Evaluation of Android Using Unconscious Recognition

In human-robot interaction, both appearance and motion are essential aspects of the robot. We study human-robot interaction using an android that has human-like appearance. Mori (1970) hypothesized the 'uncanny valley' which describes a relationship between appearance of a robot and the feeling it produces in humans. To design robots that interact successfully with humans, we must know the structure of the uncanny valley. Humans show unconscious behaviors when interacting with another human. We expect the same behaviors when interacting with a very human-like robot. Then we can change appearance and motion to study how the unconscious behavior of the human changes. In this way, we explore the uncanny valley. The unconscious behavior we consider in this study is gaze behavior. It has been found (McCarthy et al., 2001 and 2003) that eye movements are used to send social signals during conversation. In particular, when thinking about the answer to a question, humans tend to look away from the questioner. We hypothesize that if the human-likeness of the questioner changes, this gaze behavior also changes. We compare three different questioners: human, android and robot with a 'mechanical' appearance. We found that the subject changes gaze to the left of the face a longer time in case of a human or android questioner. The subject changes gaze to look down from the face in the case of a mechanical robot questioner. There is a significant difference between these two behaviors. We conclude that the android questioner is unconsciously treated in the same way as the human questioner. The mechanical robot is treated differently from the human questioner. These results will become clues to the uncanny valley and contribute to the progress of human-robot communication