Implementing traffic signal optimal control by multiagent reinforcement learning

In urban traffic environment, the traffic flow is more difficult to be predicted properly because the interaction and intertwinement among multiple crossroads, which makes preset traffic control model can not keep always high performance in all traffic situations. Considering the capability of autonomous learning being inherent in reinforcement learning, we propose a multiagent reinforcement learning based traffic signal control method. Without preset control model, multiple collaborative agents can learn the optimal control policy corresponding real traffic situation. The experiment results demonstrate the applicability and effectiveness of our approach.