
Every episode, in your inbox
One email per Travel Tech Podcast episode: the conversation, the key takeaways and the use cases worth building in aviation and travel AI.
Beyond Line of Sight: The Infrastructure Drones Need to Fly
Why beyond visual line of sight (BVLOS) drone operations are constrained by surveillance infrastructure and detection physics — not aircraft capability.
Show Notes
James Dunthorne brings expertise spanning collision avoidance research, airspace safety, sensor fusion, and complex projects including parliamentary restoration work. With a PhD in aeronautical engineering, James explains why BVLOS is fundamentally an infrastructure constraint — covering visual airspace safety limitations, low-altitude transponder fragmentation issues, edge networks for safety-critical systems, and the role physical infrastructure plays in enabling AI applications.
Read the takeaways
Beyond Line of Sight: The Infrastructure Drones Need to Fly
In this episode, I sit down with James Dunthorne, founder of Neuron Innovations and 4DSKY, and an all-round expert in drones, data, and aviation technology.
Read the newsletter issueFull TranscriptClick to expand (7,948 words approx.)
Alex Brooker: Alex Brooker here. Welcome to the second episode of the Traveltech podcast. And I have a treat for you. I just spent over an hour going really deep with James Dunthorne on all things drones, data, sensor networks, aviation, and AI. We talk about his PhD at university. We talk about flying a drone around the Houses of Parliament, over farmers fields, and up and down railway tracks and everything in between.
So, I really hope you enjoy this episode and it's sponsored by Airside Labs, airsidelabs.com. So, James, good to see you.
James Dunthorne: Yeah, good to see you, Alex. So looking forward to this show and what kind of interesting conversations we might come up with. So yeah, really good to see you and let's talk all things aviation.
Alex Brooker: Absolutely. Yeah, that sounds great. I'm really looking forward to this because I know you've got a really interesting background and story to tell. I think people will be really interested in. So, what I was thinking, if you don't mind, perhaps start at the very beginning. I know you spent some time at university and maybe that's where the passion for aviation began, but I was wondering if you maybe start us from the very beginning.
James Dunthorne: Yeah, sure. So I went to Loughborough University and studied aerospace or aeronautical engineering as they call it. It's actually part of the aeronautical and automotive engineering department because basically a plane is just an upside down car when you get up to sort of high performance vehicles. So it was really cool there. Really enjoyed it. You basically just gone upside down wing.
I think the Formula 1 guys and the people building jet fighters are kind of similar, very similar skill sets actually. But yeah, I was there and actually worked in the Loughborough University Center for Autonomous Systems. So they have like a research lab at the university where you build drones, develop the algorithms, fly them around a room with like these infrared cameras in them and develop these new control systems for landing on the back of a moving vehicle or trying to get them to autonomously land and things like this. So I really got into the drone space back in 2012. I think it was around about 2011, 2012. So yeah, over 15 years ago now almost.
Within that period went on to do some research. So postgrad research looking at collision avoidance systems for drones. Specifically looking at sort of when you've got two aircraft flying towards each other, at what point are they no longer able to avoid each other anymore? So obviously we get into collision scenarios but actually modeling when a system can no longer actually produce a result is really interesting and important from a regulatory perspective and from a safety perspective. So that was kind of where I started off.
Then I went into the agricultural sector because apparently drones were going to be massive in agriculture. So we thought, okay, we'll do our own startup. We'll build this agricultural drone. We're launching it off these catapult things and fly over crop fields. We sort of had these cameras, these infrared cameras that we used to fit to them and they'd be able to pick up sort of things like crop health. You can do things like NDVI, which is a technique you use to basically compare visible light and infrared light to be able to detect crop health.
Alex Brooker: What does that stand for?
