Abstract
Joint attention, that is a behavior to attend to the object that the other gazes at, is an important ability not only for human-human communication but also for human-robot communication. Previous works have proposed mechanisms to acquire joint attention based on the casual structure about face pattens and own gaze shift. However, a robot needs to know which pairs of states and actions it should focus on to acquire joint attention. The robot needs to detect the causality among the previous observation, the action, and the current observation. In this paper, we focus on the transfer entropy that indicates the influence of a state and an action on an experience. We calculate transfer entropies in face-to-face interaction and show that a robot can find the pair that includes the causal structure and that the robot can acquire joint attention based on the pair.
