Abstraction Selection in Model-based Reinforcement Learning

State abstractions are often used to reduce the complexity of model-based reinforcement learning when only limited quantities of data are available. However, choosing the appropriate level of abstract