Maximum Entropy Multi-Task Inverse Reinforcement Learning