Designing G-optimal experiments for robot dynamics identification

Abstract

A common way to identify the robot dynamics is to use a linear model structure in relation to the parameters and standard least-square (LS) techniques. These techniques are sensitive to noise measurements and error modeling, so that it is necessary that the trajectory is "exciting enough" to provide an accurate estimate of the parameters. This paper addresses criteria of exciting trajectories for robot dynamics identification. An optimum criterion, which corresponds to the G-optimum criteria known from the literature, is derived from the "true system" concept. It also shows that through exploitation of the model structure properties, the design of G-optimal experiments are made considerably easier. A G-optimal experiment for a two degrees of freedom SCARA robot is designed to show the simplicity and efficiency of these approaches.