Collaborative Filtering for Learning User Preferences for Robotic Tasks

Service robots are envisioned to have an increasing influence on our lives and to support us on a daily basis. From a truly effective and personalized robot, we expect the ability to learn our preferences concerning the requested tasks. Such preferences, however, are often user-dependent so that predefined strategies only match a subset of all users. In this work, we address the problem of tailoring the robot’s behavior to the preferences of its user. We present a novel solution to the problem of encoding multiple preferences for individual tasks that leverages the collaborative filtering framework. A key aspect of our method is that it does not require each user to specify preferences for all tasks. From a small number of known preferences, our approach is able to infer the user’s taste for other tasks. We present quantitative results based on crowdsourced data from thousands of users. Our results suggest the validity of our approach and demonstrate that we are able to predict user preferences with respect to two service robot scenarios.