Optimum pose measurements for kinematic parameters identification

Abstract

A six degree of freedom robot manipulator arm, a PUMA 560, is calibrated using random subsets of available experimental calibration data. Some of these subsets produce good calibration results motivating the search for an optimum calibration procedure which will use a small number of poses. Statistical analysis of the joint excursions and end effector position variation in both "good" and "bad" subsets of poses were conducted. No statistical differences between them was discovered. The condition number of the Jacobian matrix is investigated as a potential measure of the accuracy which may be obtained from the subset under consideration. The condition number thus obtained contained too much variability to be a reliable predictor of accuracy. A computer simulation was conducted using a numerical optimizer to select the joint angles to be used for calibration. The optimizer studies failed to find an optimum set of poses for calibration. The conclusions of these studies is that there is no optimum set of poses to be used for calibration. As alternative hypothesis, that the resultant calibration accuracy depends only upon the accuracy of the measurement taken, seems to be proven.