Batteries, camera, action!
Learning a semantic control space for expressive robot cinematography