Abstract

In order to allow humans and robots to work closely together and as a team, we need to equip robots not only with a general understanding of joint action, but also with an understanding of the idiosyncratic differences in the ways humans perform certain tasks. This will allow robots to be better colleagues, by anticipating an individual's actions, and acting accordingly. In this paper, we present a way of encoding a human's course of action as a probabilistic sequence of qualitative states, and show that such a model can be employed to identify individual humans from their respective course of action, even when accomplishing the very same goal state. We conclude from our findings that there are significant variations in the ways humans accomplish the very same task, and that our representation could in future work inform robot (task) planning in collaborative settings.