A human-like lane-changing behavior model for autonomous vehicles in mixed traffic flow environment

Driving models with enhanced human-likeness can greatly improve the driving compatibility of autonomous vehicles (AVs) in mixed traffic environments. By analyzing the behaviors of human drivers, we propose a human-like lane-changing driving model for AVs. Firstly, the multi-stage repeated game theory is used to describe the lane-changing decision process, and the Bayesian inference method is proposed to assess the driving preferences of surrounding vehicles online. Through the multi-stage game and evaluation, the dynamic interaction and decision-making between vehicles are realized. Then, an asymmetric driving risk field considering different vehicle types and driving styles is established based on the field theory, and the dynamic trajectory planning of autonomous vehicles is realized by combining model predictive control (MPC). Finally, three test scenarios involving different social behaviors of heterogeneous vehicles are designed to verify the effectiveness of the proposed approach. Simulation results show that the model can identify the driving styles of interactive vehicles and provide safe and personalized human-like driving behaviors for AVs in complex traffic environments.