Designing and Validating Expressive Cozmo Behaviors for Accurately Conveying Emotions

Abstract

Robots have unique abilities to influence people, but when deploying robotic systems in assistive applications, roboticists must understand how users perceive these systems’ behaviors. As part of an ongoing project to use robots as motivational break-taking aids, we present Cozmo behaviors that could function as the action space of a future robot learning strategy. Before deploying these behaviors in the wild, we evaluated them using an online video-based study with N = 113 participants. Results show that participant perceptions of Cozmo behaviors tend to match the intended valence and energy level. Furthermore, behavior valence in particular has a strong bearing on other perceived characteristics such as interaction appeal, trustworthiness, and safety. Facial expression and loudness acted as important covariates, which may help generalize these results to other behaviors and robots. The products of this work can benefit those who are interested in robot emotional expression and assistive robot applications.