The Software Agents - Episode 5: Dawn Aerospace - Stefan Powell, Koen Lucassen - Software in Space

A new remotely-piloted space plane will take off and land at regular airports, to carry global data sensors and experiments and eventually place satellites into orbit. Dawn Aerospace Founder Stefan Powell and Control Systems Engineer Koen Lucassen explain how CI/CD software techniques let them develop advanced flight systems, and refine space rockets into reliable everyday freight vehicles to monitor climate change and boost our post-pandemic communications needs.

Announcer: Welcome to the Software Agents, a podcast series that brings you the people using software to help change the world in this time of transformation. With the help of some of the brightest technology minds, we'll explore how almost every possible area of life, society, and business is being reimaged through the power of software. 

Christina Noren: Welcome to Software Agents, a new podcast on how software is helping the world survive and evolve right now as told by the people making it happen. I'm Christina Noren and my cohost is Paul Boutin.

Paul Boutin: Hello.

Christina Noren: The Software Agents is sponsored by CloudBees, the enterprise software delivery company. CloudBees provides the industry's leading DevOps technology platform that enables developers to focus on what they do best, build stuff that matters. Today's kind of a special episode because we have two guests, so it's our first one with two guests.

And our first guest is Stefan Powell who, full disclosure, happens to be one of my favorite first – of my many first cousin's sons and also a true rocket scientist trained at TU Delft. He is the CTO and cofounder of Dawn Aerospace, which has received a lot of attention, including from the software world with an investment by Erik Swan, my former colleague and the CTO and cofounder of Swan.

But Stefan is not a software geek so he's not the software agent here, and we have his colleague who is the software agent at Dawn Aerospace, Koen Lucassen, who's our real guest here because he's the one who's writing software that makes it happen. And I'll sketch a little summary of why we think Dawn is interesting. 

So Dawn Aerospace has the first space plane that flies faster than any other plane in the world and can take off and land, thanks to the software Koen and his colleagues are developing, can take off and land multiple times per day from a standard airplane runway and take payloads into space, which is absolutely transformative. And there's some interesting things around how that matters in today's world, where we have of a focus on environment and we have more software and software-defined systems that require space. 

So I'm excited to bring you this story and I have to do the full disclosure on that. So I'm going to invite Stefan to start and then Koen to carry on. So Stefan, tell me a little bit about your background and how you came to found Dawn, what Dawn is about, and then Koen tell me a little bit about the software that Stefan wants to know nothing about.

Stefan Powell: Thanks, Christina. That's for the introduction. This is really cool to be on the podcast, and yeah, I definitely am a little bit out of place with podcasts, but hopefully I add some context to the story. Yeah, so I'm a Kiwi but I went over to the Netherlands to study and that's where Dawn really started. It kind of spun out of some student project that we were building; pretty classic, regular, single-use rockets. 

You know, this was a five-year-long project with sort of 80-ish people working on it, and it all came to about 90 seconds of fun where the rocket worked well, but even after it worked, you know, it all ended up in the ocean and was completely stuffed. So it was going to take years to rebuild it. It was just kind of out of this disgust really for the status quo and just realizing that there really has to be a better way to do this that Dawn Aerospace was founded, and it was really on the principles of wanting to do things more sustainably, more – you know, higher reusability, and in a way that doesn't –

Not just affect the world in an environment sense but in terms of, you know, in space use, in terms of regulation. It all just had to attend to the existing world in a much more intelligent way, because everything that we had done in the student project was just – it was just pushing shit uphill. [Laughs] That's how it was. And yeah, to build a viable business we needed to find a better way. So, yeah.

Christina Noren: So Koen, you're a Dutchman in New Zealand right now, so tell us your story.

Koen Lucassen: So basically it started off doing my master thesis, which was in the gliding reentry vehicles. So they start off very hypersonic, and so basically a very challenging environment and you have to go back to earth basically and land it on a runway. So I was always into by and by just doing the most challenging things and the most difficult things, and of course the hardware's a big part of it. You need to have the proper hardware to make it happen, but the final thing really happens in software.