James Dunthorne: Normalized Difference Vegetation Index or something like that. I'll have to Google but it's basically - that's close enough. Basically, it works off the principle that a plant, say some wheat or barley, it absorbs visible light but reflects infrared light because it's too energetic for the plant to be able to use for photosynthesis. So a plant that's healthy will absorb a lot of natural light or visible light but reflect a lot of infrared light. But when it starts dying or its health starts reducing, it starts reflecting more infrared light compared to visible. So you can see by looking at that ratio between visible and infrared you can basically pick up the plant health.
So that was going really well like we were making some really good progress but the two problems we faced there was well how can you fly this thing long distance because the regulator said well 500 m is your limit and some fields could be sort of 3, 4 km long, so much bigger than that right?
Alex Brooker: Yeah, so you don't end up having to do multiple takeoff points and every time you're taking off and landing that's actually probably about 50% of the work is doing the takeoffs and landings. So the more you can keep the aircraft in the air in one flight and cover a large area, the more efficient it becomes and the price then drops down massively.
James Dunthorne: So that was kind of the major problem we found there. Along with the uptake of farmers was quite slow. They're not the fastest of people to sort of onboard and uptake drone technology. But as I say, this was back in 2013, so drones had only been around a few years then. The regulatory environment was very early. It was also, people were a little bit suspicious of these drones. Think they're going to be spying on you or snooping on your activities. So, we're in a very different environment back then.
So I moved into an area called GIS or geographic information systems. You may be familiar with software or companies like ESRI, for example, who make a lot of this software and what it does is it's basically a way of - think maps. It's basically maps but with vector drawings and things drawn on top and a lot of the work we were doing was around flood management, being able to measure the undulation of the land. So high points, low points, if it starts flooding which areas would be fine which areas wouldn't be.
We did a lot of surveying work there. So land surveying, building surveying and 3D modeling the environment really using that. And one of the clients when I was working there was a company called Plow and Craven. I sort of ended up speaking to the managing director at Plow and Craven. They're big huge surveying company in the UK, over 100 people working for them and they did projects like - I don't know if you know the restoration works on the House of Parliament and also did like Grenfell Tower, Tottenham Hotspur football club stadium. So all of these big projects. Yeah, they did all the surveying work. So measurement of that and I sort of said, why are you guys not using drones yet? This seems like a really obvious solution for you.
They also had a rail department as well and I sort of said to them like, I think we could replace what you're doing on the railways with drone technology. The problem was they needed like super high accuracy. It's like 5 mm accuracy on the data which is really hard to do from an aerial moving robot. But we managed to come up with a solution. We went for these medium format high resolution 100 megapixel cameras and built the system around that and we scaled that department up to about 25 people. I think we did several million in revenue during that period as well. Which went really really well and we got a lot of operational experience. I was sort of managing and guiding those operations and the drone technology that we were deploying out in the field.
And then thought, well, I should probably do my own thing now. I've worked for other people and made them successful, but I'd love to do my own thing. And the big thorny problem back then was BVLOS drones. And for those not familiar, BVLOS stands for beyond visual line of sight. So it comes full circle back to my agricultural days where it was like, well, how do you fly further than a person can see? And the regulator said 500 meter limit you can't go beyond that.
So we were trying to come up with a solution around okay how can we enable things like drone delivery, long range inspections or search and rescue missions and things like this. So we built Neuron. Neuron is sort of - we were deploying these ground-based sensors to detect air traffic and feed that directly into the drone systems that were flying. And the idea of this is that the reason why the regulator won't let you fly beyond line of sight of the pilot is because if an aircraft comes along and it's on a trajectory with your drone, they've got no way to see each other and so there's a risk of collision. So we thought well if we can provide the drone pilot with visibility of where all the aircraft are then we can mitigate that risk, overcome the regulations.
At that time I was working really closely with the regulator writing these regulatory safety cases sort of on the side as a consultant as I was developing this business.
Alex Brooker: Okay.
