Multi-Adversarial Safety Analysis for Autonomous Vehicles

Gilbert Bahati, Marsalis Gibson, Alexandre M Bayen

Abstract: This work in progress extends reachability-based safe learn-ing to autonomous driving in multi agent systems. We formulate the safety problem for a car following scenario as a differential game and show how different modelling strategies yield very different behaviors regardless of the validity of the strategies in other scenarios. Given the nature of real-life driving scenarios, we distinguish and propose a modeling strategy in our formulation that accounts for subtle interactions between agents when needed, and generate Hamiltonian numerics to investigate how our formulation can produce safe sets compared to other baselines. Our formulation encourages safe exploration for an autonomous agent for safe motion planning and advances the goal of robust safety analysis and guarantees during navigation.

Keywords: Safety, Safe Learning, Autonomous Driving, Traffic Controls, Hamilton-Jacobi Reach-ability Analysis