Vincent Lambercy:
Artificial intelligence has played a major role in aviation and air traffic control in the latest years and we've seen brilliant developments there. And this will be the topic for today. And for that I have two guests today - Max Bezzina, from INGENAV and Richard Landreville from Betterizon. My name is Vincent Lambercy, this is the Radar Contact podcast brought to you by FoxATM. So guys, welcome to the podcast. Max, can you start and introduce yourself rapidly, please?
Max Bezzina:
Yes. Hello Vincent. Thank you. My name is Max Bezzina. I'm a CEO of INGENAV a company which delivers consultancy services and training services all around air traffic, air traffic control. My background personally, I started as an air traffic controller a while ago and moved into training and proficiency and later on into management and I've been at this with INGENAV for now the last eight years.
Vincent:
Thank you very much. Richard, welcome as well. Can you please introduce yourself too.
Richard Landreville:
Hi Vincent, Thanks for having us. So my name is Richard Landreville, I'm the principal and founder of Betterizon Consultants and similar to Max, we work on many topics together, but individually I work on strategy, we do ops excellence, innovation, some technology developments, but prior to starting this company back in 2018, my background as well is as an air traffic controller back in the mid 90's starting working in Canada and then bringing my experience and working in Switzerland as an air traffic controller for Skyguide, and thereafter getting into various management roles where I held the role of head of air traffic management for Skyguide Virtual Center up until mid 2018.
Vincent:
Thank you. So today we'll talk about a project known under the code name of "CHarlie", which is an AI-based project led by different companies centered an operated by Skyguide and Skysoft mostly, but Richard, can you bring us through what brought these companies together and what the base idea was behind "CHarlie"?
Richard:
Thanks Vincent. In essence, Max and I had been working together on various projects within Skyguide from 2016 and just through time and working together, we saw that a lot of our ideas were aligned as for where we thought ATM was going into the future. And so sometime in 2018 we got together and said, well you know, Max had a couple ideas and he wantsedto bounce off me and so we sat down, had a meeting and discussed and said, well, this is what we think the future could look like. And together we've put together a bit of a one pager that we pitched to Skyguide and then from there as an idea of saying this is what ATM system support or decision support could look like in the future. And from there got into an agreement with Skyguide to move forward to elaborate an operational concept around what we were thinking and splitting a, let's call it an initiative, it wasn't considered a project at the time, but more of an initiative to delve into the researching what could be done in the future.
From there we partnered up more formally with Skyguide and Skysoft and then both representing our individual companies, Max and I and call it maybe a mini consortium of four companies where we work with Skyguide and Skysoft, obviously Skysoft being the system and the software provider Skyguide being more the sponsor now more and more being involved in the pre-implementation of some of these products, providing resources and valid and current air traffic controllers. And then Max and myself being I guess the people with the thoughts behind this kind of the vision and some of the ideas and helping the team, guiding the team into the developing to the products.
Vincent:
How you mentioned "CHarlie" is not a tool, it's a set of tools that targets ATM decision support. We will go into that a bit deeper later on, but Max maybe can you first explain to our audience what are the four tools that make "CHarlie"
Max:
What we wanted to achieve now with "CHarlie" and "CHarlie" is the name of a person and it's also, it's not an acronym. We wanted to achieve in our vision, an assistant support to decision making as if a buddy and other air traffic controller would be helping me or Richard or any other controller in our tasks. So in this thus what we have developed is a number of tools as you mentioned, to try to cover as many azimuths as we could think at the starting stage of the help that we wanted. So we wanted to cover both nominal and non nominal situations. I'm going to explain that in a minute and also longer term and shorter term support in terms of the spectrum that you would have in tactical ATC from planning to actually executing and turning aircraft or climbing and descendants. So there we divided these areas into four and I'm going to simply mention them and say a bit the objective of each one of them.
We have CORA, which is the conflict resolution advisory. With this tool, what we aim at achieving is having support to decision making into how to resolve medium term conflicts. So we're talking in the horizon somewhere around six and 12 minutes or so that covers well kind of a nominal situation where you have a conflict. Conflict in air traffic control is not something you want, but it's something that you normally get quite often as an initial state. So that was one thing. Another area we thought that supports the decision making and using data and machine learning, as we're going to discuss a bit further on, I guess, is when you have a lot of data - is to detect anomalies. So we have an anomaly detector based on data, on historical data. Normal modern ATM systems nowadays are equipped with adherence monitoring of all sorts. So adherence monitoring to routes, adherence monitoring to flight - to the level, checking whether your label is aligned with the Mode S downlink that is linked to the FMS, etc.
