The Institute for Information Systems of FHNW School of Business has various social robots at hand, among them Joey, a robot equipped with a rendered face that can display a wide range of emotions. This work aimed at designing and evaluating three different design sets of facial expressions by collecting data from potential robot interaction partners. The main aim was to investigate which facial features express the intended emotions best and how people perceive these social robot facial expressions. Finding a balance between expressiveness, likeability and suitability was the main challenge.

Through a literature review, theories relevant to the design of facial expressions were identified such as Uncanny Valley (UV), and Facial Action Unit System (FACS). The framework FACS helps to code and encode facial expressions by breaking down the individual movements of features (e.g., eyes, mouth) in a face. Informed by the findings, two facial expression sets were designed. Each design was manipulated to display the three predefined emotions: happiness, sadness, and anger. By conducting a survey (N = 109) and exposing participants to these facial expressions, recognizability, suitability, and likeability were investigated. The two different facial designs were compared to the existing face design of the robot Joey.

The ratings of the survey participants showed the following trends: an overall minimalistic/simple face design was preferred. The recognition rate of the most minimalistic design also suggests that having a clean and simple design helps to better depict the desired emotion. Through content analysis, several emotion-specific features could be indicated as an intensifier for emotions like tears for the emotion of sadness, and dominant and strongly pulled-down eyebrows for anger. These results may be taken into account when redesigning Joey's facial expressions.