A Self-learning Robust Swing-up algorithm

Abstract

A novel swing-up algorithm for an inverted pendulum under a restricted cart track length is introduced. The algorithm achieves the swing-up task by means of a control law that uses a finite input set. The algorithm neither knows nor uses values of cart and pendulum masses, or pendulum length. In the swing-up process, it self-learns a weight that is used in normalizing the energy injected to the pendulum. This weight makes it possible for the algorithm to determine the number of sampling periods needed to reach the upright equilibrium state. Also, by predicting the multi-step ahead cart position, it gives rise to a control law that keeps the cart within the track and makes the pendulum reach the neighbourhood of the upright equilibrium state in a smooth manner. The validity and robustness of the swing-up algorithm are verified through laboratory experiments.