Smart Traffic Light Controller using Deep Reinforcement Learning

Transportation has become a major priority for people nowadays. While technological advancements have made transportation easier, monitoring and controlling traffic as well as simulating it has become a significant challenge despite the significant research that has been carried out to tackle this problem. An emerging trend to solve this problem involves using deep reinforcement learning techniques which has shown significant progress and promising results in recent studies. There is an increasing demand for developing an intelligent traffic controlling agent which can dynamically adjust to real time traffic rather than just operating by handcraft rules such as timers. We employ modern deep reinforcement learning methods such as Q-learning and deep neural networks with an agent-based traffic simulator SUMO ("Simulation of Urban MObility") which provides a synthetic yet realistic environment for simulating real-world like traffic and explore the outcomes of the actions that were performed on this environment. Additionally we also aim to analyze the performance of the agent by comparing it to traditional traffic signals.