Traffic Management System Using Reinforcement Learning

An AI-based traffic signal control system that uses Reinforcement Learning to reduce congestion at a 4-way intersection, built with SUMO (Simulation of Urban Mobility) and trained using Proximal Policy Optimization (PPO).

Overview

Traditional traffic signals run on fixed timers, regardless of actual traffic conditions. This project replaces that with an RL agent that observes real-time traffic at an intersection and dynamically decides which signal phase to activate, aiming to minimize vehicle wait times and queue lengths.

A custom Gymnasium environment wraps a SUMO traffic simulation, allowing the agent to be trained and evaluated using standard RL tooling.

Features

• Custom Gymnasium environment built on top of SUMO
• Multi-lane traffic handling via edge-based aggregation
• State representation using a 2x2 spatial grid, suitable for CNN-based policies
• Reward function based on minimizing queue length and vehicle stops
• Agent trained using Proximal Policy Optimization (PPO)
• Performance comparison against traditional fixed-time traffic signals
• Streamlit interface for visualization/interaction

Tech Stack

• Language: Python
• Simulation: SUMO (Simulation of Urban Mobility)
• RL Environment: Gymnasium
• RL Algorithm/Training: Stable-Baselines3 (PPO)
• Deep Learning: PyTorch
• UI/Visualization: Streamlit

Results

The trained RL agent showed significant improvements over fixed-time signal control, particularly in reducing average vehicle waiting time and increasing intersection throughput under medium and high traffic density conditions.

Getting Started

Installation

git clone https://github.com/Rachi-collab/Traffic-Management-System-Using-Reinforcement-Learning.git
cd Traffic-Management-System-Using-Reinforcement-Learning
pip install -r requirements.txt

Usage

# Train the agent
python train.py

# Evaluate / run the trained agent
python evaluate.py

# Launch the Streamlit dashboard
streamlit run app.py

Note: Adjust the commands above to match your actual script names if they differ.

Project Structure

Traffic-Management-System-Using-Reinforcement-Learning/
├── env/               # Custom Gymnasium + SUMO environment
├── models/            # Saved/trained PPO models
├── sumo_config/        # SUMO network and simulation config files
├── train.py             # Training script
├── evaluate.py           # Evaluation script
├── app.py                # Streamlit app
└── requirements.txt

Future Improvements

• Extend to multi-intersection / network-wide traffic control
• Incorporate real-world traffic data
• Experiment with alternative RL algorithms (e.g., DQN, A2C) for comparison

Acknowledgments

Built using SUMO, Gymnasium, and Stable-Baselines3.