Energy-saving path planning method for unmanned surface vehicles based on curiosity-driven reinforcement learning

Reasonable global path planning techniques for unmanned surface vehicles (USV) can improve work efficiency. A global path planning method based on deep reinforcement learning is proposed to reduce the energy consumption of USVs in environments with current. Firstly, a simulation experimental environment is built based on electronic charts (EC) and current models to simulate the real environment. Then, the deep deterministic policy gradient (DDPG) training framework based on the intrinsic curiosity module (ICM) is built to enhance the exploration ability and improve the training efficiency. Subsequently, an external reward strategy is designed based on reward reshaping policy to suppress excessive random exploration of curiosity. Finally, simulation and comparative experiments are conducted to verify the algorithm's effectiveness.