James Dunthorne: So we had a lot of drone operators coming to me saying hey I want to do this but it's not within the rules. How do I apply for permission to do more than that? How do I fly a drone light show, a swarm of drones doing light shows? How do I fly this beyond line of sight in this specific area? And so I was literally working through the paperwork and the regulatory process with those operators. And so I was like, there's definitely an opportunity here to provide a surveillance service into these drone operators and make that data available to basically flying robots.
And since then we've been building - so that's kind of like the backstory really. But going from sort of academia through to operations and the technology behind that, the data around that and also then the regulatory process as well. It's been kind of very broad but very interesting and fun at the same time. Managed to go and get to fly around Big Ben which was quite cool. So I think I was to 3D model Big Ben with a drone. And that ended up being used by the surveying company to create the CAD drawings for the House of Parliament. So that was really really good fun.
And yeah done loads of really fun things, the palace, Buckingham Palace and at Grenfell Tower when they had the big fire we went out there and were inspecting the building afterwards. So some quite interesting high-profile projects that I was involved with directly and planned and flew and sort of orchestrated really. So it's been a good few years.
Alex Brooker: So how - I mean it sounds fascinating the sort of sheer level of flying and operations that you've led and coordinated. Dare I say it before it was cool, sort of well before UK initiatives like future of flight happened around 2020. So I mean I'm really interested in your journey from doing the flying to recognizing the gap perhaps to enable that beyond visual line of sight or different kinds of flying. And it sounds like you even started a company to help get it done overall not just for yourself but for the whole industry. So could you perhaps talk to me how did you arrive at that epiphany to create neuron and work on sensors in that way?
James Dunthorne: Yeah. So I think there were a few moments. So obviously right the way back to university days when I was researching collision avoidance systems for drones, it was obviously an up and coming area of research that was going to be of interest. And the key was how do we prove the safety of these systems? How do we build systems in a robust scalable way where you don't have single points of failure? If someone goes online to watch their grandmother flying into Heathrow airport it's not going to take down your drone delivery system. So how do we come up with this robust system and how do we make sure that drones and aircraft can stay safely separated.
So that was kind of where I was exploring the problem space. But then I had real world encounters of that, of the problems of not having these systems which was in agriculture, having only being able to fly 500 meters was a huge pain in the backside. And the aircraft was capable of flying kilometers, and I could see all of the airspace around the drone and there was all these problems that I was like, there must be an easy solution to this.
And then also when we're doing the network rail operations, when you're flying up and down railways, you're having to do them in short segments and then go to the next segment, do that one, and then go to the next segment and do that one. So, it was highly inefficient. And I knew people who were working in this domain. It wasn't just my own personal experiences but customers that I was working with on the regulatory side of things when I was consulting. I had people coming to me saying look I want to fly BVLOS in this area, can you help me do it?
And the way of doing it back then was well let's close down the airspace and therefore if we close down the airspace to other airspace users then you can fly your drone in there. But that would take like three months to go and get the approvals to do it, you'd have about 3 months worth of operation when the airspace was closed and then it would be back to normal again. And so you can't really build a scalable business model for all of these drone delivery, linear infrastructure inspection, search and rescue missions that all of these different customers I was working with wanted to do. So it was like...
Alex Brooker: Sorry. So linear inspection is like the railway use case?
James Dunthorne: Along a line inspecting or power line.
Alex Brooker: Yeah.
James Dunthorne: Search and rescue is roads.
Alex Brooker: Yeah.
James Dunthorne: Search and rescue is like okay emergency situation. Someone's lost on the side of Mount Snowdon. Where are they?
Alex Brooker: Yeah. Let's cover - we can cover a bigger area much quicker than we otherwise could. So there's like a core we can do it quicker and more efficiently if we use this technology but then it comes up against a quite slow perhaps understandably so in terms of the safety case but a big procedure to shut down the airspace to actually do the trials or whatever it is, do the flying.
James Dunthorne: Exactly. It was basically the only solution was what you would call a trial environment solution. Where you could trial something out for a short period of time. But then it was like right okay well I've proven it all works now, now what? Can I operate now? The airspace is now being opened back up again. So what we need to be able to do is how do we solve this problem in open airspace? If we ensure there's no traffic in the area then obviously I can fly safely. As soon as there is mixed traffic in the area then how do I fly safely?
