Comparative performance of two types of cerebellar model controllers for controlling robot joint size, learning and generalization

Abstract

Versions of the proportional-integral-derivative (PID) controllers are most widespread type of controllers used to control plants, in achieving satisfactory performance. Versions of PID controller are also used for controlling robotic plants. With increased speed of operation and more stringent performance requirements PID controller alone will not suffice. Control performance improve significantly by adding second controller that computes necessary driving torque from current operating condition, based on robot model. Two types of cerebellum based controllers where tested in acquiring robot model. First one is standard Albus cerebellar model articulation controller (CMAC), and second one is new cerebellar model controller with added functionality for some of cerebellar neuron types, namely Golgi cell, basket cell, and stellate cell (CMACgbs). Physical modeling of robot joint with SimMechanics was used as testing plant for rapid prototyping. Cerebellar controller features addressed are size of the neural network, learning process, and generalization.