Route Traffic Guidance Using Deep Reinforcement Learning

The developments of taller buildings have increased in the blgger cities.This has an effect on the traffic in the urban areas.Because of that,the number of people moving thought one street has increased,and it also increase the number of vehicles.Nowadays the space for moving between one building and another is narrower than before,and now the density of people in the buildings has increased in relationship with the area of the building,due to the build of higher living and working building.As the city changes,the traffic changes as well,and therefore,people have to take more time to move from one place to another and interconnecting with other paths.These complex interconnected paths are one of the reasons why we have many traffic jams nowadays.As we can see the selection of the route is very important to avoid the traffic jams and congestions that end up in loosing time for everyone.The selection of the route will not only to be the shortest path but to use the best ways to avoid the concurrent traffic jams.The estimation of the route according to the traffic state can give us a faster way to avoid traffic jams and arrive at the destination in the shortest time,saving time for us and the other cars as well,therefore releasing the stress of the traffic jams in the roads.The Modeling of the traffic has been characterized many times,but due to many factors the resulting of the modeling is always complex or somehow not entirely complete.So,in this case we are forced to neglect some factors in order to make the model usable to our purposes.The route traffic guidance is now days selected mostly based on the shortest path from one point to another,the calculation of the shortest path is easily calculated with the Dijkstra algorithm,and therefore we assume that we can get the shortest between 2 points without the problem of calculating for each agent.Then in this paper we make a model of the route traffic network to control the traffic with a policy for choosing in the agents going from one starting point in the map to another point called the destination.The deep reinforced learning model is used to select the best route according to the density of the traffic caused by other vehicles in the route network.We implement a deep neural network that combined with the reinforcement learning using the q values.It can learn how to avoid the traffic jams caused by the vehicles,frequently in some parts of the map.The deep neural network that we apply is a Deep Q Network.It uses the reinforcement learning and a deep neural network.The Deep Q Network determines the agent next action according to the Q values.The Deep Q network will be trained for the defined state of the density of the road network.Therefore,we evaluate the performance of the deep Q Network in different conditions.Firstly,in a simple and concrete example where we can easily evaluate the performance of the Deep Q Network.In this case we evaluate it with only 2 vehicles and 2 destinations fixed.Secondly,we add more vehicles to explore the ability of the Deep Q Network to learn for a more complex situation.Next,we proceed to train it with random positions of start and destinations,to prove that the Deep Q Network can adapt to different situations.And finally,we train it for a more complex situation with 20 vehicles where they can select random starting points and random destinations in the road map.