Adaptive Higher‑Order Continuous Sliding Mode Guidance and Control with Dual LSTM‑Based Gain Scheduling for Ground‑to‑Air Tactical Interceptors

Abstract This study presents an adaptive higher-order continuous sliding mode control (ACHOSMC) framework, enhanced with dual Long Short-Term Memory (LSTM) neural networks, designed for high-precision missile interception in ground-to-air tactical scenarios. Two independently trained LSTM modules autonomously adjust boundary-layer gain \(\:β\:(t)\) and switching gain \(\:{λ\:}ₛ(t)\) in real time within a fully coupled three-dimensional integrated guidance and control (3D IGC) loop. Leveraging Lyapunov-constrained online learning, the system ensures closed-loop stability while mitigating chattering during severe maneuvers and stochastic disturbances capabilities absent in recent neural network-based IGC methods. Extensive six-degree-of-freedom simulations against agile targets demonstrate a 60% reduction in integrated time absolute error (ITAE) compared to CHOSMC and a 70% reduction compared to PID. Additionally, the system exhibits lower integrated absolute error (IAE) and integrated squared error (ISE), with minimized terminal miss distance and consistent control effort within actuator limits. Real-time inference on a Jetson Nano platform (sub-millisecond latency) confirms feasibility for embedded deployment. These advancements directly enhance kill probability by reducing miss distance and engagement time while extending actuator lifespan and improving operational survivability. The innovative integration of dual gain adaptation, real-time 3D robustness, and mission-level performance marks a significant leap forward for rapid-response defense systems, such as missile interception.