However, we found scope as well that with all the historical data you have, you may augment this adherence monitoring with an anomaly alert and detector. So we included that. We thought it should be discrete, it should not be given over importance to the whole system. However, if based from data you have an extra body, "CHarlie" saying, well there's something strange going on, I do not usually see this, it's an anomaly, it can be given. So those are two. Then we try to look at something as a third area, something very normal that we normally do as controllers when we get busy - we get help by our planner or another controller to tell us "Hey, you can do this with this aircraft". Now that aircraft is clear to climb, now that aircraft is clear to descend, don't forget to send this aircraft away, etc.
So we have the third tool it's called "The ready for" which we call colloquially "the nudger". And the idea is exactly this, is that when aircraft are ready are clear of other traffic to climb or descend or to do any other action, you get a kind of a support, saying "you can do it". But we wanted to go ahead of the support of saying, well, we can kind of do this in a coded way where well, you are in your area of responsibility, the aircraft needs to climb and therefore deterministically determined to climb. No, we wanted to get previous information of data when similar aircraft with similar flights would normally be climbed or descended and then having a safety check on that. So that's the third area. And the fourth area which we have in "CHarlie" is what we call FEED standing for flow efficiency and early dispersion.
And the idea we had there is that if we wanted to reduce the workload, the technical workload of the controller, the more we can look outside and plan to reduce areas, which we call hotspots in the middle of our sectors, by taking early and predictable solutions, then the workload will be less and therefore the capacity increased for the tactical controller for the executive. And there this is a function that right now a controller, planning controller is doing. We wanted to augment that function by providing an extra tool with data to be able to expand this planning horizon from 10 minutes before entering your sector to anywhere between 20 and 10. So there the four areas if I just summarize, are looking from augmenting the planning perspective to dealing in a shorter term as much as close to anomalies and is looking for also normal things such climbs and descents and transferring aircraft to things which are less normal conflicts to quite abnormal anomalies, which we're talking about the 0.01 percentile which we as humans sometimes miss to see. And with all that, the idea behind it is that we are providing a support to our controllers to reduce the cognitive workload and therefore increase capacity and maintained or increase the safety that we have in Europe these days.
Vincent:
I find it really interesting how you speak about "CHarlie", like a buddy controller, a part of the team and assistant. And what is behind that is that "CHarlie" is not designed or is not intended to make decision to really solve the problems. But is my feeling when I hear you is that you say it's here looking above your shoulder and tells you, hey here you have something, look at that here, you have something unusual. Is that the design intention from the beginning to go in that direction? Or do you see at some point an evolution towards "CHarlie" going further down and say, Hey look, you have that conflict, you should do that and I can do that for you if you want. Or do you want always to keep a human in the loop?
Max:
Well, it is by design right now. This is what we wish to achieve. We believe in maturity and maturity models such as in quality or in value. If we think about it, a lot of the automation support we have nowadays, it goes to the level of analyzing the information, notifying me that I have a conflict and that I need to do something as a controller. Notifying me that something is happening. Going, if we want to actually taking over and implementing actions is jumping a very crucial step in which we believe that both operationally and also from the deployment of technology, we are not mature enough. There are many and when we are not mature enough and there are incognitas, then there are risks. What we wanted to do from the outset is to be pragmatic. This has been a project in which we are, yes, innovative in many ways, but we are not into fundamental research or exploring the possibilities.
We want it to be able to implement in a fairly medium term horizon. So yes, it is by design that we want to grow in maturity in a number of areas as you mentioned that us as operators and in terms of concept of operate of operations, we get used to have now an automated model agent who is giving us advice and support in decision making but not taking the decisions. We're not there. We want it as well to implement use of machine learning and data into tactical ATC. That's something innovative. The bigger the transformation, the higher the risk that we will find issues and risks in the middle. So yes, it's by design. However, this is not to exclude that once we are learning and maturing with this and once this is implemented that next steps would be starting to look and learn and how to delegate a number of actions to the system. But right now that is not the focus and this is what we want to achieve. Actually, as you said, Vincent, a supported decision making as we imagine "CHarlie" would be.
