Composition and Evaluation of the Humanlike Motions of an Android

Both appearance and motion are important considerations in designing robots for interaction with humans. In communication, humans sense intelligence in their conversation partner. We aim to develop an intelligent robot that communicates in the same intelligent and human-like way in daily life. We use the total Turing test to evaluate the human-likeness of the android. In this test, subjects are shown either the android or another human for a short time. Subjects are then asked whether it was a human being or not. The android used in this study is called ReplieeQ2, which has an appearance and motions that closely resemble those of a human being. We hypothesize that an android that is given more human-like motions will be identified as a human in the total Turing test more often. To create human-like motion we observed humans behavior when sitting. We focused on eye, eyelid, neck and chest motions and analyzed them in detail. The android motions were made from this analysis and used in the Total Turing Test. We prepared three conditions: a static android (SA), an android given natural motions (MA), and a human sitting naturally (SH). We expose a subject to one of these conditions for either one or two seconds. We then examine how often they correctly identify the condition as either human or robot. SH was identified as human most often, followed by MA and then SA, for both exposure times. MA is judged to be significantly more humanlike than SA, within the two-second experiment time. This suggests that natural motions contribute to the human-likeness of the android. We also investigate the contributions of individual natural motions to human-likeness and likeability by a pair comparison method. The results show that eyelid motion is the most effective factor for expressing human-likeness and likeability. By enhancing the android´s expressive power, this work contributes to the development of humanoids that communicate naturally with human beings.