Online Learning of Human Navigational Intentions.

We present a novel approach for online learning of human intentions in the context of navigation and show its advantage in human tracking. The proposed approach assumes humans to be motivated to navigate with a set of imaginary social forces and continuously learns the preferences of each human to follow these forces. We conduct experiments both in simulation and real-world environments to demonstrate the feasibility of the approach and the benefit of employing it to track humans. The results show the correlation between the learned intentions and the actions taken by a human subject in controlled environments in the context of human-robot interaction.