And it's pretty obvious well you need to be able to see it. How do I see traffic when I'm on the road? I use my eyes to see it but the problem with the drone is it doesn't have eyes as such and the closing velocity of aircraft is quite high so they - an aircraft looks very very tiny on the horizon until the last few seconds when it hits you. So actually even your eyesight - there was a study done by some airlines I believe a while ago which sort of said that really visual line of sight, a pilot's visual mitigation is almost meaningless. There's not a huge amount of benefit in doing so and it's not actually that effective. I can't remember the exact number but I think it was less than 50% effective. So even pilots were struggling to be able to see with their own eyes.
Could you make a machine to be able to do that? Have a drone with a camera on it? Well that's really hard to use cameras to detect aircraft. Like it's possible on the bigger drones I would say probably, but on the smaller drones certainly not. Not at that kind of distance anyway. Short range obstacles, you see these ones that fly through trees and avoid obstacles on the ground, it's a much different problem because you've got a lot slower closing velocity. But where you've got an aircraft maybe doing 100, 150 mph towards you, it's closing at a very rapid rate and so it only gets big at the last few seconds at which point based on the research I've been doing, it's generally too late by that stage. So you need to be able to see it when it's small on the horizon.
There's a few ways of doing that. One of which is using radars. These are generally really expensive to do. They basically emit a pulse out into the environment and look for reflections of that pulse. The problem with that is they're incredibly expensive. They weigh quite a lot in terms of their mass. And ultimately use quite a lot of power as well because they're emitting a lot of energy. So sticking this on a small drone and using up your available payload with the sensor that keeps everything sort of happy and sweet isn't really very useful.
So what other solutions were there? Well, you've got electro-optical, so video cameras and things like this. But as we said, actually one of the pieces of research I did when I was back at the Loughborough University Center for Autonomous Systems was looking at pixels and sort of saying, well, if I've got a video camera, how soon would I be able to detect an aircraft? And my conclusion was that with a typical high-definition camera, it doesn't even occupy a pixel until it gets very very close to you. So simply being able to detect the aircraft in the first place would be incredibly difficult. You would need to have like lots of very high resolution cameras around you with a long zoom lens on them to be able to actually pick up these aircraft in enough detail to first of all not just know that it's there but actually classify it. Is it a plane? Is it a bird? Is it a balloon? Is it a cloud that I'm seeing? Is it a speck of dust on my camera lens? There's a lot of variables there that makes it really hard actually to pick them out. So electro-optical wasn't really a complete solution. It probably has some space, it's certainly being used for obstacle detection on the ground but not for aerial to aerial threats.
So the other obvious one which sort of stands out is what they call electronic conspicuity. And this is the principle that every aircraft has this transponder on board, the size of a pack of playing cards, I would say. And a little antenna and you stick it on board your aircraft and it basically transmits its GPS location. In most cases it will have a GPS attached to it and it will ping its registration number, aircraft registration number and its latitude, longitude and altitude. And then other receivers can then pick that up.
Now that's great. Problem solved. If only that... But the unfortunate reality was that there are about four or five different types of transponder that exist out there in the market.
Alex Brooker: Okay. There's a range.
James Dunthorne: There's a range of them and they're not interoperable. They don't speak the same language. They're on different broadcast frequencies. And so you're not interested in seeing 20% of the aircraft in the sky. You want to see 100% or close to 100%. But how do you do that in an environment where it has mixed transponders? You also have the issue that not all aircraft are carrying these transponders. There hasn't been a mandate for these transponders yet. Although there is talks of this happening very very soon by the aviation regulators. So that's the key problem there is like well how do you have a sensor that can bring all these things together? And also how do you spot those legacy transponders?
