Personalized Robot Tutors that Learn from Multimodal Data

摘要

As the cost of sensors decreases and ability to model and learn from multi-modal data increases, researchers are exploring how to use the unique qualities of physically embodied robots to help engage students and promote learning. These robots are designed to emulate the emotive, perceptual, and empathic abilities of human teachers, and are capable of replicating some of the benefits of one-on-one tutoring from human teachers. My thesis research focuses on developing methods for robots to analyze and integrate multimodal data including speech, facial expressions, and task performance to build rich models of the user's knowledge and preferences. These student models are then used to provide personalized educational experiences, such as optimal curricular sequencing, or leaning preferences for educational style. In this abstract, we summarize past projects in this area and discuss applications such as learning from affective signals and model transfer across tasks.