Physics Joins AI: A Real-Time Hybrid Misbehavior Detection Framework for Vehicular Networks

Trusting autonomous, connected vehicles is necessary to build Cooperative Driving applications and enhance Smart Mobility. A necessary step is full confidence in communications beyond identification, pseudonyms and certificates: Message validity must be unquestionable to enable cooperation, even when messages originate from trusted entities affected by faults or malfunctions. Extreme reliability requires multiple evaluation tools, as independent as they can be, to fuse estimation into a dependable decision. This work proposes to combine two methods of evaluating messages, one based on Artificial Intelligence (AI) analysis and one on physical coherence of message content, achieving extremely good performance both on the VeReMi dataset, and in run-time execution in a highway scenario simulated with Plexe . Moreover, the paper proposes a simple protocol demonstrating how to safely dismantle a platoon of Cooperative Driving vehicles returning to autonomous (or human) driving when misbehaving messages are received.