Capturing Student-Robot Interactions for a Data-Driven Educational Dialogue RL Environment

Intelligent social agents provide means to improve student learning and motivation in an educational setting, and support students in a personalized manner. Current educational theory suggests that learning in an interactive setting is best when students are participating equally. We aim to use reinforcement learning (RL) to teach an intelligent social agent to use gaze, gesture, and dialogue to maintain and improve students’ participation. Performing reinforcement learning in the real world is often intractable, generally requiring some pre-training in an artificial domain. We examine a data-driven approach using previously collected data to simulate interactions in an educational group setting.