Introducing the ABOT Database
My colleagues at Brown and I recently completed the first iteration of a large (and largely fun!) project we are calling the Anthropomorphic RoBOT (ABOT) Database. Motivated by a wealth of studies about the anthropomorphic appearance of robots, the creation of ABOT was an attempt to conduct a large, systematic survey of the tremendous variety of anthropomorphic (human-like) robots and provide a useful tool for studying the human-like appearance of robots. I enjoy creating things. But more than that, creating this tool was important to me because decisions concerning a robot's degree of anthropomorphism have largely been guided by intuition, and as a result, generalizing across research studies has been difficult. What one study would consider "humanoid" could be vastly different than another study, for instance. Finally, what constitutes the human-like appearance of robots in the literature and in practice is vague and underspecified. So, we set out to address some of these problems and create something cool for others to use. It was a massive effort, but I think worthwhile in the end.
And you can read the abstract of a recently submitted paper about ABOT below:
The psychological effects of robot appearance on human perceivers is a centerpiece in human-robot interaction (HRI) research. Anthropomorphic robots, or robots with human appearance features such as eyes, hands, or faces, have drawn considerable attention over recent years. To date, decisions on what determines different degrees of anthropomorphism has been driven entirely by designers' and researchers' intuitions, because a systematic understanding of the range, variety, and constituent features of anthropomorphic robots is lacking. To fill this gap, we introduce the ABOT (Anthropomorphic roBOT) Database---a collection of extant real-world anthropomorphic robots that includes standardized images of 200 robots with one or more human-like appearance features. Using this database, we were able to decompose the human-like appearance of robots on 19 constituent features and uncover four distinct dimensions (bundles of features that characterize robot bodies and manipulators, the surface look of robots, faces, and mechanical motion) that make up laypeople's (N=1140) overall perceptions of human-likeness. We then identified the individual features that were most predictive of perceptions of a robot's physical human-likeness (N=100). These results can offer guidance on building a recipe for determining if a particular robot will be considered physically human-like in its appearance. As a result, we offer the ABOT Database as a powerful methodological tool for research on the human-like appearance of robots.
Phillips, E., Zhao, X., Ullman, D. & Malle, B.F. (Under review). What is human-like?: Decomposing robot human-like appearance using the Anthropomorphic RoBOT (ABOT) Database. Submitted to the Proceedings of the ACM/IEEE International Conference on Human-Robot Interaction 'HRI.
Photo of PR2 Robot by Beth Phillips