Detecting Slip in Robotic Grasps

This project explores the viability of using machine learning to detect when objects are slipping through the grasp of a robotic hand. To that end, this paper discusses how one might use classification to identify when objects are sliding over the surface of a load cell that could be installed in a robotic hand. It is found that through proper feature selection, slipping loads can be distinguished from arbitrary loads using logistic regression when the axis of slip is known a priori. An approach for learning the axis of slip is also presented, which generalizes the detection scheme for slip in arbitrary, unknown directions.