Robust Reinforcement Learning-Based Guidance Strategy Against Cyber-Attacks

As electronic countermeasure techniques become increasingly sophisticated, missiles that rely on ground-based systems to generate and transmit guidance commands face the risk of cyber-attacks. Most existing guidance strategies do not account for such attacks, resulting in significant degradation of guidance performance once attacks occur. To address this issue, this paper proposes a robust reinforcement learning-based guidance strategy that incorporates cyber-attack factors. First, the kinematic and relative-motion models for the missile-maneuvering target engagement are established, and typical cyber-attack scenarios are analyzed. Based on these models, a robust Markov decision process incorporating cyber-attack effects is formulated, providing a theoretical framework for applying robust reinforcement learning to guidance problems. Within this framework, a robust reinforcement learning-based guidance strategy with a robust Actor–robust Critic architecture is developed. Specifically, the robust Critic updates its parameters by minimizing the mean-squared robust temporal-difference error. Meanwhile, the robust Actor updates the policy using the clipped objective of robust Proximal Policy Optimization. Together, these updates improve the robustness of the guidance policy under cyber-attacks. Extensive simulation results demonstrate that the proposed guidance strategy successfully intercepts maneuvering targets under diverse attack conditions, validating its robustness and guidance effectiveness under cyber-attacks.