Learning friction compensation in robot manipulators

Abstract

It is difficult to represent the nonlinear characteristics of friction in terms of a mathematical model. An alternative approach of using a neural network to learn the uncertainties in the friction torque of robot manipulators is proposed. Furthermore a true teaching signal for learning the uncertainties is derived. After learning, the neural network is capable of reproducing the training data. It is then embedded in the structure of a joint torque perturbation observer to compensate for the uncertainties in friction. As a result, an accurate estimate of the joint reaction torque during electronic component insertion by a SCARA robot can be deduced. This approach offers distinct advantages over the conventional method of using a structured friction model.<>