Industrial Robot Hand–Eye Calibration Combining Data Augmentation and Actor-Critic Network

Abstract

The accuracy of hand-eye calibration significantly affects the performance of vision-guided robotic tasks. The main problems faced by hand-eye calibration are the tedious process of data collection and the low accuracy of collected data. To reduce the workload and improve accuracy, an industrial robot hand-eye calibration method combining a data augmentation strategy and an Actor-Critic network is proposed. In the proposed method, the parameters in the hand-eye transformation matrix are described by an X-Y-Z fixed angle representation. Then a parameter optimization strategy is formulated. The parameters are optimized using the Actor-Critic network. Data augmentation was performed based on ellipse features. Combined with Halcon image processing software, a bulk of unlabeled data is extracted from a single frame image of an ellipse calibration target. This method of data collection simplifies data acquisition and ensures the accuracy of data. According to the results of the accuracy analysis experiment, the proposed method significantly obtains a more accurate conversion relationship.