進化アルゴリズムを用いた移動ロボット自律化に関する研究 : 第1報:MINDSTORMSを使った強化学習の実験

概要

論文の詳細を見る
The purpose of a series of the technical reports is to show applications of evolutionary robotics approach to the development of autonomous mobile robots. The first report describes how to apply the theory of reinforcement learning to a toy robot constructed using MINDSTORMS (LEGO, Inc.). The robot can move along a straight line with a start point and an end point. He goes forward or backward seeking the goal (the end point). But he does not know where the goal is on the line. That is, the line is assumed to be an unknown environment for the robot. When he has reached to the end point, a reward (+10 points) is given to him. But when he has taken the opposite direction at the start point, a negative reward (-10 points) is given. Under the situation, the robot is equipped with a reinforcement learning strategy called Q-learning. The report shows successful goal seeking behaviors and confirms the effectiveness to realize an autonomous mobile robot under an unknown environment.
2001-08-31