Computational Neuroscience for Advancing Artificial Intelligence: Models, Methods and Applications Eduardo Alonso City University London, UK Esther Mondragón Centre for Computational and Animal Learning Research, UK Medical inforMation science reference Hershey • New York Director of Editorial Content: Kristin Klinger Director of Book Publications: Julia Mosemann Acquisitions Editor: Lindsay Johnston Development Editor: Joel Gamon Publishing Assistant: Jamie Snavely Typesetter: Michael Brehm Production Editor: Jamie Snavely Cover Design: Lisa Tosheff Published in the United States of America by Medical Information Science Reference (an imprint of IGI Global) 701 E. Chocolate Avenue Hershey PA 17033 Tel: 717-533-8845 Fax: 717-533-8661 E-mail: [email protected] Web site: http://www.igi-global.com Copyright © 2011 by IGI Global. All rights reserved. No part of this publication may be reproduced, stored or distributed in any form or by any means, electronic or mechanical, including photocopying, without written permission from the publisher. Product or company names used in this set are for identification purposes only. Inclusion of the names of the products or com- panies does not indicate a claim of ownership by IGI Global of the trademark or registered trademark. Library of Congress Cataloging-in-Publication Data Computational neuroscience for advancing artificial intelligence : models, methods and applications / Eduardo Alonso and Esther Mondragon, editors. p. cm. Summary: "This book argues that computational models in behavioral neuroscience must be taken with caution, and advocates for the study of mathematical models of existing theories as complementary to neuro- psychological models and computational models"-- Provided by publisher. Includes bibliographical references and index. ISBN 978-1-60960-021-1 (hardcover) -- ISBN 978-1-60960-023-5 (ebook) 1. Computational neuroscience. 2. Neurosciences--Mathematical models. 3. Artificial intelligence. I. Alonso, Eduardo, 1967- II. Mondragon, Esther, 1965- QP357.5.C634 2011 612.80285'63--dc22 2010018588 British Cataloguing in Publication Data A Cataloguing in Publication record for this book is available from the British Library. All work contributed to this book is new, previously-unpublished material. The views expressed in this book are those of the authors, but not necessarily of the publisher. 214 Chapter 10 Artificial Neural Systems for Robots Phil Husbands University of Sussex, UK Andy Philippides University of Sussex, UK Anil K. Seth University of Sussex, UK ABSTRACT This chapter reviews the use of neural systems in robotics, with particular emphasis on strongly biologi- cally inspired neural networks and methods. As well as describing work at the research frontiers, the paper provides some historical background in order to clarify the motivations and scope of work in this field. There are two major sections that make up the bulk of the chapter: one surveying the application of artificial neural systems to robot control, and one describing the use of robots as tools in neurosci- ence. The former concentrates on biologically derived neural architectures and methods used to drive robot behaviours, and the latter introduces a closely related area of research where robotic models are used as tools to study neural mechanisms underlying the generation of adaptive behaviour in animals and humans. INTRODUCTION isolated island. This often dark and dystopian play was a world-wide smash hit capturing the popular The idea of robots is rooted in the dreams and myths imagination as well as sparking much intellectual of old Europe and Ancient Greece in which me- debate (Horáková and Kelemen 2007). In the chanical men acted as either slaves or oppressors. process it helped forge the predominant image Indeed the modern word ‘robot’ was introduced of robots that now permeates our culture – that in Karel Čapek’s 1921 play R.U.R. (Rossum’s of life-like artificial creatures. Universal Robots) which told of artificial men In fact this image refers to what we now call manufactured as a source of cheap labour on an autonomous robots. In contrast to machines that perform precise repetitive tasks ad nauseam (e.g. robots used in manufacturing production lines), DOI: 10.4018/978-1-60960-021-1.ch010 Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. Artificial Neural Systems for Robots autonomous robots are required to behave in an (2002), Bekey (2005), Bekey and Goldberg(1993) appropriate way in a very broad range of circum- and Floreano and Mattiussi (2008). stances. Like biological creatures, their behaviour must be self-generated, making use of sensory information to moderate their responses to the HISTORY world. With these animal connotations, it is no surprise that when simple autonomous robots be- Despite the construction of many ingenious me- came a reality in the mid 20th century, researchers chanical automata over the centuries (including looked to natural nervous systems for inspiration chess playing Turks and flatulent ducks (Wood as they developed their early control systems. 