AUTONOMOUS BEHAVIOR GENERATOR FOR A COMPANION ROBOT: SELF

Abstract

We propose a robot behavior generation scheme that generates behaviors based on senses of “curiosity” and “boredom,” to create a companion robot named SELF with which humans do not get bored. The scheme was developed using a simple but unique simulation environment. A red ball and a blue ball were displayed on a computer; the human subject moved the red ball. First, the impressions of various motions and actions of the robot were examined. The actions at moderate speed create agreeable impressions. Based on natural interpersonal interaction patterns, the actions that the robot is expected to generate are classified into four types: “following action,” “fleeing action,” “no interferential action,” and “no action.” We analyzed humans’ impressions of the behaviors of the robot generated by switching the four basic actions at moderate speed. Three types of information transfer efficiency characterized these behaviors. We proposed the behavior generation scheme for SELF. The autonomous learning capability of SELF was realized by updating this conditional probability. Using the developed scheme, SELF can display curiosity and boredom resembling that seen in animals, including humans. Lastly, impressions of the robot behaviors were examined: two were characterized by information transfer efficiency; one was controlled by the proposed scheme.