This paper addresses the critical challenge of online intelligent penetration maneuver-making for aircraft operating within multi-interceptor engagement scenarios. Online intelligent penetration maneuver-making refers to the process whereby the aircraft autonomously perceives the battlefield situation and performs maneuver in the complex confrontation environment, which requires the aircraft to quickly update the strategy in a limited time. To address this challenge, a Starnet Soft Actor-Critic with Anti-Forgetting Meta-Learning (S-SAC-AFML) for aircraft intelligent penetration is proposed, which integrates a multi-objective behavior prediction network based on spatio-temporal coupling information into the Soft Actor-Critic (SAC) algorithm and a meta-learning-based optimization mechanism is established to enable dynamic parameter adaptation to the changing tasks. Specifically, the framework leverages the predicted information of non-cooperative aircraft to enhance policy training robustness while maintaining the exploration efficiency of the maximum entropy-based decision-making process. Furthermore, considering the discrepancy between penetration scenarios, particularly the uncertainty in enemy tactics, it incorporates an Anti-Forgetting Meta-Learning (AFML) algorithm to enable the aircraft to retain critical knowledge from previous tasks while acquiring the specific information of new tasks, thereby mitigating catastrophic forgetting. The superior effectiveness and adaptability of the proposed S-SAC-AFML method are rigorously validated through comprehensive simulation experiments and demonstrated within a real-time interactive environment built on Unity 3D.
Yu et al. (Sun,) studied this question.