So the software really does it at the end, but yet you need both things to make it happen.

Christina Noren: Tell me what the main functions that the software that controls the _____ aircraft are and what's involved in delivering that.

Koen Lucassen: Yeah. So the software of them, basically we're riding a full autopilot, so it starts off – we need to take off of the runway and then we kind of pitch up basically 80 degrees, so going straight up like a rocket, and yeah, we kind of do an parabolic trajectory just like a rocket.

So we kind of reach an altitude of approximately 100 kilometers and then, yeah, we have to go back, so we have to reenter in the environment and we have extreme heating and a very high mechanical load, and we basically – the software takes control of the vehicle, so it does a specific amount of maneuvers to reduce the heat load and reduce the mechanical loads and finally steer it back to the runway and make a landing so we can reuse it. 

Basically like we can attempt another flight in the same day, basically.

Christina Noren: And how many flights a day are you doing on the aircraft?

Koen Lucassen: So right now everything is in development, but on our test planes we can do multiple flights a day, yeah, in the MK-II that's outcoming. 

Christina Noren: So Stefan, I'm going to turn it back to you a little bit in terms of the why. So what are the kinds of payloads that customers, potential customers are going to take _____, and why does your approach, you know, matter? You know, there's been a lot of news in the last week about SpaceX and reusable rockets landing in the ocean and coming back to earth with astronauts onboard. So, you know, what's really driving this approach?

Stefan Powell: Yeah, so what's really driving us towards reusability is the fact that you just need to get everyday access to space for it to really be a frictionless destination, you know, somewhere that you can go on a regular basis at low cost. So what do people actually want to do in space? It's all kinds of – everything from communication, so be it internet from space or IoT services from space to earth observations. 

You know, that's taking photos of earth or synthetic aperture radar images to do everything from count cars in car parks to look at heat signatures of buildings to measure, you know, atmospheric conditions or moisture in the soil. There's just literally thousands of applications for this data. What's very near term for us that we are intending to do with the MK-II, which is a suborbital plan.

So it's not able to go to space but it's able to fly up to 100 kilometers altitude and come back down. That's actually an incredibly interesting part of the atmosphere because it's too high for balloons but it's too low for satellites, and so we actually only have a very small number of data points of what the atmosphere is actually doing up there and what it looks like, but we know that this particular section of the atmosphere, the mesosphere, it's very instrumental in long-term climate models and things like predicting weather and climate change.

But we have very little data to actually inform these models, so that – long story short, they're very poor models, and if we have more data to be able to inform those models we'll be able to make more reliable predictions of what's going to happen, and that's really important for, you know, farming in the next 20 years, and knowing what's going to happen to the climate is incredibly valuable.

Christina Noren: Yeah. It's funny because your grandfather – and again, the full disclosure – your grandfather was the first person to put me on a computer, on a Commodore 64 in 1984 [laughs] and, you know, he was the leader of your Dutch family and your cousins and aunts and uncles coming – your great aunts and uncles coming to New Zealand, and he was a pioneer in using computers in farming.

So it's an interesting kind of, you know, full circle. So on that, are you guys at Dawn actually harvesting this data on the mesosphere and climate and doing any processing yourselves as proof points? And Koen, I'm assuming that that probably falls more to you if you are doing it from a software perspective.

Koen Lucassen: We're not doing it yet. This will be – you know, the MK-II is primarily a development vehicle for us.

You know, the use case for us is to show that we had the technology do this sort of thing. We're already approaching other organizations like the New Zealand equivalent of NOAA that does a lot of atmospheric research, so they're very interested in designing a payload to be able to fly in this aircraft to be able to take measurements every day of the atmosphere. But certainly that is one area that the company could get into.

