USV maneuvering target tracking based on the HM-SAC algorithm

Objectives To address issues such as response delays and low tracking efficiency encountered by unmanned surface vehicles (USV) during high-speed maneuvering target tracking under limited information conditions, this paper proposes a target tracking control method for USV based on deep reinforcement learning. Methods This method is based on the Soft Actor-Critic (SAC) framework, designing observation space, action space, and reward functions under limited information. It integrates a Long Short-Term Memory (LSTM) network embedded within the SAC hidden layer, combining current environmental states with historical memory to optimize the reinforcement learning “state-action” mapping. This enables the unmanned surface vehicle to learn temporally optimal strategies, ultimately achieving effective tracking and control of maneuvering targets under limited information. Results Simulations demonstrate that the proposed method can achieve rapid tracking of maneuvering targets while maintaining a safe distance for continuous tracking, even in environments with wind, waves, currents, and observational delays with noise. The effectiveness of the proposed algorithm is validated through multiple rounds of random robust testing.Conclusions The proposed HM-SAC (Hidden Long Short-term Memory Soft Actor-Critic) algorithm demonstrates an efficient and adaptive tracking strategy, exhibiting significant advantages in unmanned surface vehicle target tracking control methods.