Abstract
Recent advances in developing communication robots enable us to consider a bystander robot that supports human-human communication. Current paper proposes a framework for a bystander robot to select effective actions in order to enhance the synchrony between humans assuming that the interaction among humans and the robot can be modeled by partially observable Markov decision process. In the proposed framework, the future changes in human behavior according to the robot action are predicted and utilized to select the most effective action by the robot. In this study, we begin with examining whether the predictor of human behavior can be constructed by using the boosting algorithm on the sensory data observed from the experimental communication among two humans and a small humanoid robot. The experimental results of constructing predictors are reported and analyzed in terms of applicability for action selection by the robot to support human communications.
