A Deep Reinforcement Learning-Based Guidance Law with Impact Angle Constraint

Addressing the challenge of ineffective impact angle control when employing conventional biased proportional navigation guidance for precision-guided missiles against maneuvering targets, this study proposes a guidance law with impact angle constraint by integrating deep reinforcement learning algorithms with biased proportional navigation guidance. The algorithm enhances traditional proportional navigation guidance by introducing a target maneuver compensation bias term. This bias term is treated as an action in a learning framework and optimized using the Proximal Policy Optimization algorithm. Simulation experiments validate that the proposed algorithm, compared to conventional proportional navigation guidance, not only reduces miss distance and impact angle error but also shortens flight time.