Using Rotation, Translation, and Cropping to Boost Generalization in Deep Reinforcement Learning…