Environment Guided Interactive Reinforcement Learning: Learning from Binary Feedback in High-Dimensional Robot Task Environments.