TCS Learning Classifier System Controller on a Real Robot

Abstract

To date there have been few implementation of Holland's Learning Classifier System (LCS) on real robots. The paper introduces a Temporal Classifier System (TCS), an LCS derived from Wilson's ZCS. Traditional LCS have the ability to generalise over the state action-space of a reinforcement learning problem using evolutionary techniques. In TCS this generalisation ability can also be used to determine the state divisions in the state space considered by the LCS. TCS also implements components from Semi-Mark-Decision Process (SMDP) theory to weight the influence of time on the reward functions of the LCS. A simple light-seeking task on a real robot platform using TCS is presented which demonstrates desirable adaptive characteristics for the use of LCS on real robots.