Exploring Perception and Preference in Public Human-Agent Interaction: Virtual Human Vs. Social Robot

This paper delves into a comparison between virtual and physical agent embodiments with the aim to gain a better understanding of how these are perceived and interacted with in a public setting. For our experiment we developed two agent embodiments: a virtual human and a mechanical looking social robot, that encourage passerby in public space to exercise squats through speech and non-verbal cues. We analyzed user behavior during the interaction with one of the distinct systems that differ in representation but share the same purpose and intent. We recorded 450 encounters in which a passerby listened fully to the agent's instructions and used body tracking to analyze their exercise engagement. At least one squat was performed in each of 145 encounters, which generally indicates fairly high system acceptance. Additional feedback came from 61 individuals (aged 13 to 74, 41 males, 20 females) through a questionnaire on perception of competence, autonomy, trust, and rapport. There was no significant difference found between the virtual human and the social robot concerning these factors. However, responses to single questions indicate that interactions with the social robot were perceived as significantly more responsive, and gender differences in perceived interaction pressure emerged, with women reporting significantly higher values compared to men. Despite public space challenges, the agent systems prove reliable. Complexity-reducing technical and methodological simplifications and possible sampling biases must be taken into account. This work provides a glimpse into public interactions with virtually and physically embodied agents, and discusses opportunities and limitations for future development of such systems.