A reinforcement learning system incorporating within the concept of value

Abstract

In the environment with uncertainty, to continue the appropriate behavior, the agent has to be equipped with means to judge whether change of the environment caused by his own behavior is good or not. In this paper we propose a novel reinforcement learning system which incorporates the above means, that is, a value system. Through computer simulations using maze problems, it is verified that the proposed system is valid by comparing with the conventional reinforcement learning system.