Tactile-Based Autonomous Cable Manipulation: Pose Estimation and Connector Verification

Abstract Automating dexterous tasks such as cable manipulation using robots has proven technically challenging. While vision systems and specialty grippers are available, they often prove unreliable in occlusion-heavy settings typical of cable manipulation tasks like cable tracing or electrical connectivity tests, and they lack the versatility offered by universal grippers. In this paper, we present two contributions related to cable manipulation in manufacturing operations. First, we show an approach for cable pose estimation that uses only in-hand tactile data from fingertip 3D force-sensing tactile sensors. We created a custom gripper and experimental testbed to collect a dataset of tactile cable grasps. We then trained a Long Short-Term Memory (LSTM) network to predict local cable pose and curvature within the grasp. Our second contribution is a classifier that uses tactile data while pulling on a wire to predict whether the wire is correctly seated in its electrical connector. Cable-pull tests were conducted on a testbed with variable cable attachment strengths. This method of tactile perception may better enable autonomous cable routing and contribute to a less subjective standard for cable-pull tests compared to current practices.