Attentive person selection for human-robot interaction

We present a method that enables the robot to select the most attentive person into communication from multiple persons, and gives its attention to the selected person. Our approach is a common components-based HMM where all HMM states share same components. Common components are probabilistic density functions of interaction distance and people's head direction toward the robot. In order to cope with the fact that the number of people in the robot's field of view is changeable, the number of states with common components can increase and decrease in our proposed model. In the experiments we used a humanoid robot with a binocular stereo camera. The robot considers people in its field of view at a given time and automatically shifts its attention to the person with highest probability. We confirmed that the proposed system works well in the selection of the attentive person to communicate with the robot.