Human—Hand Posture Classification For Robotic Teleoperation Using Wearable Sensor

Abstract

The aim of this work is to control a teleoperated robotic gripper to grasp objects of different shapes using an efficient posture classification algorithm. Two Inertial Measurement Unit sensors are attached to a wearable glove and using it the human operator performs 5 different postures related to grasping. The feature selected for experimentation is Unit Quaternion. Classification is done with 2 different classifier algorithms, Neural Networks and Random Forests. This is followed by a basic mapping to a 10 degrees of freedom robotic hand. A comparative study of the classifier algorithms is carried out to select the most efficient algorithm for sensor based posture recognition.