Muhilank/traffic

Meta PyTorch OpenEnv Hackathon - Round 1 Submission

Project: Autonomous Traffic Control Environment

This repository contains a Mini-RL environment built using gymnasium to simulate a 4-way traffic intersection with Emergency Vehicle Prioritization.

This submission is developed for the Round 1 of the Meta PyTorch OpenEnv Hackathon x Scaler School of Technology.

Project Structure

• env.py: The custom Gymnasium environment implementation (AutonomousTrafficControlEnv).
• train.py: A script to train a baseline agent using stable-baselines3 (PPO).
• grader.py: Evaluator script to test and grade agent performance based on emergency wait times.
• Dockerfile: Container configuration required for OpenEnv compatibility.
• requirements.txt: Python package dependencies.

Setup Instructions

Local Development

1. Clone or download this repository.
2. Install the required dependencies:

   pip install -r requirements.txt
   

3. Test the environment and run the evaluation script:

   python grader.py
   

4. Train the PPO agent:

   python train.py
   

Docker Execution (Evaluation)

As per OpenEnv specifications, you can evaluate the agent programmatically inside a container:

docker build -t openenv-traffic .
docker run --rm openenv-traffic

Features

• State Space: Tracks normal and emergency vehicle queues for 4 directions (N, S, E, W).
• Action Space: Controls 4 distinct green light phases.
• Reward Logic:
  • Penalty for global waiting times.
  • Severe penalty for delayed emergency vehicles to enforce prioritization.
• API strictly adheres to gymnasium formatting to integrate seamlessly with standard RL libraries and OpenEnv containers.

Good luck and let's build the future of AI Agent systems!