Robot Learning through Collaborative Dialog

Contents

In this paper we present our work in the domain of robot learning. Machine learning is a shared activity between the human teacher and the machine learner. Both partners have responsibility in maintaining mutual beliefs about the learner’s progress in the task. We argue that machine learning should more closely resemble human-style tutelage, and that this will benefit both the human and the machine in the partnership. We demonstrate this with a learning mechanism that shows a robot is able to rapidly learn a task from a human, it generalizes this ability to a new task configuration, and shows commitment to the overall goal of the learned task.