Vision-driven High Precision Positioning Method for Bracket Assembly with Industrial Robot

Abstract

The quality of the bracket assembly is an important guarantee for the successful launch of the rocket. The assembly accuracy of the bracket installed inside rockets is affected by bracket pose estimation, robot grasping, and positioning. We have proposed a recognition and pose estimation algorithm that integrates 2D and 3D vision to improve the bracket assembly accuracy. A position-based visual servo bracket grasping method is used to avoid the poor absolute positioning accuracy of the industrial robot. Then a small field-of-view vision system was established to achieve precise positioning of the bracket. We also developed a quick and precise hole detection algorithm based on YOLOv5 and caliper vision operator to compensate for the absolute positioning error of industrial robots. With the help of a vision servo grasping strategy and high precision re-positioning algorithm, the bracket's assembly hole positioning accuracy can finally reach 0.05mm. It is worth mentioning that the methods proposed in this paper have been successfully applied in the flexible automatic assembly line of the brackets used in rockets.