When fog rolls across a runway, the delays that follow are rarely caused by the aircraft. They are caused by what the people in the control tower can no longer see. AI air traffic control systems are now being used at some of the world’s busiest airports to fill that gap — watching runways through cameras when human eyes cannot, and telling controllers the one thing they need to know: has the last aircraft cleared the runway yet?
It sounds like a small question. At a capacity-constrained airport, it is worth roughly one landing in five.
Why fog costs an airport 20% of its landing capacity
Controllers in a conventional tower work by looking out of the window. They watch an arriving aircraft touch down, roll out, and turn off onto a taxiway — and the moment it is clear, they can clear the next arrival to land.
Take that view away and the process slows down. When visibility drops, controllers fall back on ground radar, which updates more slowly and with less certainty. To stay safe, they leave a larger gap between arrivals.
At London Heathrow, the effect is measurable. According to EUROCONTROL, the airport’s 87-metre control tower is affected by visibility constraints on roughly 12 to 15 days a year, with individual low-visibility events typically lasting 30 to 90 minutes. During those periods the airport loses up to 20% of its landing capacity.
At an airport running near 100% of its declared capacity, a 20% cut does not simply delay 20% of flights. It cascades — aircraft hold, crews time out, connections are missed, and the recovery stretches across the rest of the day.
The bottleneck in fog is not the aircraft’s ability to land. It is the controller’s ability to confirm that the runway is empty.
How is AI used in air traffic control?
AI in air traffic control is used mainly as a decision-support layer, not as an autonomous controller. Computer vision models watch runway cameras and confirm when an aircraft has cleared, speech recognition transcribes radio calls to pre-fill controller displays, and predictive models forecast fog, demand and delays. Human controllers still issue every instruction.
That distinction matters. In every operational deployment running today, the AI system is an extra pair of eyes feeding information to a licensed controller who retains full authority. No air navigation service provider is letting a model tell aircraft when to take off or land.
The four functions most commonly deployed are:
- Computer vision — tracking aircraft on the airfield through high-definition cameras, day or night
- Natural language processing — interpreting controller-pilot radio transmissions
- Flight data and surveillance fusion — combining radar, GPS and flight plan data into a single picture
- Weather processing — forecasting visibility, wind and storm impacts hours or days ahead
What happened when Heathrow tested AI on 50,000 landings?
Heathrow ran the most heavily documented trial of AI air traffic control anywhere. NATS, the UK’s air navigation service provider, began the programme in 2018 with Ottawa-based Searidge Technologies, whose neural network framework is called AIMEE.
The setup was deliberately unglamorous: 18 ultra-HD 4K cameras mounted on the tower, plus around 20 more positioned along the airfield, feeding live images into a digital tower laboratory. AIMEE’s job was to watch arrivals on the northern runway and tell controllers the moment each aircraft had cleared.
Between January and March 2019, the system was validated against more than 50,000 arriving aircraft. NATS reported that AIMEE identified aircraft reliably in poor weather and in near-darkness, and that it delivered runway-clear confirmation far faster than the airport’s existing multilateration ground movement radar — EUROCONTROL puts the difference at roughly 25 times faster.
“The AI model can operate day or night and that’s part of what the validation process is showing: that the performance regardless of whether it’s day or night conditions is equally good,” said Andrew Taylor, NATS Chief Solutions Officer.
NATS then extended the trial to full low-visibility conditions — the scenario the whole project exists to solve. The prize is recovering that lost 20% of landing capacity on fog days without touching a metre of concrete.
Three other places AI is changing the tower
Listening to the radio: speech recognition in Europe
Controllers spend a surprising amount of time typing what they have just said. Every clearance issued by radio has to be reflected in the data block on the radar screen.
