Contextual DRL-empowered integrated guidance and evasion approach for hypersonic glide vehicles

To balance the survival and mission requirements for the hypersonic glide vehicles (HGVs) when encountering multiple interceptors, an intelligent integrated guidance and evasion framework is proposed. First, the HGV model and flight constraints are established. Second, a novel decoupled interfered fluid dynamical system (DIFDS) evasive algorithm is proposed. The evasion timing, direction, and overload can be determined by three crucial coefficients in the DIFDS algorithm, and the evasion performance can be theoretically proven. On this basis, the integrated framework of guidance, fluid-based evasion, and control (IFGFEC) is established by combining the predictor–corrector guidance, the DIFDS algorithm and the prescribed-time control theory. The proposed framework ensures that the HGV performs evasive maneuvers in the presence of interceptors and follows guidance in their absence, thereby satisfying both survival and mission requirements. Third, to improve the autonomous decision-making capability of the HGV, an intelligent evasion policy based on the contextual deep reinforcement learning (CDRL) method is proposed, in which the contextual Markov decision process (CMDP) is introduced to fully represent the engagement between the HGV and interceptors. Finally, the effectiveness of the proposed method is verified by numerical simulations.