Driver Behavior Modeling via Inverse Reinforcement Learning Based on Particle Swarm Optimization

In this paper, an inverse reinforcement learning method based on particle swarm optimization (PSO) is proposed to model driver's steering behavior. Initially, the vehicle dynamics is represented by a Takagi-Sugeno (T-S) fuzzy model which provides a method of approximating Q-function. Then the driver behavior model is described as an optimal control policy with decision-making model which illustrates the driving style. Subsequently, the Q-function is approximated by a quadratic polynomial-in-memberships form and the PSO algorithm is used to obtain the decision-making model from the driving data. And the corresponding optimal control policy is obtained by using the Q-learning policy iteration method. Finally, a numerical simulation is carried to show the effectiveness of the proposed method.