Formal Estimation of Collision Risks for Autonomous Vehicles: A Compositional Data-Driven Approach

In this work, we propose a compositional data-driven approach for the formal estimation of collision risks autonomous vehicles (AVs) with black-box dynamics while acting in stochastic multi-agent framework. The proposed is based on construction sub-barrier certificates each agent via set data collected from its trajectories providing an priori guaranteed confidence estimation. our setting, first cast original risk problem as robust optimization program (ROP). Solving acquired ROP not tractable due to unknown model that appears one constraints. To tackle difficulty, collect finite numbers and provide scenario (SOP) corresponding ROP. We then establish probabilistic bridge between optimal value SOP ROP, accordingly, formally construct certificate number required level confidence. technique small-gain reasoning quantify AVs some desirable individual agents constructed data. For case compositionality conditions are satisfied, relaxed version results without requiring any but at cost potentially conservative risk. Eventually, also present approaches non-stochastic AVs. demonstrate effectiveness by applying them vehicle platooning consisting 100 1 leader 99 followers. estimate whole network collecting sampled agent.