So people have probably heard of something called ADS-B. It's quite a common transponder type and it's on your commercial aircraft. So like Boeing 737 and your Airbuses will all carry an ADS-B transponder. But you have had something before that prior to ADS-B coming out. It called Mode S. You actually had mode C and mode alpha I think before that. But they're very legacy those ones. But Mode S transponders, they tell you the call sign or the registration of the aircraft, but they don't have a GPS coordinate. So, you only know there is something out there with this call sign within range of you because you're obviously receiving it. But you don't actually know where it is.
So how do you detect those ones? Like that's they're the hard ones to detect because yes, there is a way to detect them, but it's not straightforward. And the process you use for doing that is called multilateration. And for those who aren't familiar with that, you can think of it like triangulation. Basically what - if you have say four or five sensors on the ground, what you can do is you can measure the time at which that signal was received from the aircraft from each of these different sensors and by having the differential between those times you can basically pinpoint the position of the aircraft by triangulation. And that's what multilateration is in simple terms.
But to do that the processing of these sensors has to be really really accurate on timing. You're talking sort of in the nanoseconds of accuracy. The time stamp needs to be because for every - I think for every nanosecond light travels, RF signals travel about a foot. So 100 nanoseconds is 100 feet of accuracy is the way to think of it in a rule of thumb. So how do we actually go and pinpoint - how do we get that kind of accuracy? Well the sensors need to be time synchronized with satellites or with an atomic clock of some description to be able to correct that time stamp and to be able to do it on every data packet. So it was really complicated.
Anyway, but what we did is, we found a sensor that enabled us to detect all of these different types of transponders. So, we partnered with a company called JetVision. Sort of a leading company in the flight tracking space. And they made a sensor that detected all of these different types of aircraft. And it lives on the ground. You can't really do this from the air because if you need four or five sensors to be able to triangulate the position, those sensors, you also need to know the location of those sensors. So, if you've only got this sensor in your own vehicle, you can't triangulate the position because you've got one sensor. So, you need multiple sensors, like an array of sensors on the ground to be able to triangulate these positions. And that's why we came to the solution which was that we needed some kind of ground-based sensor network to be able to pick up these aircraft and some really novel way doing that.
So, before we've gone deep which is awesome or high altitude depending on your perspective. But before we talk about the data, the reason we need sensors on the ground is because of the variety of electronic conspicuity methods that are out there and some air users don't have any at all. The balloons - would somebody in a gyrocopter or a balloon do they carry them?
Alex Brooker: Sometimes?
James Dunthorne: Sometimes, but not always. Sometimes. And the thing is as well, if you're putting these things on board an aircraft, you're just adding weight to them. And if you're having to detect four or five different types of transponder, you've got four or five different chips you need to have, maybe two or three different antennas you need to have. They may be interfering with your antennas that you have on the aircraft as well. If you're blasting out 868 megahertz and you're using 868 MHz for your communications link, then you've got conflict there. So, you don't really want to be putting all of this equipment on board your drone, which is meant to be lightweight. Lightweight things tend to fly better than heavy things. So, you want to be avoiding putting too much equipment on your aircraft just for this purpose. Can't triangulate if you've only got one sensor anyway. So...
Alex Brooker: Exactly.
James Dunthorne: Yeah, it's...
Alex Brooker: So you can't - excuse me, you can't have all the sensors on there for the electronic conspicuity. You can't use radar. Too many cameras. Again, too heavy. And there's the physics of the resolution of by the time you can see it...
James Dunthorne: Yeah. You're in - there's an issue.
Alex Brooker: Yeah. And then obviously conventional aviation right now uses air traffic controllers. They're the ones who give directions to pilots to change levels or do whatever. The pilots are not actually looking out this looking forward working out and then communicating directly with each other. It is via a controlled service. Right. So the way you've described where we seem to have ended up with drones avoiding each other is similar in that sense in that having actually a separate controlling or at least informational service about where everybody is...
James Dunthorne: Exactly.
