A psychological framework to objectively evaluate socially capable robots for interactive tutoring systems

We introduce a novel evaluation methodology to establish psychometrically validated measures to objectively evaluate socially capable robots. Our methodology involves first creating a digital computer generated face model designed to replicate the facial expression space of the robot with the highest accuracy, and second, using this model to render test sequences, which are in turn analyzed with independent facial metrics software. We compared three different facial modelings techniques to approximate our robot's face and achieved 98.86% accuracy in replicating the facial appearance as measured by facial metric software. This digital face model can now be used to create arbitrary expressions for interaction purposes and for a more detailed analysis of low-amplitude micro-expressions, which are critical for proper social communication with children in a virtual tutoring context. We argue that our methodology is a first step towards objectively assessing the believability of commercially available robots.