Feedback linearization using CMAC neural networks (Q1298257): Difference between revisions

The objective of the paper is to achieve tracking control of a class of unknown nonlinear dynamical systems using the cerebellar model articulation controller (CMAC). In fact, in recent years, learning-based control using neural networks (NN) has emerged as an alternative to the standard adaptive control techniques. The learning is then accomplished in off-line fashion. In the paper the study of the application of CMAC NNs in closed-loop of feedback linearizable systems is presented. To realize this, simple receptive-field functions that are easy computable, in fact, second-order splines, are used. To materialize such a problem, very restrictive conditions in relation to real situations must be available, e.g., to linearize a complex nonlinear system is very questionable and cannot be done in closed form. Further, the Lyapunov function for such nonlinear systems cannot be constructed in general. Also, the authors apply the method to very simple, academic examples. So it seems very probable, that the method would not work effectively for more complicated, in a sense, more realistic systems. It is peculiar that the title of the paper is not correct as the feedback linearization is not done using CMAC NNs but introduced ad hoc as a state-space representation in the controllable canonical form.