Actor-Critic Reinforcement Learning Control of Flexible Manipulator based on PDE Model

In this study, a reinforcement learning control strategy is introduced about the trajectory tracking task for the flexible robotic arm modelled by partial differential equations (PDEs) with systematic uncertainty terms. A reinforcement learning algorithm based on a structure of actor-critic is introduced for better implementation of the trajectory tracking task.The critic network approximates the cost function to evaluate system performance, while the actor network approximates the system’s uncertainty term, with adaptive laws derived using the gradient descent method. Under the proposed control strategy, the Lyapunov direct method is employed to demonstrate that the closed-loop system exhibits semi-global uniform ultimate boundedness, while the efficacy of the proposed methodology is verified through numerical simulations.