The generation of motion kinematics using a time-delay neural network

This paper addresses the simulation and generation of kinematic figure movement using a time delay neural network. This method potentially provides a means of generating realistic motion for robot manipulation, artificial limb and animated environments. The results obtained have been encouraging covering mathematical functions which approximate repetitive and revolving motions. Actual body measurements were obtained for the final experiment, which yielded the generation of a three-joint arm sequence. The shoulder elbow and wrist motions of the human arm resulted in an impressive average test error of 0.0322. These results have shown potential for the production of kinematic motions for a host of applications.