| Abstract |
A Smart city as a complex system includes many different subsystems interacting together. Smart transportation is one of the subsystems in a Smart city which consist of roads, vehicles, intersections, roadside units, traffic control systems, and so on. Many of these components needs to be connected to provide a smooth safe traffic in a smart city. The performance of these components largely depends on the computational latency of algorithms running on local or central processors. Hence, providing an optimized solution to minimize this delay is much needed. In this work, we propose a way to properly manage the flow of self-driving vehicle traffic at road junctions, taking into account pedestrian traffic. Self-driving vehicles are able to communicate with each other and smart devices along the road. However, surveillance cameras are needed to observe pedestrians' traffic at the intersection. Therefore, we use cameras, smart sensors, processors, and communication equipment embedded in self-driving vehicles and roadside smart devices to collect data, process it, and generate proper instructions to manage self-driving vehicles traffic flow at intersections. In this research we have used Simulation of Urban Mobility software to simulate traffic behaviors resulting from the use of the proposed solution. Although the simulation shows a smooth flow of traffic in a simple junction, a deep reinforcement learning approach is then proposed to manage traffic flow in multiple lane junctions. To decrease the risk of collision in intersections, a deep reinforcement approach that utilizes safety parameters for its reward function developed. The results illustrate noticeable decrease in collision at intersections. |