Reinforcement learning control of a single-link flexible robotic manipulator

In this study, the authors focus on the reinforcement learning control of a single-link flexible manipulator and attempt to suppress the vibration due to its flexibility and lightweight structure. The assumed mode method and the Lagrange's equation are adopted in modelling to enhance the satisfaction of precision. Two radial basis function neural networks (NNs) are employed in the designed control algorithm, actor NN for generating a policy and critic NN for evaluating the cost-to-go. Rigorous stability of the system has been proven via Lyapunov's direct method. Through Matlab simulation and experiment on the Quanser flexible link platform, the superiority and feasibility of the reinforcement learning control are verified.