Compositional Reinforcement Learning from Logical Specifications