Alex Brooker: Is actually the same way that we do it now with conventional flying. Right.
James Dunthorne: Exactly. Yeah. So commercial aircraft are all being tracked using, well, at an airport, they'll have an expensive radar, which can be in the tens of millions of dollars just to put one of these radars in at an airport, and they will generally have a primary radar and a secondary radar. And also generally ADS-B equipment at the airport as well to pick up these electronic conspicuous aircraft.
Now, commercial airliners are required to carry transponders. They're required to carry ADS-B transponders on board and they fly in sort of semi-segregated airspace I would say. They fly in class A and B airspace which is separated from general aviation users. So you would have class G is what most general aviation fly in. It's low down generally away from airports and places like that. As soon as you start getting up higher, when you're flying at 35,000 ft on your way on your holidays, you'll be in a class A airspace at that stage. And so when you're in those kinds of airspaces, then you have requirements to carry equipment on board your aircraft. And so all of these aircraft are conspicuous to each other and they can stay safely separated because of that.
In the lower airspace, you don't have that requirement to be conspicuous. And you've got all of these different transponder types. And so how do you pick up all of these different aircraft with different transponders? And how do you account for future transponder types as well? It's not just what do we have today? Like in Europe, EASA, they basically are now introducing something called ADSL. Now it's very similar to FLARM, which is another one of these transponder types. It operates on 868 MHz. And sorry...
Alex Brooker: Is that UK only? FLARM is that a slightly older technology?
James Dunthorne: So FLARM has been around for over a decade. It's used primarily on gliders. So these sort of long gliders with long white wings if you're familiar with them. But they also are used on general aviation aircraft. So little Cessna, people who hobby fly and things as well. But they tend to fly quite low down. So if you're flying a drone actually these kinds of aircraft are actually quite common within that airspace. You don't tend to see a 737 at 3 or 400 ft. Not very often unless it's taking off or landing. So...
Alex Brooker: Not very close up anyway.
James Dunthorne: Yeah. Yeah, exactly. So, you want to be able to see all of the aircraft that within your airspace and within that lower airspace volume, it's very mixed. There's different types of transponders and not everything is still transponding, but it's not always necessarily a requirement for everything to be transponding. As long as you can meet the required safety levels, then that's generally acceptable.
So, let's talk about some of the data involved because I think you've articulated the problem. It's knowing where everyone is and then communicating that to everyone. So, how do you deal with the data? How does it work between the sensors? And in a sort of more general sense, how on earth does this stuff coordinate together?
James Dunthorne: Yeah. So, we started off basically deploying some sensors around an area, a test area, where there's some drone activities and aviation activity. And we thought, well, let's just pull this into a data center cloud server and we'll show it on a screen and people can then buy access to this screen and use it to distribute their flights or to operate from. And the problem with that though is that when it fails, everyone loses visibility instantaneously. So across the board, every single customer who's plugged into that cloud system is now losing that. The other problem you get is say if a customer from America wants to have access to this data. Well the sensor data from America is being transported across the ocean to our server in the UK and then the customer's then getting the data from our server in the UK and taking all the way back to the US. So you've got these long hops of data going backwards and forwards.
So what was the solution to that? Well, okay, we can stick a server in every country. And therefore it keeps the latency really low. It keeps them highly performant and keeps the data pinging backwards and forwards at realtime speeds. And that also solves some of the redundancy problem that everyone loses visibility at the same time. The problem comes is these servers aren't for free. You have to pay for them. So firstly you're not getting much redundancy. If my UK server goes down, then everyone in the UK loses visibility of what's happening. You also end up with the situation where you end up with a much higher cost to these systems because you're having to churn all of this data and these sensors can generate about 2 GB of data per day each. So when you're talking about thousands of sensors, you're talking terabytes worth of data processing. And not only that but you also need to do some other processing to it. So you might need to do multilateration processing which is quite intensive as well.
