抄録
This paper describes how to develop a closed-loop controller generating competitive motions in the coupled inverted pendula model. For this purpose, we proposed an intelligent controller consisting of a classifier, a selector and an impulse generator. The classifier is identified by off-line learning of impulse responses of the model to output the indexes of equilibrium points starting from given initial conditions. The selector binarizes a series of the indexes by replacing the desired index with 1 and the others with 0. The impulse generator outputs an impulsive force when the selector outputs 1. Then, applying the obtained intelligent controller into each of pendulums of our model independently, we numerically demonstrate that the controlled model can produce competitive motions. However, since the performance of our method depends on resolution of the training data in O(n^4)-order, improving performance may cause computational time explosion. To solve this problem, we finally propose a simple method of adding a time delay element by which the control performance can improve with low resolution training data.