You know, if we've got the most capable research vehicle in the world we can – you know, instead of selling flights on that, sell data that it produces.

Christina Noren: Well it's funny, later this week we're going to be interviewing a fellow Kiwi entrepreneur who has a company, PredictHQ, that is essentially a company that sells location data and climate data and event data and data to help predict the impact on demand in different industries, so there's a lot of connection there. So Koen, I want to switch a little bit into software delivery.

So frankly, I think Paul and I are more used to the software delivery of web applications and, you know, _____ applications. What does it look like – you know, what does your software development pipeline look like? What is the technology that you're using? How frequently are you doing updates? How do you get software to the hardware? How are you controlling things remotely? Like just take us into that.

Koen Lucassen: Yeah, so we definitely do a big part of continuous integration and deployment. So mostly, like every piece of code we write we want to have at least one unit test so we at least have a test coverage of like 100 percent. That's where we aim for. So we have the individual unit. That's what we test for, like new features, like individual features, and of course we have also integration tests like software-in-the-loop or hardware-in-the-loop, and then that becomes also already closer to the deployment.

So then we just run a very big software-in-the-loop and hardware-in-the-loop test and we basically simulate a parabolic trajectory of any kind – of any sort of mission like that. So we have – we aim to have a lot of various, different missions that we can basically test our whole code and how it behaves in different scenarios.

But the real deployment really is when we have – so once we do a release and that's – we aim to do it like every two weeks, and that basically triggers a series of test flights and we do it on a small airplane first to see if it's really robust, _____ and reliable, and once that works fine we basically use it and test it on our bigger airplane, so the MK-I, and if it then works – of course we get continuous feedback and we solve all the bugs and stuff we basically encounter doing the flight test.

Christina Noren: Is the reason to test on a smaller airplane, Koen, is that, you know, if there is something drastically wrong you have less loss?

Koen Lucassen: Yeah. Well basically it's very forgiving, the very small airplane. So you could do everything wrong and you still save it, but of course that's the not the idea. But like at least if you – I mean flight testing always brings a little bit of risk, and ideally you want to take away a lot of risk already just on the smaller airplane where it's not going that fast and you still can recover if something goes wrong.

So yeah, a big part of our testing and very – like you can test a very new algorithm which has never been tested before and in real flight testing, because you have the same chances we will be using on the bigger airplanes and you can just see how the interaction with those sensors or those certain systems are, and you can get the basic major mistakes already out of the system before you really deploy it on a bigger airplane.

Yeah. I think that's really important. 

Christina Noren: What about the earth side? So I'm assuming there's control system software deployed somewhere. I'm assuming it's probably deployed somewhere in the cloud, so tell me a little bit about that.

Koen Lucassen: So yeah, we also developed a piece of – we also developed our own ground station software basically, which talks with the airplane. Yeah, so we have a couple antennas on the ground which communicate, of course, and yeah, during flight we get continuous feedback.

So we know what is happening and in which state the current autopilot is, so yeah, that's how we monitor the flight.

Christina Noren: How coupled is the delivery of software to the planes and software to the ground systems?

Koen Lucassen: Well in principle, once we release a new autopilot version we also release a new ground station version, because we have new messages – or we introduce new messages in the autopilot, so we also need to introduce them into ground station.

Otherwise we don't know what's going on. We can't read those messages. So they are linked, basically. Yeah.

Christina Noren: Where's the ground station software run?

Koen Lucassen: Just on a normal computer.

Christina Noren: An actual, physical server, not on a cloud system?

Stefan Powell: It's usually just on a laptop.

Koen Lucassen: Yeah.

Stefan Powell: Because then it's an entirely remote operation. So we pack everything into a trailer and then we're able to drive out to wherever we're flying, whichever airfield that may be.

And then we set up and we can be completely offline as we don't need internet connection to be able to run the whole thing, so run on local computers.

