EVALUATING THE EFFICIENCY OF DEEP REINFORCEMENT LEARNING IN URBAN TRAFFIC SIGNAL TIMING OPTIMIZATION

Read full text Summarise with VIRAT
Authors: Patil Ravindra Nimba
Cited by: 0
References: 6

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Because of the expanding population, the increasing number of people who own vehicles, and the constrained development of infrastructure, metropolitan regions that are undergoing rapid development are experiencing a growing obstacle in the form of congested traffic. Inefficiencies, longer travel times, and environmental issues are frequent consequences of the inability of traditional traffic signal systems to adequately manage real-time variability in traffic flow. These systems, which depend on fixed-time scheduling or reactive sensors, are often unable to keep up with the demands of changing traffic conditions. In the light of this discussion, Deep Reinforcement Learning (DRL) has arisen as a potential artificial intelligence technology that has the ability to dynamically optimize traffic signal timings through ongoing interaction with the environment. This paper reviews existing secondary literature to evaluate the efficiency of DRL models—such as Deep Q-Networks (DQN), Advantage Actor-Critic (A3C), and Proximal Policy Optimization (PPO)—in improving key traffic metrics like vehicle throughput, queue length, and average travel time. The scientific publications and research that are accessible to the public that were published over the course of the last five years are analyzed by using a qualitative narrative review approach. The analysis is constructed around three main themes: the results of traffic efficiency, the architectural design of DRL models (which includes the state space, reward functions, and learning processes), and the practical challenges that are encountered during realworld deployment. According to the results of the research, DRL has a far higher level of success than typical systems in simulations. It does this by offering adaptive signal regulation, which lowers congestion. On the other hand, real-world implementation is faced with difficulties such as constraints on infrastructure, unreliable data quality, a lack of coordination amongst junctions, and expensive retraining expenses

Fields of Study & Subjects

Traffic control and management Traffic Prediction and Management Techniques Transportation Planning and Optimization

Publication details

Journal / source
International Journal of Engineering Research and Science & Technology
Volume
22
Issue
1(S)
Pages
142–152
Publication date
21 March 2026
Language
EN