Richard:
Maybe just to bounce off some of those thoughts from Max. Absolutely. What we saw when we first started this initiative and one of the, I guess beacon what call a beacon statement or one of the visions was to say that we believe in human centricity. This is what we do here is complex air traffic control today is human-centric. The human is at the center of the decision making process. And exactly as you mentioned it, Vincent, it's really "CHarlie" is right now, it's that buddy over your shoulder giving you a hand helping you saying this is what you could do, this is what you can do. And if we look into the future, and I know we'll most likely talk about the future a little bit later in the podcast, but people talk about system centricity and there are other actors in the market that are actually starting with system centric design. And we simply believe that the steps to get these types of tools into operations is really to keep that human-centric approach. And not only because the buy-in is super important from the users is because we need the users to trust the systems that we're going to provide to them. But it also, when we talk about certification of these tools, we have to reassure a regulator that at the end that what we're proposing and the type of decision support that we're bringing to operations is robust. And I think that along with Max, that human centricity is that first step into doing that.
Vincent:
I know we had not "certification" planned when we prepare that interview, but I can't resist the question, especially now that you open that door Richard, and if you don't want to answer on that, fine with me. AI has that somehow bad reputation of giving really good results a high number percentage of the time, but in an unpredictable way be able to give really bad results and nobody can know ahead how this will happen and when this will happen. So what is your plan to go into certification in front of a CAA with such tools?
Richard:
Yeah, it's a really good question and it's super pertinent because anyone we talk to about our suite of decision support tools, this question always comes up and we don't have all the answers, but what we do know is that we start those discussions with the regulator early on so they know what we're doing, they know what we're talking about, they know what we're trying to do, we're trying to achieve. And in the end what we feel is going to be a key element is that for any certification procedure, we need to be able to explain what the system is generating as output. So without getting into super minor details here, what we've done is that we, we've adopted, let's call it the non-black box approach, whereas the models we're using to generate, let's call it generate some solutions, some ideas, call it whatever you like, but they're models where we can rationalize and explain what's been generated.
Not only we've generated a system that, let's call it a scoring system where the system will look at a bunch of different ideas being generated by the various machine learning models and then we'll prioritize those solutions based on objectives set out by the air navigation service provider for example. We will get those solutions that will score higher will be, let's say for those that are leaves aircraft up at cruising level longer and burned less fuel, for example. But in the end, any solution that's proposed will go through a set of deterministic tools that are already existing today. So if you start the process, what is generated, we can explain. So if the machine learning model generates a solution, we can explain how it generated that solution. The second point is that then we'll score them based on which ones are better for ATC based on a whole list of objectives. And then the third is that before anything is even proposed on the screen, let's call it on the glass to the air traffic controller, those proposals will have gone through the deterministic checks as they exist today to make sure that safety is all that is maintained, call it the safety nets.
Vincent:
Okay, I understand that, thank you for the clarification. Now speaking of products ATC is standard all across and all around the world, but we all know that every ANSP uses different systems. Is what you do very tight and very close to the Skysoft systems and can only be used with them? Or is it something that can be applied generically to every and any ATM system?
Richard:
It's also a very good question. When we started this out back in 2018, one of the pains that we were experiencing within an air navigation service provider, where I was responsible, was you've got these systems and you implement these new functions and then these new functionalities are so intertwined into the software code where you have a bug and then a bug generates another bug and then at the end there's so much testing required and software functions built over another function, another function. And what we said to ourselves was - we need to design something that's going to be agnostic - agnostic to software, agnostic to airspace. And so what we said is that we want these applications that can be integrated into any ATM system, something like an external module that subscribes messages from an ATM system, processes them and then publishes messages back to an ATM system, call it a microservice, call it an external module. Internally, even externally, when we're trying to explain this at a high level to people outside of Skyguide and Skysoft is we call it plug and play, of course it's not so easy.
You're not just going to plug in a module, it's going to work. But in the end, what we're looking for was to build a solution that could be easily adaptable to any ATM system. When I say easy, easily adaptable, very little adaptation. So in essence, we built, when we started this initiative, we constructed built we developed design, whatever word we want to use, a data pipeline. And a data pipeline was to say we have, we've developed methods and tools to extract a whole bunch of data to be able to transform this data into something usable that these machine learning models can work with. So in essence, we walk into a another ANSP with a totally different system than Skysofts. We need a bit of data. We can process that data and then publish system or messages back into that ATM system.
