Researchers in the UK start using AI to control air traffic

Researchers in the UK have developed a computer-based model of air traffic management that is controlled by artificial intelligence, rather than humans.

The “digital twin”, or representation of the airspace above England, is the first output of a £15mn research project that aims to find out what role AI can play in advising air traffic controllers and ultimately replacing them.

Project Bluebird is a research partnership between National Air Traffic Services (the company responsible for UK Air Traffic Control), the Alan Turing Institute (a national organization for data science and artificial intelligence), and Exeter University, funded by UK Research and Innovation. The first results of the research were presented at Exeter’s British Science Festival.

AI can be used to improve air traffic control by directing aircraft on more fuel-efficient routes, reducing the impact aviation has on the environment, and reducing delays and congestion at airports like Heathrow in London.

The shortage of air traffic controllers is another problem. They take three years to complete their training.

Richard Everson is a professor of machine-learning at Exeter University. He said that Nats has a larger database of flight records in the past than any other air traffic control body. The researchers use this data to train their AI.

Richard Cannon is the Nats Bluebird research leader. He said, “We’ve been preparing this for the last decade by recording air movements over the UK.” The data includes 10mn flights paths.

The project’s digital replica of UK airspace is now being used to simulate real air traffic to allow human controllers to work with AI agents to process aircraft.

Cannon said that by the end of the 2026 project, the AI agents would be tested in a live “shadow trial” on real-time air traffic data, allowing them to compare their decisions with those of the human air traffic controllers.

He stressed that AI would not be able to determine the actual routing of aircraft.

If the research is successful, Nats and other aviation authorities will likely first conduct extensive trials with AI, lasting several years.

Nats has an advanced computer system that processes data from one of the most congested airspaces in the world, but does not use AI for predicting future flight paths.

Last month , the system failed during the Bank Holiday weekend because it couldn’t recognise a flight with contradictory information. This led to travel chaos. The Civil Aviation Authority launched an investigation into the problem and airlines have demanded compensation from Nats.

Cannon and Everson could not comment whether AI would have prevented the failure of the system, which led to the cancellation of over 1,500 flights.

Everson, however, said that AI could increase the resilience of the air traffic control system and reduce the risk of it failing in the face of unexpected events.

Bluebird’s digital twin covers the London Flight Information Region, which includes airspace in England and Wales.

Everson says the AI system routes aircraft at a distance of 1,000ft horizontally and 5 nautical mile vertically.

This ensures that if the radio communication with the pilot fails for any reason, each flight path will remain safe and without risk of collision for a minimum of 15 minutes.

Cannon said that in every air traffic system around the world, all decisions are taken by humans. “We don’t say that we want the skies above the UK to be automated, but we are pushing as far as safely possible.”