Reinforcement learning-based affine formation maneuver control of autonomous surface vehicles with safety constraints under time-varying actuator faults

As a novel strategy, the affine formation maneuver control (AFMC) employs stress matrices and affine transformation theory to simultaneously realize various leader–follower formation maneuvers including collinearity, rotation, scaling, shearing, and translation. So far, only a few publications have reported the AFMC of multiple autonomous surface vehicles (ASVs). This study aims to explore the leader–follower reinforcement learning-based fractional-order sliding mode AFMC (RL-based FOSMAFMC) approach for ASVs with safety constraints subjected to time-varying actuator faults. A barrier Lyapunov function (BLF) is utilized to keep the ASVs operating within a constrained safety range. The problem of loss of control direction due to the time-varying actuator effectiveness faults is addressed using the Nussbaum function. The RL is formulated using the actor–critic framework. The actor approximates the uncertain nonlinearities and time-varying bias faults, whereas the critic system assesses the control action while minimizing a long-term cost function. The leader–follower AFMC uses a fractional-order sliding mode surface to enhance tracking accuracy and robustness. A Lyapunov function is used to guarantee the boundedness of the closed-loop system. Simulation experiments and performance comparisons are carried out to demonstrate the capability of the proposed controller.