Improved Deep Reinforcement Learning Decision for Integrated Control of Distributed Drive Electric Vehicle Dynamics

Abstract

To solve the problem of nonlinear coupling with longitudinal and lateral tire forces in the integrated control of distributed driving electric vehicles, an improved SAC reinforcement learning algorithm is proposed to coordinate the control method of active front steering (AFS) and torque distribution. The coordinated control adopts a layered structure, and the upper layer adopts the LQR controller to determine the additional front wheel rotation angle and yaw moment required for stability control. The three-level reward function is designed with the goal of vehicle stability and state tracking and is determined by the reinforcement learning algorithm. For the weight factors of the two optimization objectives, the SAC reinforcement learning algorithm is used to coordinate the decision-making of the active front wheel steering and the additional yaw moment; the lower controller uses the comprehensive utilization rate of the tire as the objective function to perform the optimal torque distribution. A co-simulation platform is established in MATLAB/Simulink and CarSim software. The results show that compared with AFS control, SAC coordinated control can not only ensure the stability of the vehicle and achieve accurate tracking of the reference state but also effectively reduces the comprehensive utilization rate of tires and the effectiveness of coordinated control.