Christina Noren: So what's the nature of the connection? So again, we're exploring this – what we love about geeking out on this particular project is it's so different than the software systems we're usually talking about, so what is the nature of the communications between the ground system and the truck and the trailer and the space planes in the sky?

Koen Lucassen: So yeah, the connection between the airplane and the ground station is – so we have a 900-megahertz link or an S band, so if everything goes bad we always have the S band as a backup, and that's basically the communication between ground station and the airplane.

Yeah, then we have a serial connection from the antennas to our laptop, which provides all the link to the laptop.

Christina Noren: Back to the business end of this. So how have you and James at Dawn thought about the changes to our world since the lockdown and the pandemic and the associated, you know, realization that we can have less of a climate impact? There's a lot going on right now and it impacts every business. So can you talk a little bit about, you know, how you've thought about the impacts of the last few months' changes to what you're doing at Dawn?

Stefan Powell: Yeah. So like in terms of how we run our business, we certainly used to be quite reliant on international travel and we've had to make our teams somewhat more robust or at least self-contained in their respective countries. So, you know, we have an office in New Zealand but another major one in the Netherlands. So yeah, I mean that's a very practical consideration.

In terms of our product and what we're making, we actually don't have to have a whole lot of physical interaction with our customers, and as long as international logistics can still happen, the space community was always very good at finding or using international supplies anyway, so it probably hasn't changed our customer habits too much in that respect. They're still just as willing as ever to go internationally to buy the best components to build the best satellites, essentially.

Christina Noren: Is there an impact on what your customers are doing? Like, you know, is there any sort of increased demand on what the satellites are doing?

Stefan Powell: Oh absolutely, so just because there's a lot more governance making it happen, particularly in Europe and New Zealand. It hasn't quite happened in the States yet except through military. You know, more government spending almost invariably means more spending on R&D and satellite systems in general.

So that doesn't mean that any particular type of satellite is being favored, but there's just a lot more going up in general. So yeah, the demand's going up, if anything.

Christina Noren: Paul, do you have any clarifying questions?

Paul Boutin: When you talk about taking off from a regular commercial airport runway where there are passenger planes, how does that also affect the air traffic control aspect of, say, launching satellites with a rocket versus from one of Dawn's planes?

Stefan Powell: Yeah, so that's an interesting one, but in terms of other air traffic users our plane is no different from any other plane. It sends out exactly the same signals that other planes interpret, so this is ADS-B as the generic signal that every plane has this transponder that says, hey, I'm a plane here and I'm Delta number four or whatever.

And the other part is that other planes can communicate to one another. So our aircraft has a transponder onboard that will relay any communications from an aircraft back down to us, so in terms of signals we can receive we are in exactly the same place as other aircraft, so we can talk to any other aircraft that may be high up in the sky that want to communicate, and we also need to have an ability to be aware of our surroundings.

So that's also visual. So for now, we would likely fly in experimental airspace where there will be very limited traffic so we don't need any of that long-term. We're looking into using vision assistance to be able to, you know, use basic machine learning stuff to be able to identify what's a bird, what's an aircraft and then automatically alert us and pull up a screen that the pilot on the ground will see of there's an aircraft on my starboard side.

You know, so that the pilot has equal or better situational awareness to a real pilot, essentially.

Paul Boutin: Where I was going with that was it sounds like eventually we will be able to launch satellites or high-altitude data collection flights from fairly regular airports, if not necessarily LAX down here, but without having to disrupt the normal air traffic flow. Is that correct?

Stefan Powell: Exactly. Yes, that's exactly the point. It's just an aircraft. I mean it has an engine that's pretty loud but most aircraft engines are pretty loud. It flies pretty fast. There's some established rules for, okay, you can't fly really fast at low altitude over populated areas. Fine, we can adhere to that. When we want to go really fast we're actually out of legal airspace. You know, that's above 60,000 feet, so about 20-ish kilometers. 

