An experimental comparison of robot kinematic calibration methods

Abstract

It has been shown that the kinematic calibration can help to improve the accuracy of the industrial robot. This research has involved using two different calibration methods, the CMM and VBAT, to calibrate PUMA 560 manipulator, and studied the relationship between the global and local calibration. The CMM method can measure the complete tool pose information with a special 5-tooling ball end effector, and the VBAT method can provide the partial tool pose measurement by using an automatic tracking camera which works as a single theodolite. In this work, the robot calibration procedure is based on a model which is the extension of the modified Hartenberg-Denavit model. The deviations between the kinematic model used in the robot controller and the actual kinematic structure of a given robot are in terms of the small perturbation of the parameters of the selected model. Since it is assumed that the acquired data will have measurement error, a least squares technique is employed to optimize the solution. Simulations are used to demonstrate that the selected model is adequate to model a six degree-of-freedom manipulator, to study the effect of the measurement error, and to show the data filter can work. The most important results is that the simulations show that the observability index can be misleading if the reliability of the data is not check first, this is never observed in previous work. Finally, the experimental data are used to compare the identified parameters of the two different methods, the prediction error of the global and local calibrations, and the existence of the unmodeled factors. The global and local calibrations results show that there are significant differences between them, and local calibration is an effective way to improve manipulator performance. The parameters identified by using a particular local space data can predict the position of the poses in that local space best, and get worse when the poses are further away the particular local space. The global calibration results show that the parameters identified by using the global space data can predict the position of the poses in the global space within a range which is worse than the results of local calibration in each local space.