UUV Target Tracking Path Planning Algorithm Based on Deep Reinforcement Learning

Path planning is one of the basic key problems in UUV task planning research. This paper studies the UUV path planning method in target tracking task scenario. The target is in a moving state, the moving elements are uncertain, and the traditional path planning algorithm is not applicable or easy to fall into the local optimal solution. In this paper, a tracing path planning algorithm based on deep reinforcement learning is presented, and a network parameter update method combining soft update with optimal sample training is proposed in the target network update link. The simulation results show that the algorithm can accelerate the network convergence speed while guaranteeing the stability of the learning process, and can quickly plan the optimal trajectory and maximize the time to track the target after UUV finds the target.