Termination Approximation: Continuous State Decomposition for Hierarchical Reinforcement Learning