Interpretable, Robust, and Verifiable Reinforcement Learning