Ｑ学習におけるＱ値の追加と統合の提案

抄録

This article presents a method of addition and integration of q-values, and aims to improve the performance of learning. For this purpose, we constructed a computational model of Q-learning agents employing the three types of action selections (i.e., straight ahead, left rotation, and right rotation) with two inputs of angle, and distance to a target, and compare the simulation results of lattice-shaped formation. We add many q-values on state space, and integrate which has the same action selections. In this paper, we use LBG algorithm to collect state transition vectors.