Now, right now today we're working with Skysoft in a pre-implementation phase. Of course we're using their system, their HMI, their screen, the look and feel, and we're adapting, we're adapting some of those HMI invisible solutions based on the Skyguide concept of HMI. But having said that, that's exportable to any HMI because in the end it's really what we're trying to do is the intelligence behind it saying we have a conflict, for example, in the case of CORA - system will provide a solution. And then you get with the local experts together and say, well, how do you want that solution displayed on your screen? And it really comes down to that. And that's really our mantra or our objective here is as we work and even towards pre-implementation at Skyguide, is to say, no, no, we don't want this intertwined into your code of the present HMI, but let's have this as an external module. And that comes with as challenges. It's new, it's a new way of doing things, but it's also something that we think is going to age well in the future if that can be said.
Vincent:
That's really interesting indeed, looking forward to see that spread out. And now to wrap up, and I know this will be a longer wrap up than usual, and we've been asking that question to every guest and we're at more than 50 episodes and I never looked forward so much as out of today because I want to hear you on how you see ATC evolving in 5 and 50 years. And I suggest, Richard, we start with you for 5 and then we hand over to Max for 50.
Richard:
So again, thank you. And I think I'm taking the easy question at 5, but I think that if you look, and it's not without speaking negatively because we see the situation as it is and as it is today, we see a need for more integration, more harmonization. And I'm not just talking with one ANSP, we're just talk, we're talking generally in even within ANPS within the network, if you talk about Europe, yeah, you talk about even globally, is that we need more integration, more harmonization, less fragmentation. The systems should be able to talk to each other and we'll talk to each other in a way that connect system connectivity. We shouldn't need to be making as many phone calls to adjacent sectors and controllers as we do today. And I think that ATC needs to embrace technology and grow with it and say that, but doing that is that you need to get your people on board and you need to, because of course, I guess the end users, whether be it an air traffic controller or an air traffic controller assistant or any type of supporting role within the operations rooms across Europe or anywhere in the globe, they see technology coming and they say, oh, well what's going to happen there?
We're going to need maybe less people or that's going to transform the way I work or transform my job, or are they still going to need me? And I think that if you look over let's say at least a 5 year period, I think that there's a long way to go in a sense that we need to harmonize the way we work. We need to get that technology in place, which we see it's hugely challenging from procurement to implementation. We've seen use cases across Europe where system procurement is taken from procurement to implementation, has taken over 10 years to get a system online. So we know that it's very long, but I think we need to be more efficient. So within 5 years, let's say, let's learn from our mistakes in the past, our lessons learned, if that makes sense. And to say how can we do things a little bit differently to take steps together, whether that be ANSPs or a couple ANSPs.
We have FABEC for example, even within the FABEC, if those ANSPs can get together and say, well, let's take some concrete steps together and get our technology being more interconnected. I think those are first steps into the future. And then providing the much needed capacity. We're coming off COVID now. I think that during the worst days of COVID, I think people there, there's some people out there that would said the traffic's never going to come back. And then we saw very quickly that recovery was a lot steeper than anybody would've anticipated. And traffic is growing. People want to travel. So within the next 5 years, we need to build the capacity into the system. And I think that people are our biggest asset, but it's our people. They're going to help us bring that technology in to help us build that extra capacity.
Vincent:
It's very interesting to see how you still see the people being in the center of everything for the next 5 years. So Max, now over to you. Do you also see people at the center in 50 years or do you think we'll all be retired and nobody else will do ATC and it'll all be automated?
Max:
Well, let me look at the crystal ball and see what we see. But it's a question I'm going to take in a way. Well, 50 years is less than twice as much as well, Richard and myself have been in ATC now and before going into the next 50 years or in 50 years time. The reflection is that if I reflect conceptually, it doesn't seem that we are way different the way we do things now, then we were 28, 30 years ago, at least in certain parts of Europe. However, if we think about the complexities that have happened in the last 30 years, we managed to keep abreast with these. So yes, we have advanced a lot in 30 years, even if at the first thought things seem to be the same, and we are encountering problems like Richard said, but we are probably also controlling a hundred percent traffic increase over what we were doing 30 years ago.
