Real-time neurofuzzy control for an underactuated robot

Abstract

We use a neurofuzzy approach, the NEFCON model, to generate and optimize a fuzzy controller for real-time control of an underactuated robot: the Pendubot, which consists of a two link inverted pendulum actuated only at the first join. The NEFCON learning algorithm is able to learn fuzzy rules as well as fuzzy sets. We present the results of the learning process for a fuzzy controller to balance the Pendubot in its highest inverted position, simulation results, and real-time results. The extension of this work to include the learning process of a swing-up procedure is in progress.