Active robot learning with human tutelage

Abstract

In this paper we describe how a robot may benefit from active learning in a human-robot tutelage setting. Rather than passively absorbing conceptual knowledge, the robot learner actively tries to influence the human teacher in order to improve its learning experience. We compare the performance of agents that employ this strategy in simulation to a robot that interacts with a human teacher. It was found that people respond to the robot's social cues and that this can improve learning, albeit less expressed than in simulation. Moreover, we found gender effects indicating that robot learners might benefit from even more specific tailoring towards their human tutors.