Synthesis of Hierarchical Controllers Based on Deep Reinforcement Learning Policies