Vision Based Deformable Wires Recognition using Point Cloud in Wire Harness Supply

Abstract

Wire harness assembly is one of the essential tasks in the automation industry. However, it is still being completed by human work owing to several difficulties. In particular, the wire, which is the target object, is a deformable object, and is significantly thin. The features of the wire require other approaches that are suitable for working conditions in terms of recognition as well as robot manipulation. The dual-arm robot must preemptively hold the wire before the wire harness assembly. In this study, we presented a methodology for recognizing cables using point clouds obtained from RGB-D sensors on wire harness supply designed to perform this task. In the proposed three stages, we presented a method for distinguishing the sub-cables and connectors of the wires by applying several clustering methods and color similarity comparisons, and calculating the gripping position of the robot. In addition, we verified our method through experiments on several cases.