Introduction to Reinforcement Learning for Torpedo Guidance with Obstacle Avoidance

This paper explores the application of reinforcement learning (RL) to torpedo guidance with a focus on obstacle avoidance and target acquisition in dynamic environments. By employing a dual-actor network approach and a regularization technique for the critic network, the proposed method demonstrates robust performance in scenarios with random obstacle placements and target directions. Our results show that the system effectively navigates toward the target while avoiding obstacles, leveraging reinforcement learning to optimize torpedo trajectories. The study addresses the challenges of overestimation and underestimation biases common in reinforcement learning methods, such as deep deterministic policy gradient (DDPG), by utilizing the double actors and regularized critics (DARC) algorithm. This approach has potential implications for enhancing the autonomy and efficiency of military applications in complex and uncertain settings.