key insight is that each layer has a different trust model. The neural network is untrusted by default — it proposes, but never commits.

Источник

Training the network with a verification‑aware loss dramatically improved the acceptance rate.

Источник