Analyzing human gaze adaptive to robot gaze in real-time human robot interaction

Eye gaze is very important to adaptive behaviors in human-human communication because it can both indicate human's visual attention and perceive the attentions of others. Human-robot interaction experiments can be used to study real-time human adaptive behaviors, the timing of gazing and exact timing of multimodal interactions between communication partners. One advantage of doing so is that the robot's behaviors can be systematically manipulated by different strategies so that the social interaction between two social agents can be decoupled and easier to analyze. For this purpose, a human-robot interaction system was built. The evaluation of this system shows that the robot can respond to the human's gazing behavior in the same time scale of human's responsive behavior. Therefore it can serve as a research platform for studying human-robot interaction. This thesis also presents an experiment designed on this research platform. The human subjects interacted with the robot controlled by four different interaction strategies: no looking back, looking back, extended looking back and more looking back. The main results are as following: (1)human subjects were sensitive to robot's different gazing behaviors and adapted to the changes; (2)when the robot gazed more on human's face, more mutual gazes happened; (3)when the robot gazed more on human's face, the human subjects were more likely to look at robot's face right before they speak of the name of the objects; (4)the human's gaze duration on the objects was much shorter when the robot did not follow the human's gaze; (5)compared with "more face looking" condition, "extended face looking" condition had more transfer entropy from robot's gaze to human's gaze, which indicates better communication.