Feedback error learning and nonlinear adaptive control (Q705594): Difference between revisions

Feedback error learning (FEL) was originally proposed as a computational model of learning motor control in the cerebellum using internal models of the nervous system. The authors of this paper suggest that there is a close relationship between FEL and nonlinear adaptive control algorithms with adaptive feedback linearization. The authors' insight was motivated by their work on an adaptive control framework for advanced statistical learning. In this paper, the authors present an analysis of the formulation and stability properties of FEL for a class of nonlinear systems. The general structure of the control system includes a class of nonlinear many-input, many-output systems of the form \[ \begin{aligned} x & = f(x)+G(x)u\\ z & =h(x)\end{aligned} \] where \(x\in\mathbb R^n\) is a state, \(z\in\mathbb R^p\) is an output, \(u\in\mathbb R^m\) is an input, \(f:\mathbb R^n\to\mathbb R^n\) and \(g:\mathbb R^n\to\mathbb R^{n+m}\) are nonlinear functions and \(h:\mathbb R^n\to\mathbb R^p\) maps states to outputs. Two different formulations are considered: one uses an adaptive state feedback controller, and the other an adaptive feedforward controller. The authors provide a sufficient condition on the choice of feedback gains to guarantee stability of FEL based on a Lyapunov analysis. That this condition is also necessary is related to the strictly positive real (SPR) nature of the tracking error dynamics. Numerical examples are provided that illustrate the stability properties of FEL.