Bidirectional Task-Motion Planning Based on Hierarchical Reinforcement Learning for Strategic Confrontation