Representation Learning For Perception And Control