An open-ended approach to evaluating Android faces

Expectation and intention understanding through nonverbal behavior is a key topic of interest in socially embedded robots. This study presents the results of an open-ended evaluation method pertaining to the interpretation of Android facial expressions by adult subjects through an online survey with video stimuli. An open-ended question yields more spontaneous answers regarding the situation that can be associated with the synthetic emotional displays of an Android face. The robot used was the Geminoid-DK, while communicating the six basic emotions. The filtered results revealed situations highly relevant to the portrayed facial expressions for the emotions of Surprise, Fear, Anger, and Happiness, and less relevant for the emotions of Disgust, and Sadness. Statistical analysis indicated the existence of a moderate degree of correlation between the emotions of Fear-Surprise, and a high degree of correlation between the pair Disgust-Sadness. With a set of validated facial expressions prior to nonverbal emotional communication, androids and other humanoids can convey more accurate messages to their interaction partners, and overcome the limitations of their current limited affective interface.