A New Approach to Reinforcement Learning for Sequential Robotic Tasks using a Chained Options Model and Subtask-Focused Rewards