Air traffic management (ATM), even at its most efficient, has traditionally been a real-time affair: directives are issued as situations arise and information is processed by human controllers balancing a range of monitoring and decision-making responsibilities.

But as global commercial air traffic, according to the International Air Transport Association (IATA), pushes toward returning to 2019 levels by 2024 and resuming robust pre-pandemic growth trajectories, air navigation service providers (ANSPs) increasingly are embracing artificial intelligence as a means of keeping pace. Worries that ANSPs will be challenged to hire and train sufficient numbers of controllers are also driving the ATM sector towards AI.

Indeed, airlines in the U.S. partly laid the blame for the extensive flight delays and cancellations in the summer of 2022 on FAA controller staffing shortages, particularly at the agency’s Jacksonville, Florida en route center that is critical to managing air traffic in the crowded region of the southeastern U.S.

The idea behind adapting AI capabilities to ATM is to ease the burden on controllers, enabling better decision making based on more complete information, and—longer term—to allow ANSPs and airports to do much more detailed advanced planning.

“I think, in simple terms, AI can take us from reactive to proactive planning,” NAV Canada program director of service delivery Blake Cushnie told Avionics International in an interview. “Reactive planning for a major airport, particularly an international airport, means that by the time [controllers] want to make a decision, a lot of [incoming] aircraft are already airborne. The average flight time to Toronto is well over three hours.”

This leads to longer-haul flights getting priority over shorter-range flights, he explained, simply because the one- or two-hour flight can still be canceled or delayed when a situation develops. But technology solutions that could reliably predict the air traffic scenario at an airport a day or more in advance would give controllers higher levels of flexibility and make the system more efficient, according to Cushnie.

NAV Canada is prioritizing AI research and development, partnering with the Massachusetts Institute of Technology (MIT) Lincoln Laboratory to develop high-end ATM technology and processes. It is also building a digital twin of Canadian airspace.

“The controller role will evolve significantly,” Cushnie said. “It’s important to remember that humans are good decision makers, but they do not necessarily make good monitors. Systems are fantastic at monitoring for faults and errors. But somebody still needs to make the decision.”

At least for the foreseeable future, AI will “support the decisions, but the controller will still be there to make decisions and to manage traffic accordingly,” Cushnie explained.

Where AI is already playing a role in ATM is in managing airport surface movements. “It's quite a heavy workload” for a controller to monitor everything happening on and around an airfield, Marco Rueckert, head of innovation at Ottawa-based Searidge Technologies, told Avionics. “That's something that AI is very good at: really looking for these patterns in the video.”

Searidge’s air traffic control AI platform, Aimee, is used at London Heathrow Airport and a host of major airports around the world. Utilizing an extensive array of cameras covering an airport’s surface, Aimee teaches its artificial neural network to detect aircraft and other airfield objects and provide a detailed, unified view on controllers' digital monitors.

Controllers using Aimee are given more time to make decisions and increased situational awareness, Rueckert said, adding: “We’re taking some of these monotonous tasks away from the controller and then having the controller become a supervisor saying, ‘I'm going to watch over what the AI is doing and I'm going to monitor the output before I give any instructions.’ AI can feed the controller information that is of a much higher value than just looking out of the window.”

For example, Aimee ground level cameras at runway exits (being trialed at Heathrow) could give controllers much greater awareness in low cloud and reduced visibility scenarios, potentially allowing more flights to land and take off.

“By mounting cameras at ground level facing runway exits, Aimee will have full visibility of the aircraft,” according to Seabridge. “Aimee uses advanced AI segmentation algorithms to determine the aircraft positioning and informs the end user when the aircraft clears the key thresholds.”

Learning From The Past

The constant, expansive video surveillance of the airfield also gives controllers data on past performance and tendencies that can be deployed in future planning, a primary goal of using AI in ATM.

NAV Canada is developing a digital twin of Canadian airspace, still in a pilot phase, with the goal of enabling significant advanced planning. The digital twin will access vast amounts of data on air traffic in a given market—Toronto, for example—to provide controllers a view of likely future scenarios.

“This is a cloud-based application that will merge data sources together,” Shavin Fernando, NAV Canada’s manager of advanced analytics products, told Avionics. “We’re able to quantify and measure things we haven’t been able to before … We’re trying to provide more data points and a more holistic view for controllers making decisions.”

Eventually, NAV Canada expects the digital twin could give controllers the capability to consistently plan 24 or more hours in advance by predicting the most likely traffic scenarios. “The aviation system is an ecosystem,” Fernando said, explaining that air traffic can be most efficiently managed by understanding how different components of the system affect one another.

“This is not a silver bullet,” he cautioned. “It’s a tool allowing controllers to connect the dots.”

NAV Canada is still at an early stage of developing the digital twin, Fernando noted. Cushnie said the digital twin “is going to give us a snapshot of recommended sector openings and closings. Airspace is generally split into sectors [monitored by controllers]. A lot of these sectors get combined when it's quieter and split open when it's busier. This will help us predict operations so we can better know when we need to open and close sectors” and how to better plan controller staffing.

Of course, the most difficult challenge in managing airspace is predicting weather and determining how it will affect air traffic. NAV Canada and MIT Lincoln Laboratory in July 2022 announced a partnership to “develop state-of-the-art technologies for managing capacity and demand imbalances for airports, terminals, and enroute airspace under challenging weather conditions,” according to a joint statement.

The first phase of the project will combine various weather models “to gain a more complete and more accurate picture of weather impacts,” NAV Canada explained, adding that the technology solutions being developed “will apply sophisticated algorithms to the data that generates these pictures to predict capacities at various key points in the aviation system. They will then compare those capacities to traffic demand expected at those same points up to 12 hours into the future.”

Cushnie said NAV Canada will work with MIT “from a machine learning or AI perspective on better weather predictions, better storm predictions, particularly in areas where we don't have radar coverage today, which is a lot of Canada's north.”

He noted that decisions about canceling or delaying flights are made by airlines and ANSPs that may be using different weather forecasts, so “it's very hard to get everyone on the same page.” But a system that could provide all stakeholders with a cohesive forecast and specific predictions could give controllers and airlines much more planning flexibility.

Cushnie explained: “We say, ‘tomorrow, we think between this time and this time we're going to lose a couple of runways, so let’s plan accordingly.’ That's really the level of weather prediction accuracy we're working on with MIT.”

AI tools could quickly gather data from previous days when a similar forecast was made and then make predictions based on that data. As Cushnie explained, an AI system could “look at what happened with the same forecast, look at wind models and other data pieces, and use algorithms to try to send a very clear picture about what the operational environment likely looks like and what the probabilities of an aircraft being able to be on a certain runway at a certain time are.”

He added that AI’s biggest benefit to ATM will likely be in reducing unpredictability. “The most important thing that we need in aviation is predictability,” Cushnie said. “And I think this is the area where machine learning is most valuable—learning from yesterday to try to predict tomorrow better. That's what we're trying to do. We're very keen to harness new technology that allows us to be more predictable.”