System identification approach to robot learning: application to behavior learning of mobile robot based on visual information

摘要

A word is enough to the wise. We are pursuing research issues focused on the realization of such learning capability. This paper proposes a robot learning method based on system identification approach. Our method utilizes GMDH algorithm which is a kind of system identification method and expands it so that multiinput-multioutput type system can be applied to. As an example task of our robot, this paper deals with behavior learning problem based on visual information. Suppose that a set of visual information is input data and continuous value of motor command to a robot is output data, identifying a mapping function between input and output data is equivalent to resolving behavior learning problem.