Intercept Guidance of Maneuvering Targets with Deep Reinforcement Learning

In this paper, a novel guidance law based on reinforcement learning (RL) algorithm is presented to deal with the maneuvering target interception problem using deep deterministic policy gradient descent neural network. We take missile’s line-of-sight (LOS) rate as observation of RL and propose reward function, which constructed miss distance LOS train network off-line. process, trained has capacity mapping normal acceleration missile directly, so generate commands in real time. Under actor-critic (AC) framework, we adopt twin-delayed (TD3) by taking minimum value between pair critics reduce overestimation. Simulation results show that proposed TD3-based outperforms current state law, better performance cope continuous action space, also faster convergence speed higher reward. Furthermore, accuracy robustness when intercepting target, converged.