A Pipe Defect Localization Method for Robotic Repair Manipulation

Current manual-based oil and gas pipe repair operations are dangerous and need to be replaced by robots. Although robots are attempted in many complex disposal tasks, they mostly work under the control of humans, which lacks autonomy. In this paper, we propose an effective vision localization algorithm for pipe surface defects, which combines 3D reconstruction of pipe surface, the principle of camera projection, and the YOLOv5 detector to realize the 6D pose estimation of pipe defect from point cloud data and color image. A series of vision localization experiments are conducted from multiple imaging perspectives to intuitively evaluate the localization algorithm's effectiveness. Moreover, a two-stage manipulation, utilizing the vision localization to approach and the force sensor to contact the defect, is proposed to evaluate this algorithm with digital-form estimation errors. One hundred manipulations are conducted, achieving a 95% success rate. And the translation and rotation errors are smaller than 1.5mm and 0.02rad respectively. The experimental results show that this proposed method can effectively estimate the pose of pipe surface defect for robotic repair manipulation and alleviate the lack of pipe surface defect samples, even in the background clustered environment.