2003)), it was not until the 1930s that devices The huge advances in our understanding of recognizable as robots (in the present day sense real neural networks over the past few decades, of the term) appeared. Early mobile robots, coupled with our development of increasingly such as Thomas Ross’s ‘robot rat’, completed sophisticated artificial varieties, has led to sig- in 1935, were designed for narrowly focused nificant growth in research on artificial neural single behaviours (often maze running) and em- systems in robotics. This chapter concentrates on ployed highly specific mechanisms to achieve the use of explicitly biologically inspired artificial their intended task (Cordeschi, 2002). These neural networks in autonomous robotics, review- ‘artificial animals’ inspired what were probably ing key applications and forms of neural system the very first examples of more general mobile used. There are a variety of drives underlying such autonomous robots – W. Grey Walter’s tortoises work, ranging from straightforward engineering (Walter, 1950). These robots were also the first to motivations – the desire to build better, smarter employ an early form of neural network as their machines – to purely scientific ones, particularly artificial nervous system. They were born out of in the use of robots as tools to study mechanisms the cybernetics movement, a highly interdisci- underlying the generation of adaptive behaviour plinary endeavour – drawing together pioneers in animals and humans. Often varying degrees of of computing and modern neuroscience – which both these motivations are present in any particular was the forerunner of much of contemporary AI project. This chapter will highlight work that falls and robotics, and the origin of artificial neural at the two extremes of this spectrum, as well as networks and evolutionary computing, as well much that rests in between. as control and information theory (Boden, 2006; The next section gives some historical back- Husbands et al., 2008). ground to the area in order to motivate the remain- In 1949, Walter, a neurologist and cyberneticist der of the chapter. Following that, there are two based at the Burden Institute in Bristol, UK, who major sections that make up the bulk of the paper: was also a world leader in EEG research, com- one on the application of artificial neural systems pleted a pair of revolutionary machines he called to robot control, and one on the use of robots as ‘tortoises’. The devices were three-wheeled and tools in neuroscience. The chapter closes with a sported a protective ‘shell’ (see Figure 1). They general discussion of prospects for such research. had a light sensor, touch sensor, propulsion motor, There is not enough space to give a comprehensive steering motor, and an electronic valve (vacuum review of all major work in the area, instead we tube) based analogue ‘nervous system’. Whereas have concentrated on a few important topics that earlier machines such as Ross’s were constrained to give a good overall flavour of research in the field. run on rails, the tortoises could roam freely around For more detailed coverage see e.g. Siciliano and their environment. Walter’s intention was to show Khatib (2008), Webb and Consi (2001), Ayers et al. that, contrary to the prevailing opinion at the time, 215 Artificial Neural Systems for Robots Figure 1. Grey Walter watches one of his tortoises even a very simple nervous system (the tortoises push aside some wooden blocks on its way back had two artificial neurons) could generate complex to its recharging hutch. Circa 1952 behaviour as long as there was a sufficiently large number of possible interaction patterns between the neurons (Walter, 1950). By studying whole embodied sensorimotor systems acting in the real world, he was pioneering a style of research that was to become very prominent in AI many years later, and remains so today (Brooks, 1999; Hol- land, 2003). Between Easter 1948 and Christmas 1949, he built the first tortoises, Elmer and Elsie. They were rather unreliable and required frequent attention. In 1951, his technician, Mr. W.J. ‘Bunny’ Warren, designed and built six new tortoises to a much higher standard. Three of these tortoises were exhibited at the Festival of Britain in 1951; others were regularly demonstrated in public throughout the 1950s. The robots were capable of phototaxis (steering