Observer-based optimized backstepping control using critic-actor reinforcement learning for morphing aircraft

This study developed an optimized backstepping control scheme for morphing aircraft, integrating observer design with Critic-Actor reinforcement learning (RL) theory to achieve optimal control. The aerodynamic model of the morphing aircraft was first analyzed, leading to the formulation of a nonlinear dynamic model. An improved extended state observer (ESO) was designed to estimate lumped disturbances and imprecise system states, enhancing observation accuracy. A Critic-Actor RL network, combined with the ESO, was then implemented at each control order to generate optimal control commands, ensuring optimal performance while maintaining disturbance resistance. A lightweight RL framework was developed using an adaptive law update method, enhancing practical applicability and differentiating it from traditional RL approaches by eliminating the need for repeated trial-and-error processes. The boundedness of the closed-loop signals was theoretically proven using the Lyapunov function, ensuring the stability and safety of the controller. However, challenges remain when addressing scenarios where the nonlinear dynamics of the aircraft are entirely unknown. Simulation results demonstrated that the proposed algorithm exhibits high robustness and stability under two simulation cases, with optimality guaranteed.