Cable manipulation with partially occluded vision feedback

Abstract

Installation of a deformable object likes cables is crucial for robotic manipulation, which includes several key problems: 1) estimating the state of a cable; 2) tracking a cable when it is partially occluded; 3) manipulation considering the environment. To address these problems, we proposed a robot framework based on vision. To estimate the state of the cable, our method 1) Uses YOLO to detect the crossings and cable ends; 2) Extracts the contour of a cable from its projection on an RGB image. Our method uses an algorithm based on Coherent Point Drift (CPD) to track a cable. To install a cable efficiently, we designed a gripper equipped with a motor for sliding a cable in hand. Our method does not rely on any prior knowledge. This feature helps us adapt to different situations without too much preliminary preparation. Despite that, our method successfully verified by experiments that mimic the practical installation environments.