Biomimetic composite learning for robot motion control

This paper focuses on biomimetic hybrid feedback feedforward (HFF) learning for robot motion control. Existing HFF robot motion control approaches have a major problem that accurate estimation of the robotic dynamics, which is crucial for mimicking biological control, is not taken into account. In this study, a composite learning technique is presented to achieve fast and accurate estimation of the robotic dynamics in robot motion control without a stringent persistent-excitation (PE) condition. The control architecture includes a proportional-derivative (PD) controller acting as a feedback servo machine and an estimation model acting as a feedforward predictive machine. In the composite learning, a time-interval integral of a filtered regressor is utilized to construct a prediction error, and both the prediction error and a filtered tracking error are used to update parametric estimates. Semiglobal exponential stability of the closed-loop system is rigorously established under an interval-excitation (IE) condition which is much weaker than the PE condition. Simulation results have been provided to demonstrate effectiveness and superiority of the proposed approach.