Hybrid Calibration of Industrial Robot Considering Payload Variation

Abstract

Absolute accuracy of industrial robot is required for most of industrial applications. However, positioning errors of several millimeters are induced by many factors. Hybrid calibration, combining analytical model and learning-based regression, can compensate for most of the positioning error, including payload effects. However, when the payload changes, hybrid calibration has to be performed again. In this paper, hybrid calibration is applied on an industrial robot in two different sub-workspaces, with two different payloads. The results of this method have been compared to other calibration approaches, and highlight that hybrid calibration provides a higher final accuracy. Moreover, two data-efficient and pragmatic approaches are proposed, to address the issue of changing payload. Both methods are based on hybrid calibration. The first one uses previously-acquired knowledge to drastically reduce the number of measurements necessary to update a trained learning model with another payload. The second one uses a model trained separately for two different payloads and interpolates the outputs to compensate for new payloads without any additional measurement. The datasets used are available at: https://doi. org/10.57745/DWUC0H. The methods have been experimentally validated using a compensation algorithm and compared to other approaches, and show that the positioning error can be reduced by 95%.