AI & CLOUD
Feature: Putting the AI into AIrspace management
Airspace World 2024 offered a window into how digital technologies and AI are expected to transform even the most safety-critical industries, though the palpable excitement from some was partly offset by warnings around reliability.
As has been the case with conferences covering almost any industry in the last nine months, AI and generative AI (GenAI) were seemingly omnipresent at the event held in Geneva last week, though the former was given a much more positive outlook from several prominent speakers.
Industry areas covered under the event’s umbrella including Air Traffic Control (ATC) management are faced with an increasingly complex operating environment. An ever-growing volume of passenger and freight aircraft is being joined by emerging drone and electric vertical take-off and landing (eVTOL) vehicles operating at lower altitudes.
Another issue facing organisation of the skies are an increasing number of rocket launches, including for satellite communications infrastructure, requiring areas above sites to be cleared around planned take-offs.
In these increasingly hard to navigate environments, some see increased adoption of AI technologies as playing a central future role in ensuring safety and the efficiency of airspace management. However the reliability and security of such systems, as you would expect, was a frequent concern.
During a panel on day one entitled The Skies of Tomorrow: How AI is Transforming Air Traffic Management, a group of experts from across the sector gave a thorough overview of the potential role, risks and challenges of AI implementation within air traffic control and associated fields.
Adam Wheeldon, CTO at UK air traffic control company NATS, said over a number of years the organisation had used AI to “farm” its wealth of data “for insights into operational concept improvements and safety performance indicators”.
He added it had more recently used the technology in an intelligent approach and predictive decision-making tool (related to runway use) and a demand capacity balancing platform with London Heathrow Airport as its launch location for the two.
Discussing the intelligent approach platform, he clarified the AI was algorithmic and was “not using deep learning or things like that, but gave controllers support they wouldn’t normally have”. Wheeldon added its most advanced AI was being tested in an R&D project named Bluebird, which uses machine learning agents within a digital twin to take over and control a task. The project included trials with both AI agents and humans.
“We do half an hour with the AI agent controlling traffic in the sector then we switch over and let the human do the same task. Then the human scores the AI as to how effective it is at the control,” he explained. “It’s been remarkable how these agents have been able to manage the traffic safely. More difficult is the expeditious nature: trying to get the flows”.
The AI tends to take a “safer strategy then maybe a human would”, he added. “The next stage of that project is looking at trustworthiness in AI and making sure you can trust the outcomes”.
"It’s been remarkable how these agents have been able to manage the traffic safely."
ADAM WHEELDON, CTO NATS
“What we are finding there is the AI would really like a lot of the human constraints like loops, sectors, small pieces of airspace removed. It would prefer to have the whole flight information region and just control the flights across the entirety of the airspace. That is great, but unfortunately to make it trustworthy you really need to have a human [who] understands what they are doing,” Wheeldon concluded.
No saviour
Later in the panel, head of ATM portfolio and international affairs at vendor Frequentis Markus Klopf highlighted “you can see AI as your personal assistant”, adding it had been working on related projects for decades but “only recent started calling it AI”.
Klopf claimed the technology could increase efficiency in the sector and add safety, but cautioned “I think it can only increase automation to a certain level. We’re not yet ready to go full automation…where everything is done by AI”.
“Research is going in that direction obviously, but in the real, daily life it’s all about supporting [air traffic] controllers”.
Among the applications Frequentis is looking into is use of AI to analyse data streams, speed fault detection and spot patterns which “usually humans would not see”.
It was also showcasing the potential impact of AI on improving voice communication as “pilots and controllers communicate a lot and there is a lot in that speech that can be analysed and help your automation system”.
“The bad news is AI won’t save air traffic management”, he cautioned, adding there were issues in the industry where it “can help, but can’t solve everything”.
As would be expected in an industry with extremely high safety requirements, standards and regulations were always a frequent reference point, especially in regards to GenAI and machine learning systems.
Director General of The European Organisation for Civil Aviation Equipment, Anna von Groote, speaks at an industry event Director General of standardisation body The European Organisation for Civil Aviation Equipment (EUROCAE) Anna von Groote was broadly optimistic on the potential of the technology, highlighting she was “always amazed by the multitude of applications of AI”.
She added the body started looking into the area with its working group in 2019 and is now assessing requirements. “We hope the standards will make the engineering ecosystem more solid and compatible, and really compliment this development of AI from innovation into the marketplace”.
Mark Cooper, VP and CTIO at civil aviation system provider Nav Canada, added “AI is a real conundrum for us”.
“AI won’t save the world, but you can’t ignore it,” he added. “We have a lot of legacy systems, but you can’t ignore the productivity aids and benefits that come with it.”
“We want to embrace AI and one of the first things we’re doing is looking at the policy so we know what we are doing and how we’re going to work with it”.
Going large
However, when it comes to GenAI, despite a number of sessions around the show on the potential of large language models (LLMs), the focus was very much on non-critical use cases, at least until the level of trust is increased.
Elsewhere at the event, CEO and founder of unmanned aircraft player ANRA Technologies Amit Ganjoo highlighted AI was limited to the quality of the data sets, a factor raised regularly across most sectors when this highly-hyped topic is raised.
Across the sessions, many speakers had something good to say about what AI brings to the table but, as with many a debate in the tech industry, the concerns were not about the ability of the technology, rather the reliability and security aspects of new digital systems.
In an industry which lends itself so well to automated and semi-automated applications, it was no surprise to see a strong level of enthusiasm for AI’s potential here. However, given the amount of work which still needs to be done, a world where computers have full control of the skies may still be a long way off, for better or worse.