I think one of the things about reinforcement learning is that it tends to require exploration. So using it in the context of physical systems is somewhat hard.