Deep Reinforcement Learning-Based Intelligent Torpedo for Target Tracking

This study examines deep reinforcement learning (DRL) to improve torpedo target tracking. Considering the limitations of traditional control methods in handling nonlinear and uncertain dynamics, a control strategy developed with the Proximal Policy Optimization (PPO) algorithm is proposed. A simulation environment with realistic underwater dynamics was created using Unity, and the torpedo model was trained against stationary and moving submarine targets. During this process, a particular reward function was designed to facilitate rapid and effective learning by the torpedo. Simulation results demonstrate that the torpedo trained with the PPO algorithm achieved an 85% success rate against stationary targets and a 79% success rate against moving targets. These findings indicate that DRL-based control strategies are a powerful tool for enhancing the operational effectiveness of torpedoes. The results mainly highlight the significant potential of DRL in improving the target tracking and hitting performance of torpedoes in dynamic and uncertain underwater conditions.