Image-based robot motion simulation

Abstract

This paper describes an image-based robot animation technique. The technique utilizes eigen space method to generate compact representations of images from a set of reference images. A wavelet neural network is used to learn the relationship between robot joints and image representations. The learning algorithm is Quasi-Newton algorithm with Levenberg–Marquardt modifications. The trajectory in the joint space is first planned to generate a joint sequence. A corresponding sequence of representations can be obtained through the trained wavelet neural network and the image sequence can be synthesized using the result of eigen space method. No a priori models of robot and camera or any calibrations are needed. The virtual motion of the robot UP6 is demonstrated.