I. Introduction
Reinforcement learning's super-human performance on Atari games [1] and Go [2] has led to a new wave of interest in this method. In robotics, reinforcement learning (RL) has made headlines by learning intricate and impressive motor skills such as juggling [3], ball-in-cup [4], in-hand manipulation [5], pick-and-place [6], or locomotion over highly variable terrain [7].