Artificial IntelligenceCryptocurrencyMetaverseCybersecurityTech Forward

Alaska Airlines pioneers A.I. to plan flight routes, saving fuel and time

July 27, 2021, 11:00 PM UTC

One morning last week, Alaska Airlines flight 1380 thundered down the runway of the Seattle-Tacoma Airport and ascended into clear blue skies, bound for San Diego.

The flight pushed back from the gate five minutes ahead of its 6:10 a.m. scheduled departure time. But its journey really began several hours earlier, about a mile away from the airport, on the sixth floor of Alaska Airlines’ new steel and glass headquarters building, known as “The Hub.”

That’s where flight dispatchers working for the airline plan flights, deciding on the precise route each aircraft will fly to reach its destination. Dispatchers are licensed by the FAA and share legal responsibility for an aircraft’s safety along with its pilots. For the past year, some of those dispatchers have had help from an adept new colleague: an artificial intelligence system created by a small Silicon Valley startup that can often make better predictions about variables such as weather and air traffic than even experienced human flight planners.

During a trial of the software, the airline achieved big savings in jet fuel and lower carbon dioxide emissions, as well as seeing improvements in its on-time performance and reliability. Now it is rolling out the system, called Flyways, to help dispatch all of its flights in the lower 48, making Alaska the first airline to use this kind of technology so extensively. “This is as game changing for aviation as Google Maps and Waze has been for driving,” says Pasha Saleh, Alaska’s director of flight operations.

While Alaska Airlines is the first commercial carrier to use this software for flight dispatching, it’s unlikely to be the last. Increasing use of such systems could make a big impact on the aviation sector’s carbon dioxide emissions and lead to fewer flight delays. What’s more, the software, which was created by a company called Airspace Intelligence, could help pave the way for a revolution in U.S. commercial aviation that many in the aviation sector as well as some environmentalists and politicians have long-sought: an “open skies” air traffic system (technically called “trajectory-based operations”) that will allow planes to fly the most direct route to their destination, doing away with the system of pre-defined waypoints.

Selecting the right ‘highway in the sky

Today, airplanes in the U.S. don’t fly a straightline path between takeoff and landing. Instead, they zig-zag in the general direction of their destination, dog-legging between a series of navigational waypoints. These waypoints form what are essentially highways in the sky. But there are many different possible routes between waypoints and it is up to the dispatchers to determine which of these skyways the plane will travel to reach its destination.

To do this, the dispatchers take into account weather and wind forecasts, as well as military and other airspace restrictions, and sometimes reports of turbulence provided by pilots who have already flown that same route earlier in the day. Finding this information often requires the dispatcher to flip back and forth between a dozen different government, private weather forecasting, airport, and internal airline databases. In addition, dispatchers may consider current airport congestion levels, as well as their own intuition, based on hard-earned experience, about how crowded runways and landing approaches will likely be at the scheduled departure and arrival times.

But what the dispatchers don’t factor into their calculus is the actual FAA-filed flight plan of every other aircraft planning to be in the sky that day and how all those planes might interact, leading to cascading congestion along the flight route. It’s simply too much information for any human to juggle. “It would be like doing nine Rubik’s Cubes at the same time to do that,” Saleh says.

In fact, dispatchers rarely plan an entire route of waypoints from scratch for each flight; doing so would be too time-consuming. Instead, they select from a limited menu of “canned routes” worked up in advance, Saleh says. If the flight is from Seattle to Boston and there’s a strong westerly wind, the dispatcher might choose “Seattle to Boston Route 1,” whereas the go-to routing when the wind is from the south might be “Seattle to Boston Route 3.”

But Flyways, the A.I. software package, allows Alaska’s dispatchers to break away from canned routings. Unlike a human, the A.I. system—actually a collection of different machine learning modules linked together—can calculate the probable position of every other aircraft in the sky that day and how it will impact congestion along the route. It can better predict how weather systems will form or dissipate, opening new routing possibilities. It can plan a brand new, custom route between waypoints to take advantage of these factors. And it can do all of this in mere seconds. When it sees a custom route that will save fuel compared to the canned route, it will suggest this alternative to the dispatcher.

Flyways doesn’t replace the need for a human dispatcher, who still has a legal duty to make the final call on the flight plan. In fact, in Alaska’s trial of Flyways, its dispatchers chose to accept the A.I.’s alternative route suggestion only about one third of the time, Saleh says. “Sometimes the dispatcher might not feel comfortable with the weather,” and the consequences if the software gets its prediction wrong, he says. For instance, the A.I.-selected flight plan might be sending an aircraft directly into what looks like a dangerous storm because it is forecasting that the storm will have dissipated by the time the flight reaches that location, but the dispatcher may feel it is better to err on the side of caution and select a different route. There are also certain routes the FAA prefers commercial aircraft use and dispatchers know the agency may reject alternatives, even if they are more efficient. “At the end of the day it is the dispatcher’s judgment which is why they are licensed, and safety is their number one priority,” Saleh says. But Alaska also expects that dispatchers will gradually start accepting more of Flyways recommendations as they become more accustomed to using the software.

