Using Human Demonstrations to Improve Reinforcement Learning