MPC-TD3 Trajectory Tracking Control for Electrically Driven Unmanned Tracked Vehicles

To address the trajectory tracking issue of unmanned tracked vehicles, the majority of studies employ the Model Predictive Control (MPC). The MPC imposes high demands on model accuracy. Due to factors such as environmental interference, actuator constraints, and the nonlinearity of vehicles under high-speed conditions, dynamic and kinematic models fail to accurately delineate the motion process of tracked vehicles. Aiming at the problem of insufficient trajectory tracking precision of unmanned tracked vehicles, a trajectory tracking controller jointly controlled by the Twin Delayed Deep Deterministic policy gradient (TD3) algorithm and the MPC algorithm is developed. During offline training, the agent acquires the discrepancies between the model and the environment under various working conditions and optimizes its own network; during online reasoning, the agent adaptively compensates the output of the MPC based on the vehicle state. The experimental results indicate that, compared with the pure MPC algorithm, the MPC algorithm compensated based on the TD3 algorithm reduces the lateral errors by 41.67% and 22.55%, respectively, in circular and double-lane-change trajectory conditions.