Alaska put the system in place primarily to make strides towards its sustainability goals, says Diana Birkett Rakow, the airline’s vice president for public affairs and sustainability. The airline has announced a five step “roadmap” to reach net zero carbon emissions by 2040. It includes bringing on more fuel-efficient aircraft like Boeing’s 737 Max, switching to newer, more sustainable forms of jet fuel, and purchasing carbon offsets, as well as eventually migrating to a new form of propulsion—either electric engines or ones that run on hydrogen.

But the first step on Alaska’s path is what the airline calls “operational efficiency.” That means ringing fuel savings from its existing fleet and routes. And that’s where Flyways comes in. The airline’s pilot of the software, on a small fraction of its normal flight schedule, saved 5.3 minutes of flying time per flight. That translated into a total of 480,000 gallons of jet fuel and 4,600 tons of carbon emissions. For an airline that before the pandemic was burning 750 million gallons of aviation fuel annually, that may not sound like much. But Birkett Rakow says the savings will grow as Alaska returns to a normal number of flights and as more dispatchers begin to use Flyways.

What’s more, Saleh says, Flyways had some unexpected added benefits: Because it could foresee and avoid congested air routes, it improved the airlines on-time reliability. And that has meant the airline has actually been able to take some of “the padding” that Alaska, like all airlines, builds into its schedules to account for taxiing times and air traffic delays.

Green screens and paper maps

Airspace Intelligence, Flyways creator, is a Silicon Valley startup founded by a group of engineers who had previously focused on simulators that could help self-driving cars navigate crowded cities. But Philip Buckendorf, Airspace’s chief executive officer, says he and his co-founders soon realized there were a lot of other companies out there already making similar systems. So they decided to look at other forms of transportation. Visiting an airline’s dispatch center was eye-opening, Buckendorf says.

“We thought it would be very sci-fi, and it was the exact opposite,” he says. “Lot of IBM green screens and weather maps on paper.” The company’s three co-founders decided to focus on airline routing.

Airspace wound up working with Alaska because it was simply most receptive to trialing Airspace’s tech. The airline is proud of its history pioneering new technology, particularly when it comes to navigation. Executives like to talk about how the carrier was one of the first airlines to deploy GPS and one of the first to use automated landing systems that could be deployed at smaller, remote airports.

Airspace Intelligence’s biggest competitors are flight dispatch software sold by Boeing and its subsidiary Jeppesen, called JetPlanner Pro, and a system called Flightdocs sold by aviation software firm ATP. Germany’s Lufthansa Systems has also built its own flight routing system called NetLine. Several startups are offering software to help plan flights for fleets of autonomous drones. These include Skyward, SkyGrid and Dronebuddy. But Buckendorf says most companies offering routing software to airlines are focused on optimizing individual flight plans.

What makes Airspace Intelligence’s Flyways different, he says, is that it tries to find the most fuel-saving routes across the entire fleet of aircraft, and takes into account how all the other planes using the skies will impact that fleet-level solution. Another key difference: In the past, airline dispatchers would make a single routing decision, usually two hours before takeoff time, and not modify it after that. Now, with Flyways, the route can potentially be updated several times during that two hour window to take advantage of routes that will offer better fuel savings, Buckendorf says.

The future of U.S. air traffic management

Flyways incorporates several algorithms, using a variety of different machine learning techniques. Some involve supervised learning, where the system learns from historical data. Others are based on reinforcement learning, where the system learns from trial-and-error in a simulation. The modules are then put together into a single user interface. One benefit of this modular approach, though, is that it is relatively easy for the Flyways A.I. to explain to a dispatcher why it is making a particular recommendation. That is important, Buckendorf says, if dispatchers are going to learn to trust Flyways’ suggestions. It’s also important for helping them decide when they might not want to follow the A.I.’s advice.

Today, Airspace Intelligence employs about 30 people. It has raised what Buckendorf says is “a quite substantial” amount of venture capital funding—although he declined to give an exact figure—from investors that include Spark Capital, Google’s A.I. fund Gradient, Bloomberg’s venture capital arm Bloomberg Beta, Renegade Partners, and mutual fund company Franklin Templeton.

Alaska is Airspace’s first customer; it is in active discussions with other U.S. airlines as well as various parts of the U.S. government, Buckendorf says. He says the more airlines that use Flyways, the better the routing ought to become for all of them—much like the more drivers who use Waze improves the road navigation guidance for each of them—because the system has more data from which to learn. There is also a benefit to every airline of having less congestion, as it means fewer air traffic delays and better on-time performance for everyone, he says.

Flyways—or a system like it—could also serve as the basis for a new air traffic control system if the U.S. ever moves to a kind of “open skies” model where aircraft can fly any route they choose to reach their destination, so long as they don’t violate restricted airspace. Airlines have long pushed the FAA to move to this model, but to date, the U.S. air traffic control network doesn’t have the kind of infrastructure to support it. Birkett Rakow says Alaska is among the carriers that have been talking to the Department of Transportation about budgeting more money for upgrading the air traffic management and control systems across the country to hopefully enable a move to open skies.

Correction, July 28: A previous version of this story misspelled the last name of Alaska Airlines’ Vice President for Public Affairs and Sustainability Diana Birkett Rakow.

Subscribe to Fortune Daily to get essential business stories straight to your inbox each morning.