Learning motion from images

Abstract

Describes a method of determining a robot end-effector's motion required to achieve a standard position and orientation relative to an object through learning. By using a back-propagation network, the authors establish the direct mapping from 'what is seen' to 'what should be done'. The method does not need camera calibration, nor hand-eye calibration, nor explicit object model. Some general rules for correct learning are presented. A recursive scheme of movement control is designed with convergence proof. The method is simulated on an application object and shows promising application potential.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>