Agile Legged Robots through Reinforcement Learning and Optimal Control