Exploring Avatar Facial Fidelity and Emotional Expressions on Observer Perception of the Uncanny Valley

Abstract

Avatars or digital human characters are traditionally associated with digital games, where they enhance story-based interaction and player engagement. However, interest in accurate and cost-effective avatar development has increased in fields such as serious gaming and simulation training. The interest from these fields stems from training needs where avatars, used create human emotional experiences, could aid in appropriate education. In this context, the area of avatar emotional expression is not well understood. Applications often result in avatars that produce feelings of uncanniness in end-users, which may have a negative impact on training outcomes. This paper aims to firstly explore how avatar fidelity or realism could influence the emotional experience of interactions between end users and virtual humans. Secondly, we examine how avatar facial features displaying emotional expressions affect participants perceived valence of the avatar. These affects were assessed through a combination of experimental and survey methodologies. An existing ‘Godspeed’ survey was used to measure an end-users’ perception of an avatars ‘humanness’, ‘eeriness’, and ‘attractiveness,’ together with a three-part experiment that measured participants’ startle reflex responses to avatars with differing fidelity and emotional expressions. The results indicated that participant gender played a role in their perception of avatars. Also, the gender of the avatar appeared to have a significant impact on participant responses. Avatars displaying sad expressions emerged as less unpleasant and possibly less uncanny than smiling expressions. This research represents an entry point in a broad, cross-disciplinary area of research. While there are important findings and a significant amount of data generated, these elements pose questions for future work in this area.