Efficient Deep Reinforcement Learning Via Planning, Generalization, And Improved Exploration