Although deep reinforcement learning (DRL) methods are promising for making behavioral decisions in autonomous vehicles (AVs), their low training efficiency and difficulty to adapt to untrained cases ...