In this paper the authors distill a Deep Reinforcement Learning such as DeepQN into a decision tree policy which can then be automatically checked for correctness, robustness, and stability. The ...