“The Q-learning approach was not well-suited for the robotic locomotion tasks pursued by John Schulman during his PhD.”