Title:	EVOLUTION OF ARTIFICIAL NEURAL NETWORKS FOR ROBOT CONTROL USING SPECIATION AND COMPLEXITY MEASURES
Author(s):	Thomas Jorgensen , Barry Haynes
ISBN:	978-972-8924-60-7
Editors:	António Palma dos Reis
Year:	2008
Edition:	Single
Keywords:	Complexification, ANN, neural complexity, robot, anthropomorphisation
Type:	Short Paper
First Page:	198
Last Page:	201
Language:	English
Cover:
Full Contents:	click to dowload
Paper Abstract:	This paper describes the preliminary results obtained using a novel methodology for complexifying and growing artificial neural network weights and topology using a neural complexity measure and speciation. The use of continuous complexification combined with speciation reduces the anthropomorphisation, i.e. the constraints imposed by a designer, of the artificial neural network compared with other standard methodologies. The continuously complexified neural networks are tested and simulated in a simple three in a row robot alignment task. This research investigates the developing relationship between brain/controller and environment, through continuous complexification and quantitative measurements of complexity. The preliminary results obtained indicate that using speciation and neural complexity measures reduces anthropomorphisation and speeds up the evolutionary process compared other methodologies. In addition, the complexity of the evolved controller matches and reflects the complexity of the environment in which it develops, better than fixed structure networks.
	Go Back