Doubly Mild Generalization for Offline Reinforcement Learning Yixiu Mao 1, Qi Wang 1, Y un Qu