Once you're that high you're no longer in what's called airspace. You know, there is enough air to still fly if you're going fast enough, which we are, but we're heading straight up at that point, so that's when we're into rocket domain. And as soon as we come back down we're in aircraft – we integrate back into airspace, other aircraft can see us, we can see other aircraft. For all intents and purposes we're just a regular aircraft. It's just that we have a rocket motor on board and not jet engines.

Paul Boutin: And so my point is that getting these satellites or flights into actual space is not the special disruptive event of a rocket launch as we know it today.

Stefan Powell: Exactly. You should be able to do it ten times from a pretty regular airport with passenger aircraft flying in and out of that same airport.

Christina Noren: That brings me to the next area I wanted to explore a little bit, which is, you know, we have a pretty global audience at CloudBees.

You know, we're a global company with people in 18 countries and are customers are similarly distributed. You know, there are some special characteristics of the Netherlands and New Zealand respectively that benefit Dawn and, you know, the rocket culture coming out TU Delft and the airplane culture that comes – you know, the independent air flight culture that comes out of New Zealand. So I'd love for both of you to talk a little bit about that and how that helps Dawn.

Koen Lucassen: So I actually myself was not part of that, which is the rocket club in the University of Delft, so I was like first going to a different university, and then doing my master's I was actually looking for like a challenging environment, and space, of course, is the most challenging and it's really booming at the moment, so that's what brought – drew me into it. I've done – this is a bit of a tangent, I guess – yeah, so I've done my internship at a spaceport bay – well not a space – a rocket launch platform in the north.

So that's where my really passion of rockets developed. So I was doing really hands-on work with, like, student _____ rockets and integrated some systems there. 

Christina Noren: It matters that you're doing these launches in New Zealand, right?

Stefan Powell: Oh, yeah. Absolutely. Like from a regulatory point of view it's really hard to do to this stuff in the Netherlands. Even though – I mean like there's a lot of people, that makes it harder, but it's actually just the way that the regulations are set up.

They just prescribe exactly what you have to do to be safe, and if you can't do that you can't fly. So they say, "Oh, your jet engine has to have such-and-such amount of redundancy," and then we say, "Well we don't have a jet engine, sorry. We have a rocket engine." They say, "Well, sorry. I don't know what you're supposed to do for rocket engines. You can't fly." 

Christina Noren: Could you do this in California? 

Koen Lucassen: Not exactly under the ruleset that we would like to, but we believe we're going to be able to transfer a New Zealand-based certification to the US. There's a lot of precedent for doing that. So that's our kind of entry into the market, if you like.

Christina Noren: Right. We've had another guest on here who's a Greek entrepreneur who sold his company to Splunk and, you know, he made a big deal about Silicon Valley being the only place to innovate and, you know, he's got the starry eyes about the sort of Silicon Valley cult and, you know, I kind of wanted to feature you guys as the kind of startup that couldn't happen in Silicon Valley.

Stefan Powell: Well there's at least one Silicon Valley aerospace company that's testing their aircraft, you know, about 100 miles from us, specifically because the airspace regulations in California make it difficult. You know, even though California has the Mojave Desert, it's like this perfect testing ground, but it's still harder to do stuff there than it is to just go to a small airport in the central part of the South Island of New Zealand with clear skies and stuff all going on and a regulator that you can call up on the phone and have a conversation with.

That's just not the case in the US. 

Christina Noren: Yeah. And also, you know, so this is an audio-only podcast, but I'll say that I'm looking at your two smiling faces. I'm looking at two people working in an office without masks on in a perfectly safe way at the beginning of August 2020. In ten years, assuming that, you know, your test flights go well, you got your early customers, you keep developing aircraft, keep developing software –– what do you think the space flight that Dawn does looks like in ten years, and then, Koen, if you can sort of say, like, you know, when you've grown large and you have a team of 30 avionic software engineers, what does that look like?

Stefan Powell: So in ten years' time a pilot should almost be an afterthought. It might be a legal requirement, but essentially the aircraft is completely able to fly itself.

