TCB Publications - Abstract

Kakali Sarkar and Klaus Schulten. Topology representing network in robotics. In J. Leo van Hemmen, Eytan Domany, and Klaus Schulten, editors, Models of Neural Networks, volume 3 of Physics of Neural Networks, pp. 281-302. Springer-Verlag, New York, 1996.

SARK96 We consider the visually guided control of the grasping movements of a highly hysteretic five-joint pneumatic robot arm. For this purpose we apply a modified version of the so-called topology representing network algorithm, a vector quantization algorithm which learns to represent also neighborhood relationships. Notion of neighborhood relationships allowed us to average over the behaviour of neurons which represent similar tasks, both during the training and in generating control signals in the mature state. Based on visual information provided by two cameras, the robot learns to position and orient its end effector properly for the object to be grasped. For simplicity we consider grasping of cylindrical objects only. The control comprises two stages. At a first stage, the end effector approaches the side of the cylinder facing the robot base, at a second stage, the end effector grasps the cylinder. Training of the first stage involves a brief episode of supervised learning to prime the network. The control is achieved through a training which yields a vector quantized representation of a zero-order signal of joint pressures and a first order correction through Jacobian tensors which relate the error, expressed in terms of camera coordinates, to corrective joint pressures. The network is trained satisfactorily after about 300 trial movements, with a residual average error of 1.35 camera pixels. Beside a demonstration of the technical feasibility of control through topology representing networks this article provides a tutorial for technical applications of such networks. The algorithm behind topology representing network, its training and employment for task control is described in complete detail to provide the reader with a comprehensive view of this important class of neural networks in the context of a technical application.

Download Full Text

The manuscripts available on our site are provided for your personal use only and may not be retransmitted or redistributed without written permissions from the paper's publisher and author. You may not upload any of this site's material to any public server, on-line service, network, or bulletin board without prior written permission from the publisher and author. You may not make copies for any commercial purpose. Reproduction or storage of materials retrieved from this web site is subject to the U.S. Copyright Act of 1976, Title 17 U.S.C.

Download full text: PDF ( 1.4MB)