A neural network for robot control: Cooperation between neural units as a requirement for learning

Thomas Martinetz, Klaus Schulten

Research output: Contribution to journalArticlepeer-review

Abstract

We present a self-organizing neural network for learning visuo-motor coordination and describe its application to the task of learning the end effector positioning of a simulated, three-jointed robot arm. In contrast to a neural network algorithm that we introduced earlier in Refs we now employ visual feedback which enables the robot arm to position its end effector with an accuracy of about 5% of the size of the workspace after only 100 learning steps and with a final accuracy of about 0.06% after 6000 learning steps. Learning proceeds without the need for an external teacher by a sequence of trial movements using input signals from a pair of cameras. The use of visual feedback enables the robot arm not only to adapt to slowly occurring miscalibrations, but also to compensate for sudden changes in its geometry, e.g. when picking up a tool. The results of a simulation and a mathematical analysis of the learning procedure show that cooperation between neural units during the course of learning, incorporated in the network, is essential for the robot arm system to succeed in learning.

Original languageEnglish (US)
Pages (from-to)315-332
Number of pages18
JournalComputers and Electrical Engineering
Volume19
Issue number4
DOIs
StatePublished - Jul 1993

ASJC Scopus subject areas

  • Control and Systems Engineering
  • Computer Science(all)
  • Electrical and Electronic Engineering

Fingerprint

Dive into the research topics of 'A neural network for robot control: Cooperation between neural units as a requirement for learning'. Together they form a unique fingerprint.

Cite this