A method is presented for calibrating and compensating for the kinematic errors in robot manipulators. A method of selecting a set of independent kinematic errors for modeling any geometric errors in a manipulator's structure is developed. A calibration algorithm is presented for finding the values of these kinematic errors by measuring the end-effector position. These kinematic errors are experimentally determined for a PUMA 560 six-joint manipulator. Two general-purpose compensation algorithms are developed and the improvement in the Cartesian position of the end-effector is experimentally measured and these results are presented. The results show that the positioning accuracy of a robot manipulator can be substantially improved using a relatively simple technique for measuring the Cartesian position of a tool attached to the end of the manipulator.>

This paper presents a method of calibrating and compensating for the kinematic errors in robot manipulators. A method of selecting a set of independent kinematic errors for modeling any geometric errors in a manipulator's structure is developed. A calibration algorithm is presented for finding the values of these kinematic errors by measuring the end-effector Cartesian position. These kinematic errors are experimentally determined for a PUMA 560 six joint manipulator. Two general purpose compensation algorithms are developed and the improvement in the Cartesian position of the end-effector is experimentally measured and these results are presented.

The positioning accuracy problem of robot manipulators has long been one of the principal concerns of robot design and control. In a previous work, a linear model that described the robot positioning accuracy due to kinematic errors was developed. However, the previous work considered only the small errors by ignoring the higher order terms and did not address the special case of two consecutive parallel joints. In this work a more detailed model is given that applies to consecutive parallel joints and includes the second-order terms. By comparing the results of the linear model and the second-order model, the accuracy of the linear model can be evaluated for a given manipulator and range of input kinematic errors. The error envelopes obtained using the linear model and the developed second order model for the Puma 560 are plotted and compared for various sets of input kinematic errors. A comparison of the computation complexity for the two models is also given.