You'll service it on the ground, you'll payload it in and a satellite in it and you'll tell it to go. It'll autonomously do a whole bunch of checks. It'll taxi itself out onto a runway, it'll integrate with the airspace, take off and fly itself to space and then deploy a payload. And the end product is a satellite in space that the user can then communicate with and use.

And the impact on the environment is essentially just burning a bit of fuel, you know, just like any other transportation is. So the aircraft comes back and it's able to, you know, refuel, reload new payload and fly again probably a couple of hours later, realistically. Software-wise, that's incredibly intense and I would imagine you're going to need more than 30 people to do that. [Laughs]

Koen Lucassen: Yeah. Definitely the goal is to get like a full, autonomous flight. Well, it's a goal. Yeah. 

That's what we work towards, and yeah, probably requires quite a lot of people as well, because operations is such a big part of the whole development as well. I mean you can do like a big part of both software and hardware-in-the-loop, especially if you validate all your aerodynamics and then you can basically do everything, but still the flight testing is the real test and that also requires time.

Christina Noren: Stefan, I wasn't counting operations people when I said 30. 

Stefan Powell: Oh, no. Yeah, I'm meaning just people that –

Koen Lucassen: Yeah, yeah, yeah.

Christina Noren: There's a significant number of reasonably productive software organization. Please go ahead, Koen, I'm sorry.

Koen Lucassen: Yeah, of course we were – yeah, we want to launch payload basically, like we have specification and requirements for payload for a customer, and then like we want to have also a flexible autopilot so we can aim for like a certain amount of microgravity time and adjust our mission for that, and that also requires big team because every new payload with new requirements would require a slightly different trajectory, which we have to simulate and check if everything – if all our systems work well for that.

So yeah, we really need to – especially –

Christina Noren: So give me an example, Koen, of two different kinds of payloads and the reasons and the nature of the different trajectories.

Stefan Powell: Yeah, I could probably speak to that.

Koen Lucassen: Yeah. 

Stefan Powell: Say two different payloads, one wants to take an atmospheric sample at 80 kilometers altitude.

So, you know, they just want to measure what the atmosphere is doing at that point and they probably want to do it probably once every six hours or so. So yeah, the software has to be able to accurately get the aircraft to that altitude. Now the problem with that is that your stations essentially sit at much, much lower altitude and, you know, from about 25 kilometers you've essentially got all the speed and all the moment you're going to get and your engine is off.

And it has to be able to forward that trajectory to be able to, you know, hit that altitude that you want to get to. Another mission might be a low-altitude, high-speed, hypersonic research flight, so for, say, investigating new heat shield materials or something. That's certainly one thing we want to look at for improving our space planes.

So that would mean flying what we'd call a constant dynamic pressure trajectory, so you'd want to fly at such a speed that you have the same amount of heat and air essentially hitting the vehicle. So you have to match not just the speed but the altitude and the resulting total wind effect hitting you, the total pressure hitting your vehicle. So it has to, you know, have a smart predictor of what's going to – you know, how is the atmosphere going to change with the altitude and how is it going to change –

– you know, how is your acceleration going to change with the vehicle weight changing and thrust change and all these things?

Koen Lucassen: And of course, above all you need to manage constantly your energy, so you need to be able to get back to the runway. So energy is a very important parameter there, so basically down range as well. So you need to manage those two to get back and land safely.

Christina Noren: In any of the test flights have you had to abort because the calculations have been wrong?

Stefan Powell: We've definitely had to abort flights where it hasn't responded to, say, like wind gusts or outside _____ as we had expected or where it didn't fly a flight plan quite as we thought. Because this is an aircraft and it flies in a very robust way, you know, it can accept very generic inputs or, hey, I want to fly over there and it'll just figure it out.

