| Popis: |
When designing controllers for machines that interact with human users, it often becomes necessary to adapt control policies to user preferences, even when these preferences are not aligned with the optimal policy. In this paper, we present a reinforcement learning approach that allows to take into account both a classical control performance and end-user feed- back. We aim to learn policies that adapt automatically to the needs of a set of users, that rely on the devices. These human- friendly schedules accommodate to user-specific requirements, while simultaneously minimizing operational costs. |
