STOCHASTIC GAME THEORY AND MEAN-FIELD MODELS FOR AUTONOMOUS VEHICLE COORDINATION

Abstract

Coordinating large numbers of autonomous vehicles (AVs) on roads is a complex, uncertain, and decentralized control problem. Each vehicle must decide in real time how to accelerate, brake, change lanes, merge, or cross intersections while anticipating and influencing the actions of many others—some of which are human-driven and inherently unpredictable. The problem is made more difficult by partial observability (no vehicle sees everything), communication constraints, noisy sensors, delays, heterogeneous objectives (comfort, energy, time), safety requirements, and non-stationary environments.