On-the-Fly Proof-Based Verification of Reachability in Autonomous Vehicle Controllers Relying on Goal-Aware RSS

Peter Rivière, Tsutomu Kobayashi, Neeraj Kumar Singh, Fuyuki Ishikawa, Yamine Aït Ameur, Guillaume Dupont

Abstract: Autonomous vehicles (AVs) are expected to satisfy not only safety, but they also shall achieve specific goals, e.g., stopping at a particular location on a shoulder lane of a highway for an emergency evacuation while avoiding collisions. The Goal-Aware Responsibility-Sensitive Safety (GA-RSS) framework was proposed to derive control strategies guaranteed to satisfy safety and goal achievement. This framework extends RSS rules, originally designed for safety in basic traffic situations, with a program logic allowing to reason on goal achievements in complex situations. In [11], the Event-B proof-based formal method was used to design a correct-by-construction model of the whole AV controller with a safety architecture and control strategies derived with the GA-RSS framework. This work is extended to handle liveness properties, which are extensively used to model complex goals and achieving them employing the EB4EB reflexive meta-modelling framework. As a result, relying on the EB4EB meta-model, an on-the-fly verification of temporal properties such as deadlock-freeness and goal reachability has been formalised and performed for advanced reasoning. Furthermore, the case study demonstrates additional strengths of the EB4EB meta-modelling approach, such as improvement of modelling and proof understandability and reusability.