Before robots can be truly useful in homes, schools, and hospitals, they must become capable of learning new skills. Andrea Thomaz, an assistant professor of interactive computing, wants them to learn from their users, so that experts don’t have to program every task. She aims to make robots that not only understand a human teacher’s verbal instructions and social signals but give social feedback of their own, using gestures, expressions, and other cues to let the person know whether they have correctly understood the directions.
Thomaz has designed machine learning algorithms based on human learning mechanisms and built them into her robots Junior and Simon, which have faces that make basic expressions and hands that can grasp simple objects. In experiments with people untrained in formal teaching, Junior has quickly learned enough about things in its environment to catch on to tasks such as opening and closing a box. –Kristina Grifantini