Abstract: Despite the significant achievements of deep reinforcement learning in a wide range of domains, there is still the generalization problem. Traditional transfer methods cannot adapt quickly ...