Coordinating Fully-Cooperative Agents Using Hierarchical Learning Anticipation