Composite Error Learning Robot Control Using Discontinuous Lyapunov Analysis
Yongping Pan, Kai Guo, Alexey Bobtsov, Chenguang Yang, Haoyong Yu
Abstract
A feedback-error learning (FEL) framework, which is characterized by internal dynamics modeling and hybrid feedback–feedforward (HFF) control, provides a computational model for motor learning control in the cerebellum. For FEL-based adaptive robot control, closed-loop stability and parameter convergence involve in stringent conditions, such as high-gain feedback and persistent excitation. This article proposes a composite error learning framework for adaptive robot control under discontinuous friction, where an HFF control structure with forward and inverse models is introduced to mimic the cerebellar motor learning control mechanism, and a composite learning technique with memory regressor extension is employed to capture the robot dynamics. Using discontinuous Lyapunov analysis with Filippov's differential inclusion, we rigorously prove that semiglobal stability of the closed-loop system is ensured without high feedback gains, and exponential parameter convergence (implying accurate robot modeling) is guaranteed by a weakened condition of interval excitation. Experiments on an industrial robot manipulator have demonstrated the effectiveness and superiority of the proposed method.