Promoting Coordination through Policy Regularization in Multi-Agent Deep Reinforcement Learning