Toward Information Theoretic Active Inverse Reinforcement Learning