Robot calibration is very important to improve the accuracy of robot manipulators. However, the calibration procedure is very time consuming and laborious work for users.
In this paper, we propose a method of relative error compensation to make the calibration procedure easier. The method is completed by a Pi-Sigma network architecture which has sufficient capability to approximate the relative relationship between the accuracy compensations and robot configurations while maintaining an efficient network learning ability.
By simulation and experiment of 4-DOF SCARA type robot, KIRO-3, it is shown that both the error of joint angles and the positioning error of end effector are drop to 15%. These results are similar to those of other calibration methods, but the number of measurement is remarkably decreased by the suggested compensation method.