Skip to main content

Research Repository

Advanced Search

X-TCS: Accuracy-based learning classifier system robotics

Studley, Matthew; Bull, Larry


Profile Image

Dr Matthew Studley
Professor of Ethics & Technology/School Director (Research & Enterprise)

Lawrence Bull
School Director (Research & Enterprise) and Professor


Although most learning classifier system (LCS) research uses the accuracy-based XCS, it had never been used to control a physical robot before. In comparison to purely evolutionary or purely reinforcement learning approaches, an LCS should be faster to learn than the former, and require less operator input and additional techniques than the latter. Increased learning speed removes the need to build simulations, and all learning can take place in situ on the robot. Learning is online and continuous, not 'learn, then perform'.

This paper presents some simple extensions to XCS which enable it to learn to optimally control a robot in a simple task. There is little need for tuning of the algorithm, less need to make decisions a priori to guide the solution of the problem, and the robot learns optimal behaviour in hours rather than the week previously reported with evolutionary techniques in a similar problem.


Studley, M., & Bull, L. (2005). X-TCS: Accuracy-based learning classifier system robotics.

Journal Article Type Article
Publication Date Sep 5, 2005
Journal 2005 IEEE Congress on Evolutionary Computation
Peer Reviewed Peer Reviewed
Volume 3
Pages 2099-2106
ISBN 0780393635
Keywords robotics, learning classifier systems
Public URL
Publisher URL