Airbus’s flying AI with no pilot: learning to fly ourselves, sort of

When the Airbus A5P was developed, its designers imagined it was capable of flying fast, safe, with good fuel efficiency. Perhaps that was a bit optimistic, but it was used by a Japanese airline until the 1970s. Then Boeing built a much bigger version, the BA50, and the A5P – with its four engines and 16 seats – vanished again.

Their replacements are giving hope that the once-defunct type can someday fly itself. Artificial intelligence called “neural networks”, first used by Nasa to investigate time-delay and vibration detectors, are being harnessed to predict where the plane should take off and how quickly it should take off. Thanks to artificial intelligence (AI), an AI programme called Predictive Flight Management System can take off and land automatically in safety, using data from the plane’s sensors to assess turbulence, wind strength and temperature.

This is an opportunity, according to experts such as Philippe Huët, Airbus’s chief digital and innovation officer. Huët says that, with AI flying itself, a plane operator can “automatically take over and analyse the data stream”, such as from ADS-B (also known as air traffic control) to assure the best possible performance. The result could be lower fuel consumption and higher profit margins.

“If you just look at traffic data from the end of 2017, there was a 40% decrease in travel compared to last year. In other words, we need to do everything we can to minimise the effect of the falling traffic. As a European air transport [company], we must demonstrate that flying itself is the future,” says Huët.

An Airbus A4X, a second-generation A5P (above), and a prototype for its flying-life successor, the A5P-RAM. Photograph: Airbus

This would indeed fly first into very complex safety tests, before it would be cleared for passengers in 2023. Even if it succeeds, however, it would likely not replace our beloved flights – it is still 10 times heavier than an airliner – and Airbus has not ruled out such things as airports, batteries and communications.

The A5P, an early example of AI and in-flight systems, was developed by Peter Kiilinger of the Swiss Federal Institute of Technology. It was created to resist the often-dangerous way that propeller-driven planes would produce a clap of thunder or shake violently as they reached higher altitudes. In the 1960s, when Kiilinger received a grant to develop the system, it was considered too dangerous and was kept under wraps. “There were a lot of problems,” recalls Kiilinger. “The propeller and the landing gear are often behind the cockpit and very close to the pilot’s head, so there could be a lot of violent shaking, or, worst of all, colliding forces.”

The A5P solved these problems by using four computers within the propeller, which allowed the computers to sense and predict what was happening around the plane. This allowed the engines to start rotating less quickly, less forcefully, therefore reducing shaking.

The company said that the A5P ran at a rate of around 100 flight cycles a minute, and often managed 100 hours a week of flight time. However, it wasn’t well-designed for the jet age and the piece of tech had a very short life span after flying for three months. The aircraft that does manage to make it to 2023 is a revamped version of this A5P.

What remains to be seen is whether it will find a following from the general public. For instance, if AI pilots weren’t willing to take the risks themselves, it’s quite a stretch to imagine they will abide by an autonomy policy that puts them in charge of the plane.

It may ultimately depend on the airlines. According to Luke Stevens, project leader of Predictive Flight Management Systems, US airline JetBlue was the first to deploy the system on a Boeing 767-300ER. In November 2018, they took the experimental software and flew it using sensors on the runway. They see a possibility of it being integrated into future flights.

But if you are an enthusiast, consider the example of EVA Airways, a Cyprus-based carrier that flew autonomously and quickly from its main runway on 4 April to a secret location five miles to the west in the production region of Toulouse. It performed its scheduled flight using GPS technology, rather than by “waving” the plane around the airport. Here, an “autonomous” flight simulator was used to simulate the selection of flight route.

Leave a Comment