Abstract
This paper describes an interactive neural network based system for specifying robotic tasks using virtual tools. This virtual environment allows an operator to reach into a live video scene and direct robots to use corresponding real tools to carry out complex metal finishing tasks. The virtual tool concept provides a human-machine interface that is robust to unanticipated developments and tunable to the specific requirements of a particular task. This interactive specification concept is applied to robotic deburring processes. A function is formulated to map the end-effector position of this robot to corresponding set of joint angles through a neural network learning process obtained through examples. The experimental result of such a system that has been implemented on the Mitsubishi RV-M1 robot shows the efficiency of the approach and its potential for use in virtual reality based interactive robotics.
