The Series Elastic Gripper Design, Object Detection, and Recognition by Touch

Abstract

Abstract In recent years, robotic applications have been improved for better object manipulation and collaboration with human. With this motivation, the detection of objects has been studied with a series elastic parallel gripper by simple touching in case of no visual data available. A series elastic gripper, capable of detecting geometric properties of objects, is designed using only elastic elements and absolute encoders instead of tactile or force/torque sensors. The external force calculation is achieved by employing an estimation algorithm. Different objects are selected for trials for recognition. A deep neural network (DNN) model is trained by synthetic data extracted from standard tessellation language (STL) file of selected objects. For experimental setup, the series elastic parallel gripper is mounted on a Staubli RX160 robot arm and objects are placed in pre-determined locations in the workspace. All objects are successfully recognized using the gripper, force estimation, and the DNN model. The best DNN model is capable of recognizing different objects with the average prediction value ranging from 71% to 98%. Hence, the proposed design of the gripper and the algorithm achieved the recognition of selected objects without the need for additional force/torque or tactile sensors.