Accepted by AROB 2021. For letting agents in traffic simulation behave more like humans, we propose a unified mechanism for agents learn to decide various accelerations on deep reinforcement learning and generate a traffic flow behaving variously to simulate the real traffic flow.
