Text this: Qualitative Spatial Abstraction in Reinforcement Learning