2001 年 121 巻 4 号 p. 762-768
In this paper, we propose a behavior learning method of a mobile robot using visual information in real environment. View information, which is a characteristic of image, is utilized for input value in proposed method. The input value is a sort compressed data that original image from camera is projected. Proposed system consists of two networks for state expression and behavior generation. State expression network, which is based on self-organizing algorithm, categorizes the robot's states from view information. Therefore, state expression network acquires how to express the state of system in real environment. Also, behavior generation network autonomously acquires the robot's behaviors using output value of state expression net-work. For training behavior generation network, we utilize Actor-Critic method as a sort of unsupervised learning scheme. We report some experimental running results using a mobile robot in real environment. As the result, the mobile robot generates adaptive behaviors utilizing visual information, autonomously.
J-STAGEがリニューアルされました! https://www.jstage.jst.go.jp/browse/-char/ja/