Artificial Neural Network Guided Compensation of Nonlinear Payload and Wear Effects for Industrial Robots

Abstract

The absolute accuracy of industrial robots is influ-enced by numerous geometric and non-geometric errors. Most state-of-the-art calibration and compensation methods consider only the geometric errors and neglect the non-geometric ones. In this paper, a hybrid compensation approach is proposed that combines well-known kinematic models with a model-free data-driven component using a neural network. This allows the use of established calibration methods for the geometric influences captured in the kinematic model to improve the non-geometric error compensation by the neural network. The proposed approach is applied in two use cases: payload and wear compensation. Simulations and real experiments show the improved absolute accuracy of the hybrid compensation approach compared to classical calibration methods.