A SAC-Based Deep Reinforcement Learning Approach for Autonomous Underwater Vehicle Combat

In order to adapt to the needs of underwater combat under low maneuverability conditions, this paper applies a soft actor-critic (SAC) based deep reinforcement learning algorithm to the strategy research of autonomous underwater vehicle (AUV) suicide attack. Firstly, the kinematic model of AUV is constructed and the underwater combat scenario and rules are established. Then the SAC algorithm is used in the strategy training for under-water combat environments, and the optimizations of reward functions are formulated for the characteristics of the environment. Finally, through the simulation and comparison with the DQN algorithm, the superiority of the proposed method is verified.