Physics-informed neural predictor-corrector guidance for time-coordinated reentry of hypersonic glide vehicles

Abstract

This paper addresses the time-coordinated reentry guidance problem for multiple hypersonic glide vehicles and proposes a physics-informed neural predictor-corrector guidance method. In the proposed framework, the path and terminal constraints are mapped to the bank-angle corridor, and a two-parameter bank-angle magnitude profile is designed in the velocity domain. Combined with heading-angle corridor-based lateral guidance and a cooperative flight-time determination strategy, the method enables rapid online generation of time-coordinated guidance commands. To reduce the computational burden of conventional predictor-corrector guidance, a physics-informed neural predictor is developed to estimate the time-to-go and range-to-go without repeated trajectory propagation. Unlike a purely supervised surrogate model, the predictor is tailored to the correction loop by embedding range-evolution and time-energy consistency constraints into the training process, thereby improving both prediction accuracy and physical consistency. Numerical simulations show that the proposed method achieves accurate and adaptive time-coordinated reentry guidance. Compared with numerical integration, the neural predictor requires only about 1.1% of the computational time while maintaining comparable prediction accuracy. Monte Carlo simulations further demonstrate the robustness of the proposed method under initial-state and aerodynamic uncertainties.