Research on robot path tracking method based on IDDPG-MPC

In complex marine environments, path-following control of unmanned surface vessels (USVs) faces numerous challenges, including environmental disturbances, dynamic nonlinearities, and underactuated systems. To overcome the limitations of traditional line-of-sight/PID control in terms of robustness and adaptability, this study proposes a hybrid control architecture combining improved deep deterministic policy gradient (IDDPG) and model predictive control (MPC). The IDDPG algorithm, as the upper-level decision-making module, utilizes deep reinforcement learning to generate optimal heading angle increment commands by learning the environmental state. The MPC, as the lower-level execution module, optimizes control variables such as thrust and rudder angle through rolling optimization based on the USV’s three-degree-of-freedom nonlinear dynamics model. This study constructs a closed-loop “perception-decision-execution-learning” paradigm and employs gradient pruning and a customized reward function to ensure the stability of algorithm training and the optimality of control decisions. Lateral deviation and heading angle error are used as evaluation metrics to verify the control performance. Simulation results show that this method effectively solves the adaptability challenge of traditional control strategies in complex environments. Compared with the traditional ALOS-PID method, the average lateral deviation is reduced by 37% and the heading angle error is reduced by 21%, thus realizing high-precision path tracking control for unmanned surface vessels and providing a new method for autonomous surface vehicle navigation.