So you have all these problems around scale of firstly how do you make a network that's robust and it's not - it fails locally not globally. How do you keep latencies low so that you're getting real-time information to the pilot as and when they need it. A second or two can make a big difference if you're in a collision scenario. So you need to be able to see things in as real time as possible. And you also have this problem around well, okay, if an operator needs to assure their operation and assure the systems that they're using. Well, they would have to come to us and we would have to tell them how our system works, how does our algorithm work and go through that whole process of how all of this fits together. So there's a lack of transparency around that. A lot of these systems would basically be like a black box. If you go to flight radar 24 or one of these other flight tracking networks...
Alex Brooker: Other networks are available.
James Dunthorne: Yes, exactly. There's - and you end up with all of these networks competing against each other. So how do you actually - the problem space here isn't about how do we create a network because there are networks out there already. The problem is that those networks there's a lack of transparency, performance and robustness to these systems and regulators are not happy about us using fragile systems that have lots of latency and there's actually requirements in a lot of the regulation that require it to be less than two seconds worth of latency. We need to prove its availability and prove its performance and robustness and all of these kinds of factors. And especially when you start getting up to national or international scale, those questions are just going to get harder and harder and harder, which is how, as soon as there's an incident and everyone loses visibility and you've got drones falling out the sky, because they've all lost surveillance capability. It's the equivalent of having a centralized network for all of the lights on the front of everyone's car. Imagine if everyone's driving down the motorway and then suddenly the server goes down and everyone's lights go off on the front of their car and in the dark. It's going to cause absolute chaos.
So what you need is a system where okay well one person's lights can fail and sometimes it happens and sometimes somebody loses visibility but you don't want everyone to lose visibility at the same time because it creates a big problem.
So just giving that context was how do we stop the server from going down? How do we make it more robust? There's a few ways. We can create lots of servers and put them everywhere, but we've already discovered that that's quite expensive. It's also quite architecturally complicated. But I love the comment which is the best part is no part. So the best way to stop a server from failing is to remove the server from the network. If you've got a drone pilot or an end user who's consuming some data, then the quickest way you can - the quickest and safest way to go and get that data from the sensor into their platform is to connect them directly to each other with essentially like a long cable but like an internet cable.
And so how do we orchestrate a network of sensors so that any end-user application can access those sensors directly? And that was basically the conclusion we came to which was how do we create a network where sensors and applications are on the same network and can orchestrate connections between those different nodes. And the common way of doing that is what they call an edge network. So an edge network is that very thing is every node in the network is connected to every other node in the network and they can communicate with each other, transact, stream data between each other.
Alex Brooker: Okay.
James Dunthorne: And the thing with that is if a node gets eliminated well it doesn't affect any of the other nodes, only affects the nodes that it's directly servicing. It doesn't affect the network as a whole. Whereas if you have this sort of hub and spoke model where you've got a centralized server in the middle and all of the data coming in and then all of the data coming back out of it. You can imagine if you eliminate that then you've lost everything.
So we went for that edge network design and that's caused - that created a lot of other engineering problems that we had to solve. We went through that process. We basically created - we can think of it like a telephone directory. And that telephone directory is the way that all of these nodes in the network find each other, connect to each other, and then stream data with one another as well.
So, if I could just take a step back, I'm trying to keep up with your explanation because it sounds so fascinating and detailed. And I love the example of car headlights. So if literally if everybody has the same subscription to headlights and suddenly the server that's managing that goes down then everyone's headlight switch off is I think a really good analogy for the risks of like super centralization of that kind of service. And so what we're talking about when you say direct connection and one node being removed and the network can adapt. So that's similar to how the internet's designed now. So if one node goes down, you can still watch Netflix might be slightly slower, might go via a different - as long as your connection's still there and you get to the backbone, you can still get to those other websites, those other servers over the internet, right? So the network adapts.
James Dunthorne: Exactly.
Alex Brooker: You still get a connection. So what you're talking about is linking the consumer of the data directly with the sensor that they need.
