Consolidating AI data to benefit all stakeholders

An abstract illustration of a network.

Partnership is the key.

EUROCONTROL has taken the first steps to developing artificial intelligence (AI) solutions to predicting traffic loads – the next big step will be to start researching ways AI can be used to assist the controller.

For researchers at the EUROCONTROL Experimental Centre (EEC) in Brétigny-sur-Orge, the application of artificial intelligence to safety-critical air traffic control operations has always been a distant prospect. Until now.

“We have been working with AI for eight or nine years now. Three years ago we started a programme in cooperation with a number of universities to use AI for predicting controller workload and sector capacity, aircraft trajectories and runway capacity. Previously, the mathematical models on aircraft performance we used had clear limits and levels of uncertainty. When we applied the statistical approach of AI we had better results and saw the possibilities of moving quickly from exploratory research to deployment. Now, for the first time, we are starting to consider whether we could use AI for some safety-critical tasks – an area we have not been able to consider before. We think now is the time to start considering this, with the aim of producing the first tools within 10 years.”

So what has changed? Why are researchers only now starting to consider what kind of research will be required to move AI into the control room, with the ultimate promise of changing the nature of the controller’s task from tactical air traffic control to airspace manager – something which has been an industry goal for decades? There are three reasons why the time might now ripe for such exploratory research to start.

First, there is the work being undertaken elsewhere, by aircraft manufacturers for example, who have begun to look at how AI could be introduced into the cockpit.

“We are now starting to see other stakeholders looking at AI,” says Pierre Andribet. “We are seeing airlines starting to optimise their operations using AI while Airbus, in cooperation with universities such as the famous Montreal AI centre of excellence, is beginning to look at how AI could be introduced eventually into safety-critical zones of aircraft operations, such as single-pilot configurations. When I talk to manufacturers such as Airbus they say they are convinced many things are possible but it will take time.

“Until now, these safety-critical zones have been a no-go zone for us but we should now start looking at how automation could eventually help the controller, perhaps during low traffic periods.”

Initially new AI tools could help the controller at the planning stage; more accurate traffic-load predictions will mean Network Manager (NM) operators could start to reduce some of the buffers placed into the system as a result of uncertain predictions of traffic loads in a particular sector.

“The more we improve the prediction, the more we can relax the buffers while maintaining safety – you can only remove these buffers if you have clear evidence we are not touching safety,” says Andribet.

Second, AI technology has evolved in terms of data management, tools, processing power and data manipulation while new skills have been developed in university research centres. As the technology has matured, the comfort levels in the Agency with AI and machine-learning techniques have started to grow as the benefits of AI have become apparent. Over the last few years the Agency has looked at introducing AI techniques for more than traffic load forecasts. For example, it has also developed new tools to forecast the impact of the ionospheric effect – where global satellite navigation system signals slow down as they pass through the charged particles of the ionosphere, then through the water vapour in the troposphere – on navigation performance and improve predictability of the system’s performance.

Interested? Subscribe!

Never miss out on the latest developments from EUROCONTROL

Third, there is a new understanding of AI that can be specifically applied to aviation to improve the performance of all stakeholders. And here, believes Pierre Andribet, datasharing is the key.

“I think the development of AI in aviation is a question of synergies,” he says. “We have many data sets in aviation but we only use a small portion of the data and the benefits will be realised in proportion to the amount of data we can access. If we can find a way to federate data from airlines, manufacturers, NM, airports and air navigation service providers while protecting confidentiality, then the benefits for all stakeholders will be substantial. However, at the moment data is in many places. We have started to talk with partners to discuss how NM might in the future be able to share data collected on board aircraft and we are talking to airports, too. What we must demonstrate is that by sharing, all individual actors will benefit more.”

In May 2019 EUROCONTROL organised a multi-stakeholder AI event in Brussels to raise the aviation community’s awareness of potential benefits of AI, and the need to start the process of data-sharing for AI advancement was loudly voiced. It is a role that EUROCONTROL is ideally placed to play, Pierre Andribet believes. The Agency is neutral, pan-European; it has a history of being able to find common solutions and is linked to all the major actors, including regulators and research agencies.

“We have started to launch a partnership network with aviation stakeholders to see what can be done – what we would like to do is create an innovation lab for AI using data federated among actors,” says Andribet. “Partnership is the key.”

But there will be many challenges to overcome before AI can deliver the potential network management benefits which experts believe are possible. It will take time to convince all stakeholders that data-sharing is really in their interests and that a framework has been put in place which achieves the right balance of protecting confidentiality while ensuring clear operational benefits. New levels of trust will have to be built among pilots and controllers that AI-based technologies will reduce workload, increase performance and enhance, rather than compromise, safety. A common understanding needs to be reached on cyber protection and data management. Universities must be brought into the aviation AI community in a systematic way, so people with the appropriate skills can appreciate the benefits of working in the aviation sector. Most importantly, regulators will also need to have access to AI knowledge and skilled personnel in a way which will allow them to make independent judgements on the safety of new AI technologies. Regulators are currently staffed by professionals with many decades’ experience in aviation and aerospace; it will be impossible to simply employ AI technologists directly from universities and research centres.

Then there is the question of global competition. There are no Googles, Facebooks, Microsofts or Amazons based in Europe, attracting skilled technologists into the private sector straight from university. China is starting to invest heavily in AI. But Pierre Andribet is optimistic that Europe has a key role to play in this area, despite global competition.

“Most of the AI tools we are using are public and widely used and I don’t think Europe is lagging behind in terms of skills and education; I know it is a priority of the European Commission to develop even more new technical universities. It is true we are not filing as many patents as in the USA but we are developing the schools and aviation is an attractive employer. I am confident we have the capacity to retain the skills – but we will need access to volumes of data and if we cannot accelerate this access, we cannot develop.

“AI is embedded in our work now; it figures in at least 15 research programmes,” he says. “It will be a necessary tool in many aspects and used in almost all our different areas of research. The goal is not to build a team dedicated to AI, but to have teams dedicated to managing data, with people using AI technologies in many domains – NM, ATC, airports and even communications, navigation and surveillance (CNS) technology research. The key issue is to bridge the gap between the people who know their domain and the people who know AI. But we need to act now, we can’t wait for five years.”

Read our full Skyway issue

Learn more about the new era for aviation powered by AI and our role in it.