Consideration of multiple objectives in neural learning classifier systems

Abstract

For effective use in a number of problem domains Learning Classifier Systems must be able to manage multiple objectives. This paper explicitly considers the case of developing the controller for a simulated mobile autonomous robot which must achieve a given task whilst maintaining sufficient battery power. A form of Learning Classifier System in which each rule is represented by an artificial neural network is used. Results are presented which show it is possible to solve both objectives when the energy level is presented as an input along with sensor data. A more realistic, and hence more complex, version of the basic scenario is then investigated.