Abstract
Role-taking between subjective-self awareness (I) and objective-self awareness (Me) is significant to control interaction with others. In objective-self awareness, we estimate what our partners expect ourselves to do. The behavioral response of the partners includes somewhat the desired interactive patterns, which allow us to maintain fluent communication. So in this paper, we propose a new learning framework of interaction patterns. The robot learns the partner's motion patterns during their interaction as itself's desired response patterns. We verified the validity of this mechanism by constructing a virtual world where the robot can physically interact with its partner.
