To develop truly autonomous and intelligent mobile robots, the developmental approach is indispensable. Considering the development of human beings, internal rewards which are related to the instinctive desire, motivation, curiosity, novelty, etc., are important factors contributing to the development. For the robots, there are several studies in which curiosity and novelty are used to promote the progress of learning. In this paper we propose an internal reward called "appetitive reword" and a function to control it, for the robots which have to acquire an action policy for a mission to "survive and perform a instructed task effectively". The "appetitive reword" is inspired by "appetite" of human beings, which is known to be well regulated by a mechanism between the digestive tract and the brain. By using an example of a security guard robot, we show the effectiveness of the proposed reward.
抄録全体を表示