Neural-network-aided integrated control design for launchers via observer-augmented high-order fully actuated system approach

This paper presents a neural-network-aided integrated guidance and control strategy for launchers based on an observer-augmented High-Order Fully Actuated System (HOFAS) framework. In contrast to conventional cascade architectures, the proposed approach unifies translational and rotational dynamics into a single HOFAS model, eliminating the need for sequential attitude reference generation. To compensate for unmeasurable high-order states, an extended state observer is integrated into the HOFAS framework, enabling recursive controller synthesis with parametric eigenstructure assignment. The observer gains are rigorously computed via linear matrix inequality conditions to ensure exponential stability of the augmented closed-loop system. Furthermore, a neural-network-based gain tuning strategy is proposed to achieve real-time adaptability under varying initial conditions, with controller parameters optimized online via Bayesian optimization. Comprehensive numerical simulations, including Monte Carlo trials, validate the effectiveness and robustness of the proposed method in orbit insertion scenarios. Compared with a first-order state-space-based approach, the proposed framework demonstrates superior tracking performance and improved propellant efficiency, underscoring its potential for future high-performance space missions.