A novel pose estimation method for robot threaded assembly pre-alignment based on binocular vision

Threaded assembly plays a critical role in industrial manufacturing; however, achieving a fully automated threaded assembly remains challenging. In this study, an automatic robot thread assembly system based on binocular vision was developed, along with a novel approach for spatial circle pose estimation. Notably, this method utilises the chamfering circle of the threaded hole as the recognition target and achieves precise pose estimation without requiring any prior knowledge, from a geometric perspective. Utilising only a chord of the ellipse projected from the circular feature of the threaded hole, the method effectively addresses the traditional reliance on complete target features. Additionally, it avoids the need for point cloud fitting, which is commonly used in conventional 3D pose estimation, thereby significantly reducing computational complexity and improving both efficiency and accuracy. An innovative method for verifying the spatial circle positioning accuracy is proposed based on the calibration plate coordinate system. The proposed method achieved position error ranges of [0.0419, 0.0837], [-0.0864, 0.0148], and [-0.0434, 0.0286] in mm along the x, y, and z axes, respectively. Furthermore, the orientation error ranged from 0.649° - 1.752° To comprehensively consider the origin of the various errors, a workpiece was designed to conduct robot alignment experiments. The average errors along the x, y, and z axes were -0.23, -0.57, and -0.45 mm, respectively. Overall, the proposed vision measurement method demonstrated excellent pose estimation accuracy and significantly enhanced the automation of robotic threaded assembly processes. This advancement holds great potential for widespread applications in industrial manufacturing environments.