Future Traffic Management System Empowered by Reinforcement Learning and V2X technologies

Traffic congestion in major cities around the globe is a daunting problem that is getting worse as the speed of urbanization keeps increasing. The role of traffic lights at intersections in regulating flows cannot be underestimated. Though there are a variety of actuated traffic signals based on cameras or loop detectors, the cost of these technologies is prohibitively large. This is why the vast majority of existing traffic lights in the world employ a timerbased decision logic which is clearly not very effective.
Dedicated Short-range Communication (DSRC) is a relatively new short to medium range wireless communication technology used for inter-vehicular communication. With this communications technology or other existing technologies such as cellular, Wi, Bluetooth
or UWB, an intelligent traffic signal control (ITSC) can be much more cost-effective since the majority of intersections can be equipped with actuated traffic signals with low cost.
On the other hand, wireless Vehicle-to-Vehicle (V2V) or Vehicle to Infrastructure (V2I) communications based ITSC is considerably different from traditional ITS; hence, adapting
current ITS to V2X based systems is pretty difficult. Traffic control algorithms will also need to be re-designed to implement the unique properties of wireless communication.
In this thesis, several new forms of possible future infrastructure-based and infrastructure free ITSC systems are researched, in terms of simulation results, prototype system design, algorithm studies. The future traffic systems introduced are Virtual Traffic Lights (VTL), DSRC-Actuated traffic signal control (DSRC-ATSC), and a more advanced version of DSRCATSC: Partially Detected Intelligent Traffic Signal Control (PD-ITSC) System. We explore the capacity of state-of-art Reinforcement Learning algorithm on the PD-ITSC systems.