reinforcement learning agents have been mostly developed and evaluated under
the assumption that they will operate in a fully autonomous manner -- they will
take all actions. In this work, our goal is to develop algorithms that, by
learning to switch control between machine and human a