Applying Hierarchal Clusters on Deep Reinforcement Learning Controlled Traffic Network

Traffic congestions is a crucial problem affectingcities around the globe and they are only getting worse as thenumber of vehicles tends to increase significantly. Traffic signalcontrollers are considered as the most important mechanism tocontrol traffic, specifically at intersections, the field of MachineLearning introduces advanced techniques which can be appliedto provide more flexibility and adaptiveness to traffic controltechniques. Efficient traffic controllers can be designed using areinforcement learning (RL) approach but major problems offollowing RL approach are, exponential growth in the state andaction spaces and the need for coordination. We use real trafficdata of 65 intersection of the city of Ottawa to build oursimulations and show that, clustering the network usinghierarchal techniques has a great potential in reducing the stateactionpair significantly and enhance overall trafficperformance.