Research led by DLR, the German Aerospace Center, working with Austro Control, DFS and other European providers, has demonstrated that automatic speech recognition can pre-fill those radar labels directly from the controller’s voice — reducing workload and improving flight efficiency. EUROCONTROL is also applying the same technology to training, using speech recognition to replace human “pseudo-pilots” in simulators so trainees can practise without a full support team.
Predicting fog days in advance: Delhi
Delhi’s Indira Gandhi International Airport faces a harder version of the problem than Heathrow: dense winter fog that can close in for weeks. Its answer combines infrastructure with prediction.
The airport’s Airport Predictive Operations Centre fuses real-time airside data with predictive weather models. Those forecasts draw on WiFEX, a long-running fog research programme run by the Indian Institute of Tropical Meteorology with the India Meteorological Department, which reports around 85% accuracy in anticipating dense fog up to 36 hours ahead.
Paired with CAT-III instrument landing systems now installed at both ends of all three main runways, the airport can handle roughly 30 landings an hour in low visibility, and expects fog-related recovery time to drop from about six hours to two.
Towers without windows: the United States
The US is moving in a related direction. The FAA has gone to industry for remote and digital tower systems — where the out-of-the-window view is replaced entirely by cameras, sensors and panoramic displays — with a potential procurement of up to 50 units over five years. The FAA Reauthorization Act of 2024 directs the agency to run a formal approval programme extending to at least three airport locations.
Digital towers are not inherently AI systems. But once the view is a camera feed rather than a pane of glass, every AI capability described above becomes far easier to bolt on.
What AI in air traffic control cannot do
It is worth being blunt about the limits, because aviation coverage tends to overstate them.
AI is not separating aircraft. It is not issuing clearances. It is not replacing controllers, and the certification path that would allow any of that does not currently exist. Aviation safety regulation is built around demonstrable, deterministic behaviour, and machine learning models are difficult to certify against that standard.
There is also a workload trap. A decision-support tool that is right 99% of the time can quietly erode a controller’s own vigilance — the same automation-complacency problem the industry has studied in cockpits for decades. This is precisely why deployments so far are narrow: one clearly defined question, one verifiable answer, human in command.
The pattern mirrors what we have seen in other transport sectors, from AI cutting cargo ship turnaround time in ports to AI predicting satellite collisions with space debris: the models handle perception and prediction, humans keep the authority.
What this means for passengers
You will never see this technology. There is no app, no announcement, no visible change at the gate.
What you may notice is a foggy morning that does not turn into a ruined travel day — an airport that keeps landing aircraft at close to its normal rate instead of dropping a fifth of them, and recovers in two hours instead of six. For airports that are physically full and cannot build another runway, reclaiming capacity that weather takes away is one of the very few options left.
Frequently Asked Questions
Does AI control aircraft at airports?
No. AI systems in air traffic control provide information to human controllers — such as confirming a runway is clear or forecasting fog — but licensed controllers issue every instruction to aircraft. No regulator currently permits AI to separate or sequence traffic autonomously.
Which airports use AI air traffic control today?
London Heathrow ran the most extensive trial, using Searidge Technologies’ AIMEE system with NATS. Delhi’s Indira Gandhi International Airport uses AI-driven predictive analytics for fog operations, and several European providers including DFS and Austro Control have trialled speech recognition tools.
How much capacity do airports lose in fog?
Heathrow loses up to 20% of its landing capacity during low-visibility periods, according to EUROCONTROL, because controllers must leave larger gaps between arrivals when they cannot see the runway. The airport experiences such conditions on roughly 12 to 15 days a year.
What is a digital or remote tower?
A digital tower replaces the physical out-of-the-window view with high-definition cameras and panoramic displays, letting controllers work from a remote location. The FAA is pursuing deployment across US airports, with procurement of up to 50 systems over five years under consideration.
Will AI replace air traffic controllers?
Not in the foreseeable future. Aviation certification requires predictable, auditable system behaviour that machine learning models struggle to demonstrate. Current deployments are narrow decision-support tools designed to reduce controller workload, not to remove the controller.
