Paper Title
Deep Reinforcement Learning with CNN for Adaptive Traffic Signal Control

Abstract
Various studies have introduced signal control techniques for adaptive traffic to enhance the flow of traffic efficiency. Implementing traffic signals in an adaptive traffic management system proves effective in minimizing congestion. Reinforcement learning, a modern approach, seeks to develop a policy function through a trial-and-error process, optimizing rewards by interacting dynamically with its environment. This study presents a traffic signal control framework for an oversaturated urban network using Deep Q-Network. The learning process is improved by integrating diverse state information, incorporating detailed traffic conditions from both upstream and downstream perspectives. Experiments are usually done using the representation of Urbanmobility, a traffic simulator made for the management of traffic signals on a large scale. Numerous studies have explored control techniques for ATSfor improvingflow of traffic efficiency. An adaptive traffic management system that utilizes traffic signals can effectively reducing the congestion by adjusting the timings of the signalsaccording to real-timeconditions of traffic. One of the advanced approaches in this domain is reinforcement learning, which aims to develop an optimal decision-making policy through continuous interactions with the surroundings using a hit-and-miss process. By refining these interactions, the system learns to maximize rewards and improve traffic control strategies over time.