Q-Learning Based Parameter Adaptation in Model-Free Control Systems for Vehicle Trajectory Tracking

This paper addresses the precise modeling challenge in autonomous vehicle trajectory tracking control by proposing a data-driven model-free adaptive control (MFAC) approach. To solve the performance degradation problem of MFAC with fixed parameters under varying conditions, a Q-learning-based parameter optimization strategy is presented. The method constructs a state space incorporating tracking error, road adhesion coefficient, and vehicle speed, and evaluates tracking accuracy and control smoothness through a composite reward function to dynamically adjust the controller parameters. CarSim-Simulink co-simulation results demonstrate that the proposed parameter-adaptive control strategy significantly improves vehicle tracking accuracy and stability, reducing tracking errors by 19.6% under challenging low-friction scenarios compared to fixed-parameter approaches.