Inverse Reinforcement Learning Based Stochastic Driver Behavior Learning

Drivers have unique and rich driving behaviors when operating vehicles in traffic. This paper presents a novel driver behavior learning approach that captures the uniqueness richness of human realistic scenarios. A stochastic inverse reinforcement (SIRL) is proposed to learn distribution cost function, which represents with given set driver-specific demonstrations. Evaluations are conducted on data collected from 3D driver-in-the-loop simulation. The results show learned model capable expressing strategies under different Compared deterministic baseline model, reveal can better replicate driver’s variety traffic conditions.