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