Socially Learning Robot

Can a robot learn how to engage people on its own? SoLeRo learns how to capture people’s attention and keep them engaged by trying out many of its facial expression and seeing people’s response. Guess what? SoLeRo learned, that “no one leaves a sad robot”. SoLeRo learned this engagement strategy by itself, just like babies, that making a sad face makes people not want to leave.

SoLeRo used a reinforcement learning algorithm to learn which facial expressions make people engage with the robot. In order to measure engagement, we used a Kinect sensor to detect faces and how close they were to the robot. Engagement was thus defined as how many close faces were looking at the robot. We captured data at the World Science Festival to train SoLeRo during the “Ultimate Science street Fair”.

Publications

  • G. Gordon and C. Breazeal, “Learning to maintain engagement: no leaves a sad DragonBot”, AAAI, Fall Symposium, AI-HRI, 2014.

No one leaves a sad DragonBot!