Lyapunov-guided SRWNN-based safe reinforcement learning control for morphing flight vehicles under non-Gaussian aerodynamic uncertainties

Purpose This research seeks to provide a stability-guaranteed intelligent control scheme that addresses the morphing flight vehicles (MFVs) working in the presence of nonlinear and non-Gaussian aerodynamic uncertainty. Design/methodology/approach To address the aforementioned challenges, this study proposes a Lyapunov-based self-recurrent wavelet neural network (SRWNN)-assisted safe reinforcement learning architecture for coordinating morphing actions and flight control. Specifically, the SRWNN will be used to estimate the nonlinear aerodynamic uncertainty in real time based on its high capacity of approximation and temporal learning. Then, a Lyapunov-constrained safe reinforcement learning strategy is introduced to ensure safe policy evolution and stability during learning. In addition, an adaptive dynamic programming-based controller is developed to realize robust attitude stabilization and optimal morphing actions. Findings As shown by the simulation studies conducted on a nonlinear 6DoF MFV, the presented approach demonstrates better tracking performance, faster convergence, lower control action oscillation and stronger robustness under non-Gaussian disturbances than conventional schemes. Originality/value In this study, a stability-guaranteed safe reinforcement learning scheme combined with SRWNN is developed for estimating nonlinear aerodynamic uncertainty and coordinating morphing actions under uncertain flight environment.