An Enhanced Path Tracking Control Method for Unmanned Surface Vehicles Using Combined Reinforcement Learning DDPG Algorithm and MPC

To address the problem of path tracking control for the USVs in highly dynamic environments, we propose a trajectory tracking control algorithm based on DDPG and MPC. DDPG's Actor network generates preliminary control actions, optimized by MPC for final control. LSTM networks process sequence data for Actor and Critic networks. Future state predictions are stored in a “Prediction Pool” to enhance Critic training and accelerate convergence. The outcome of the simulations demonstrates that the introduced algorithmic approach exhibits efficient convergence in terms of path-tracking precision, significantly reducing path error and heading error, demonstrating considerable potential for practical applications.