Robot Deep Reinforcement Learning: Tensor State-Action Spaces and Auxiliary Task Learning with Multiple State Representations - Robotics Institute Carnegie Mellon University