An Evaluation of Methods for Assessing Robot Kinematic Model Accuracy in the Presence of Noise

Published

March 25, 2024

Author(s)

Mitchell R. Woodside, Patrick Bazzoli, Philip A. Olubodun, J. Adam Nisbett, Helen Qiao, Douglas A. Bristow

Abstract

While there are many works developing methods for modeling and calibrating robot kinematics, assessing the accuracy of those models has received little attention. However, accuracy assessment is critically important for applications where the robot must operate with absolute accuracy over a large region of workspace, such as in robotic machining. When the model of such a system is well calibrated, the remaining deterministic error can be quite complex, owing to complicated gearing errors, deformations, and quasi-static thermal changes. Locating the largest deterministic error requires an exploration over the workspace, but assessing the largest error is complicated by repeatability error and measurement noise. How then to assess the largest error from such a measurement set? This paper evaluates the efficacy of two conventional methods, maximum measured error and outlier rejection, and a novel method based on model invalidation that uses a hypothesis testing framework. A machining robot is used to develop a numerical study for evaluation of these methods under differing magnitude of measurement noise. A high-order kinematic model of the robot is constructed as used as the true robot kinematics, and the workspace for that system is used as the region of interest. A best-fit DH-model is used as the model whose accuracy is to be measured. The study shows that the largest deterministic error can be difficult to locate with just a few percent of points approaching the defining accuracy limit. As expected, the largest measured error provides a poor (over)estimate of the error as noise is increased, but outlier rejection can be equally as bad as rare large deterministic errors can be easy mistaken for low-probability large random error. The novel model invalidation method, however, performs well across all noise levels.

Conference Dates

August 28-September 1, 2024

Conference Location

Bari, IT

Conference Title

2024 International Conference on Automation Science and Engineering (IEEE CASE 2024)

Pub Type

Conferences

Keywords

Assess Robot Accuracy, robot kinematics, measurement error

Citation

Woodside, M. , Bazzoli, P. , Olubodun, P. , Nisbett, J. , Qiao, H. and Bristow, D. (2024), An Evaluation of Methods for Assessing Robot Kinematic Model Accuracy in the Presence of Noise, 2024 International Conference on Automation Science and Engineering (IEEE CASE 2024), Bari, IT (Accessed July 19, 2024)