Probabilistic Game-theoretic Control for Autonomous Vehicles on Lane-free Roads

Abstract

We propose a probabilistic game-theoretic controller (PGC) for autonomous vehicles on lane-free roads in decentralized and communication-free environments. At each time step, each vehicle solves a nonlinear Model Predictive Control problem and predicts the behavior of nearby vehicles through multi-level game-theoretic reasoning. We represent interaction uncertainty using structured trajectory hypotheses, and we incorporate collision risk through a Conditional Value-at-Risk-based cost term and iterative risk allocation constraints. We then compare the proposed PGC with the deterministic counterpart. The results show that the PGC handles uncertainty more effectively. In particular, it maintains safety while reducing unnecessary hesitation, avoiding overly conservative maneuvers, improving overtaking behavior, and increasing speeds along track. This study establishes a novel framework for probabilistic lane-free autonomous vehicle control under uncertainty.