BluebirdATC is an open-source digital twin of en route airspace, developed by Project Bluebird, a collaboration between the Alan Turing Institute, the University of Exeter and NATS. It provides a safe, reproducible sandbox to simulate realistic air traffic scenarios, develop autonomous ATC agents, and benchmark their performance.
This repository contains the following packages, each with their own README for more information:
| Package | Purpose |
|---|---|
 |
The digital twin — simulate airspace, aircraft, and actions. Docs |
 |
A REST API server for the digital twin. Docs |
 |
Gymnasium environments — train RL agents, single & multi-agent. Docs |
bluebird-hmi |
An optional web-based visualisation package. Docs |
Project Bluebird are hosting an AI agent development competition, the AI(r) Traffic Controller Challenge.
To get started with the competition specific setup see the docs here.
To get started with viewing a scenario in the HMI - make sure uv is installed (installation guide) and then run the following command in a terminal:
uvx bluebird-api@latestThen navigate to http://localhost:8000/hmi/.
You'll see a radar HMI with no scenario loaded. To load a simple I-Sector scenario:
- Select Load new scenario in the top left
- Choose Artificial → I-Sector Two Aircraft → Load
- Press the play icon in the top left
Aircraft will appear in the sector and begin moving. Each label shows the callsign, current flight level, groundspeed, and cleared and exit flight levels - the same information a real ATCO sees on their radar display.
To get started with agent development, we have provides some examples for interfacing with the digital twin:
- here to directly interact with the digital twin
- here for using the gymnasium
- here for using the REST API from any language
Full documentation for the latest release is at https://docs.projectbluebird.ai.
Please see the contribution guidelines if you would like to contribute to BluebirdATC.
