Distributed optimal formation control for UAV-USV multiagent systems: A FEWNN-based RL approach with self-adjusting prescribed performance

Abstract

This paper designs a fuzzy Elman wavelet neural network (FEWNN)-based reinforcement learning (RL) optimal controller with self-adjusting prescribed performance for the formation tracking control problem of unmanned aerial vehicle-unmanned surface vehicle (UAV-USV) systems. First, a self-adjusting prescribed performance function (SAPF) is developed to constrain tracking errors and adaptively adjust the error constraint boundaries. This function ensures that tracking errors converge within desired bounds while avoiding singularity issues during formation reconfiguration, addressing the limitations of traditional fixed-boundary prescribed performance control. Second, an FEWNN is proposed to approximate the lumped uncertainty and is embedded into the reinforcement learning strategy for solving the Hamilton-Jacobi-Bellman (HJB) equation. Compared with the traditional radial basis function neural network (RBFNN), the proposed FEWNN incorporates the dynamic memory feature of the Elman neural network, which can effectively enhance the function approximation capability. Third, an RL-based optimal formation controller with self-adjusting prescribed performance is designed. Compared with traditional optimal control methods that lack the prescribed performance constraint, the proposed controller can not only achieve optimal control but also guarantee the steady-state performance and precise trajectory tracking. Finally, the effectiveness of the proposed algorithm is validated through Lyapunov stability theory and numerical simulation analyses.