Hierarchical Reinforcement Learning and Value Optimization for Challenging Quadruped Locomotion