Scribd
Upload
20 views

Improving Traffic Light Systems Using Deep Q-Networks

Uploaded by

Janhavi Patil
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
20 views

Improving Traffic Light Systems Using Deep Q-Networks

Uploaded by

Janhavi Patil
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 2

Title-Improving traffic light systems using Deep Q-networks

Abstract-

As urban traffic complexity increases, managing intersections efficiently becomes challenging. This
work proposes an agent-based traffic light control system, where each intersection has an agent that
dynamically adjusts light phases based on real-time traffic conditions. Using Deep Networks for agent
training, our system significantly reduces waiting times, improving transit times by 44% compared to
traditional fixed-timing methods. New performance metrics are also introduced to evaluate the
effectiveness and reliability of our approach.

Authors name-

Juan Moreno-Malo , Juan-Luis Posadas-Yagüe , Juan Carlos Cano , Carlos T. Calafate, J.


Alberto Conejero , Jose-Luis Poza-Lujan

Year of publication - 15 October 2024

Technology –

 Deep Q-Networks (DQN): A reinforcement learning algorithm used for making decisions in
complex environments with large state spaces.

 SUMO (Simulation of Urban Mobility): A microscopic traffic simulator employed for testing
and validating traffic management strategies.

 Internet of Things (IoT): Provides real-time traffic data and enables dynamic traffic control.

 Deep Learning Models: Used for traffic analysis and prediction, including Yolov3 for object
detection.

 Neural Networks: Applied within the DQN framework to approximate action-value functions.

These technologies are integrated to develop and test a multi-agent traffic light control system.

Algorithms-

Q-learning: For developing the multi-agent traffic control system.

Deep Q-Network (DQN): To optimize traffic light control and improve decision-making.

Findings-

Reduced Waiting Times: The Q-learning traffic control system significantly cut waiting times at
intersections.

Environmental Gains: Lowered CO2 emissions, fuel use, and noise.

Performance Metrics: New indicators were created to measure system effectiveness.

Future Work: Enhance agent coordination, include public transport, adapt to dynamic events, and
test in real urban settings.
Methodology –

Model: Multi-agent traffic light control using Deep Q-Learning (DQN) to minimize vehicle waiting
times at intersections.

Components:

Environment: Traffic lanes and intersections with cyclic light phases.

Agents: Each manages one intersection and adjusts traffic light phases periodically.

Action Management: Agents change phase durations to optimize traffic flow.

Reward Function: Minimizes waiting times at the agent's and neighbouring intersections.

Q-function Approximation: Uses a deep neural network for Q-value estimation, trained with DQN.

Implementation: Python and PyTorch for training, SUMO for simulation with a Manhattan-like
topology.

Limitations-

Limited Scope: The model currently doesn’t account for public transportation or dynamic urban
events.

Decentralized Control: Coordination among agents could be improved.

Real-World Testing: The system needs further validation in diverse and evolving real-world traffic
conditions.

You might also like