As a human being, you already know that facial expressions greatly influence everyday interactions between people. You might scan someone’s face to determine if they’re lying, lose a hand of poker after momentarily grinning at your cards, or wonder what is displeasing someone who appears to be unhappy. It should come as no surprise, then, that social roboticists have found robot faces and facial expressions to have a strong effect on human-robot interaction.
While beginning our social robotics research at the University of Pennsylvania GRASP Lab, we knew that our choice of robot face would be important. Despite the social impact of robot faces, though, no other researchers had developed and evaluated the kind of faces we sought for robots with LCD screen heads. Rethink’s proprietary set of faces for the Baxter collaborative robot was an interesting proposition for nonverbal status communication, but not quite emotive enough for a social robot. Conversely, some past academic research on LCD screen-faced robots involved hyper-realistic humanoid faces so humanoid that they might raise users’ expectations beyond the capabilities of modern robots!
In our work, we decided to design a new set of faces for Baxter that was humanoid enough to make the robot seem like a playful social agent but not hyper-humanoid enough to make it seem uncanny.
We created faces of many colors and expressions, accompanied by information on how each one is perceived by human onlookers. Our initial set contained 49 Baxter faces, including seven colors (red, orange, yellow, green, blue, purple, and gray) and seven expressions (afraid, angry, disgusted, happy, neutral, sad, and surprised). Online study participants (N = 568) drawn equally from two countries (US and India) rated photographs of a Baxter displaying these faces. Their evaluations of how pleasant and energetic the robot was, alongside their reported feelings of safety and satisfaction while looking at the robot, could help elucidate the emotions conveyed by the robot faces, while also providing useful robot affect information to other researchers.
Overall, the face color and expression, and the rater’s country of origin all significantly affected how pleasant and energetic our Baxter robot was perceived, as well as the onlooker’s feelings of safety and satisfaction. Facial expression had the largest effect size. As commonly seen in nature, red seemed to be a special color for influencing people’s ratings, appearing almost universally less pleasant, more energetic, less safe, and less pleasing. For example, our results indicate that a happy red robot face would generally make raters feel less safe than a happy face of any other color, but more safe than an angry robot face of any color.
While designing our Baxter faces, we found that other researchers were interested in this face database. In response to this interest, we released the Baxter faces, source files, and photographs from the study in the public GitHub repository. Individuals who are interested in more details on our face design and study results can also consult our paper from the 2016 International Conference on Social Robotics here. We hope that Baxter and Sawyer users will find these resources useful in their research efforts.
Read more about guest blogger/roboticist Naomi Fitter and her human interests in Women Who Reign. And see what other researchers have been up to with their Baxter robots in the Rethink Robotics video gallery.