Learning of Generalizable and Interpretable Knowledge in Grid-Based Reinforcement Learning Environments