Human-Likeness in Android Facial Expressions: Effects of Embodiment and Dynamics on Behavioral and EEG Responses

Abstract

Android robots increasingly rely on facial expressions to communicate internal states during human-robot interaction. However, it remains unclear whether human observers respond primarily to the appearance of the agent (human versus android) or to the dynamics of the facial expression itself. We present an experimental paradigm that enables the assessment of human-likeness of android facial expressions while studying these two factors. Participants observe 6 -second video clips depicting three expressive states (happy, sleepy, and surprised) under three conditions: (1) an android displaying the expression, (2) a human producing the expression naturally, and (3) a human imitating the android's expression style. Following each clip, participants categorize the perceived emotion and rate its intensity. Stimuli are repeated ten times and presented in counterbalanced and randomized order while electroencephalography (EEG) is recorded. The proposed design probes agent embodiment and expression dynamics, which are typically confounded in studies of robot facial-expression perception, and provides a tool to assess the human-likeness of android faces at behavioral and electrophysiological levels. The paradigm offers a principled framework for evaluating alternative android facial-expression generation mechanisms, including manually designed and learned controllers. Overall, this work contributes towards understanding human perception of expressive states in androids and provides a methodological tool for improving expressive robot design.