James Dunthorne: Exactly.
Alex Brooker: Yeah. And that's a really good example because then the internet doesn't go down across the world. Your internet service provider may go down and you might lose internet in your building or even in your street or something like this. But failures are localized there. Just because you lose the internet doesn't mean that someone on the other side of the country loses their internet as well. And so the internet is in itself like an edge network. It's connecting people in a point-to-point manner. And it has multiple routes through that network as well. From a geography perspective, you have all of these sort of points within the internet that basically reroute data to the next point and to the next point. And if one point goes down, it just goes round them and goes through another route to the end user. And that's essentially similar kind of analogy to what we're doing is removing those sort of single points of failure within the network and making this more of a more robust, more highly available solution.
So, why would being hyper local at the edge be a good idea for drones operating at speed close to each other? I think actually maybe it just
Alex Brooker: I've obviously known you since you were back at Snowflake and then went on to Cyrian and doing some really cool work there, and you know obviously did quite well through that process. Yeah, I'm excited to see what your sort of next big thing is going to be because I see you doing a lot of stuff with AI and testing of AI and things like this. You got your fingers in a few little areas at the moment so it's going to be exciting to see that progress and how that sort of works out.
Are you kind of feeling your way along that journey at the moment or do you have any kind of plan yet, or are you just seeing which way the wind blows?
James Dunthorne: You're turning the tables on me, Alex. Is this not supposed to ask me questions? No, it's all good. It's all good. Yeah, no, I'm a solopreneur and learning AI, building AI, building AI systems and chatbots and working with awesome startups and yeah, doing lots of cool things. Let's see how far I can go as a solopreneur and not have any humans other than myself, although I've learned that context switching is the big – that's the real bottleneck there.
Alex Brooker: It's hard, isn't it? The brain context switching between – there's only so many instances of Claude code I can have in terminals across my display before I literally can't cope.
James Dunthorne: You got your workbench with about 40 machines all coding simultaneously for you?
Alex Brooker: No, that would be insane, but there's some really cool automation stuff that's coming out soon, definitely.
James Dunthorne: Yeah, no, thanks again, Alex. All good. Best of luck with the podcast and yeah, look forward to seeing it going live. And if anyone wants to get in touch, you know, obviously they can get in contact at neuron.world. There's links on there to contact us if you want to send a message and find out a bit more then check it out.
And if you want to track some planes in 3D, then 4dsky.com is the one. There's a – you can get in there, click on the map at the top right and you'll be able to see like a really futuristic view of the airspace which is kind of cool. And there's some new releases coming out in the next few weeks, so that's going to be really cool and going to be some exciting new features in there.
So none of this boring 2D map display, but we're going 3D. And as the name suggests, eventually 4D as well with time, so be able to rewind. We haven't worked out whether you can fast forward yet, but you can certainly rewind. But yeah, it's going to be really fun and excited what the next 12 months will bring, but yeah, thanks very much for inviting me on, Alex. Good to see you again as always and hopefully have some more exciting updates for you in the future.
Alex Brooker: Love it. Thank you so much. And all the best for this year. I know you and the team are going to smash it. So good stuff, James.
James Dunthorne: Nice one.
Ep 1: Jevons Paradox for Knowledge Work
Ep 3: This Ex-Pilot Is Building AI for the Cockpit
Related episodes
Building AI for aviation or travel?
Airside Labs builds working AI prototypes on your data and tests them against the edge cases of your operation. Our agent tools give Claude, ChatGPT or your own agent a catalogue of 6,700+ aviation AI use cases and aviation lookups it can trust.
About the Travel Tech Podcast
Hosted by Alex Brooker, founder of Airside Labs, the Travel Tech Podcast features authentic long-form conversations with leaders, innovators, and startup founders shaping the future of aviation and travel technology. No scripts, no sponsors, no sales pitches — just real conversations about building technology that matters.
Have a story from travel or aviation tech? Come on as a guest.