So we need to acknowledge that. And when now looking at the future from one side, I always introduce ATC to my students as I'm also an instructor, as a very pragmatic sector. We do things with purpose when there is a need for them and we feel the need. We may anticipate that need by a few years, but we do not invest into creating a situation where the need doesn't exist. So I believe that that trend will continue in terms of that we will continue advancing in terms of technology deployment. We're going to talk about what I think about humans or not in a minute. So we'll start seeing action implementation by systems. We will start seeing an integration about in terms of that. But that's because the complexity, U space, different types of aircraft, the volumes of traffic will probably continue to increase, especially if the whole sector, and this includes the flying part, finds a solution to the environmental issue, which is another issue, which needs to be probably tackled first or equally.
Now in 50 years time, I am not sure that a world in which we are all on vacation without working as humans is utopic. I believe that that's more dystopia, that the world is a human world. And in that human world, we have a need actually to be self-fulfilled. And that self-fulfillment, talking about Maslow and all these motivational in big part comes from work. So I believe that the humans need to be part of the system cause it's a human world. And in this human world, yes, we need to make things more efficiently. Yes, if we can do more with less or with equivalent, we should do that. But we shouldn't simply go into a place where we are only two risks in our own world. So from societal, it wouldn't be the world I would want to see in 50 years time. If I were the one who would write the concept of operations, I would think of a centralized ground-based system.
Again, highly automated with the human in the center, but really being supported by a lot of technology. So I do not necessarily see that the airborne part should be confided with decomplexifying their needs. They have their own business requirements, be it U space, be it commercial flying, etc. They're there to fly, to take care of passengers of their cargo, of whatever the reason they're to transport. And there needs to be a system which allows them to do this in the most safe and in the most efficient manner. I see that that is a ground-based way. I do believe that we will shift from tactical ATC into supervisory and management mode with possibly less people executing because of thing. Now, assuming that complexity is growing within reason. But I still see, Vincent, you're right, the human in the center of this, because I see in general that our way of evolving as a society in 50 years time could still have us finding purpose. And if we want to travel from here, Madrid to Geneva to do work, we want to be participant in both the sense that we want to be the tourist who goes elsewhere, but we want to be part of the system which makes this possibility there for us, ourselves or for the service. So yes, more intelligent, yes ground based, yes - highly automated, but in a way that resembles us. It's a human system that I see in 50 years time.
Vincent:
Richard, I see you want to complement on that? Go ahead.
Richard:
And just to complete that a little bit is if we look at both Max and I, who back then we didn't knew each other, but we started in our careers around the same time in the mid 90's. The job looked the way it looked back then, and we learned the things we learned to do the job. And if you look at the evolution over 25, 26 years now, or a little bit more even is, you see the job has changed. It's not the same job today as when I started and when Max started in '95, '96. It's not the same work. And I think that as we project ourselves into 50 years, as Max said, the humanist still has a big part to play in there, but the job is not going to be the, it's not look the same.
We talk about sector concepts. Well maybe there's no more sector concepts. Maybe the intelligence of the systems is as such that you don't need to break these airspaces down into as many sectors. Maybe it's just controlling individual flights within a larger block of airspace. And there's many, many projects that have started discussing that. So there's those such of things. So the job will just look different, I think. And then that brings another aspect that I really hope that air navigation service providers are taking into account is that the future generation of air traffic controllers and demographic that you're looking for, you need to be, I think start looking for a different maybe profile of candidate to recruit to do that job into the future. Maybe the skillset that's required is going to change as you move forward. So I would invite the air navigation service providers maybe to start looking at that aspect and saying, well, the job is evolving, technology is evolving. Well, is the profile of the air traffic controller evolving into the future? And what are we needing? What type of skills do we need? So those are all questions that we certainly don't all have the answers to. If we all had crystal balls though, life would be a lot easier. But those are, anyways, some of our thoughts on that topic.
Vincent:
Yeah, very true. Thank you for taking your crystal balls out today and have a look with me. It was really nice. I look forward to see more of your results. We will put all contact details in the episode notes so the audience can get in touch with you, with us if required. Richard, Max, thank you very much. It was a pleasure to review.
Max:
Bye
Richard:
Thank you Vincent.