Unlike a rocket where, you know, a rocket starts vertical on a pad and it flies a very precise trajectory, and if it goes off that trajectory even a few percent they have to blow it up. You know, because it's this aircraft that has to interpret very generalist inputs the algorithms have to be very broad and very robust, and so sometimes it responds to them in ways that you wouldn't have predicted and that makes it – can make it really, really tricky.

You know, that's what's really hard about making – you know, like an autonomous car is in that way similar to an autonomous aircraft, which is totally dissimilar to telling a robot to go forward three steps, left three steps. That's easy. [Laughs]

Koen Lucassen: Yeah. For that I'm going to say, like, we have software-in-the-loop, so if we have a new mission we can always test it in software-and-hardware-in-the-loop, so we kind of know what to expect. So yeah, we don't have to abort that much because we just know what's going to happen because we already flew that mission basically in software or hardware-in-the-loop.

Christina Noren: Is the test designed that that enables you to do?

Koen Lucassen: Yeah, exactly. That's really important. Because if you do the test in software or hardware-in-the-loop and you know what should happen and then something happens that shouldn't happen, then you know when to take over as well, so you know to identify the things that should not happen.

Christina Noren: This last passage makes me think that it's almost like the difference between your space plane and a standard rocket is the difference between a car and a train.

Stefan Powell: Yeah. Absolutely. Yes. That's a very good analogy, actually.

Christina Noren: Any last thoughts for people who are generally interested in interesting applications in software in terms of things that are – you know, if you were at a cocktail party with a software engineer working on standard web applications, what would you tell that person to get them interested in software for space?

Stefan Powell: Well, I think like the continuous integration and testing is really cool, because like basically once you develop a really – or a new feature, you want to have immediate feedback of like how it performs, how it performs in this flight, because like you have a sort of anomaly in your previous flight and let's say that's a bug and you need to solve it. I mean you have an idea of the direction you want to go, you implement it, but of course you don't know if it works directly. So that's really nice to have those pipelines and a program mission to really have direct feedback of what you did.

That makes you really happy when they succeed and it actually works.

Christina Noren: Paul, what's your takeaway from this conversation of what really grabbed you?

Paul Boutin: Well, what I didn't realize is that the most advanced software part of flying a space rocket plane out of a regular airport is really about its guidance and about having it be reused, especially that latter is not something where I would've thought, oh, software can help with that. Well, there we are.

Stefan Powell: So the guidance is actually such a complicated problem. There's just so many degrees of freedom and how you could fly this vehicle that it's even hard to predict how much performance we can get or even how to design the vehicle. So very early on in the development we actually just used a generic algorithm to help us decide, you know, how big should we make the wings, how slender should we make them. You know, all the aerodynamic design was put into a generic algorithm with a trajectory simulation so we could allow it to change the design of the aircraft and see how high and how fast they would let us go.

Because this is such a complex analytical problem it's damn well near impossible to solve on paper, but through brute-force methods we were able to get at least somewhere close to what we think is optimal.

Christina Noren: So in terms of doing that kind of analysis up front, are you using Python notebooks? Are you using data science tools? Are you building specialized software internally that helps you do those calculations?

Stefan Powell: It's all run in Python, I believe. Isn't it? 

Koen Lucassen: Well it's just C++ code.

Stefan Powell: Oh, C++, okay.

Koen Lucassen: Yeah. So there are some tools like – also TU Delft comes back into that, because I think that TU Delft – what is it, like– it's like an astrodynamics toolbox basically and it has some atmospheric models as well as the flight dynamic model in it. So you can also model your airplanes and put the aerodynamics in them and you can simulate your trajectory.

And then they have a genetic algorithm which is basically a random optimizer, so it chooses to point you randomly and then for the network evolves and only chooses the best solutions and then tries to combine them to find the best trajectory possible. Yeah, so that's mainly within C++, and like we can use the main toolbox of TU Delft for that.

And it's also quite nice because I think the people who are mostly working at Dawn are from TU Delft, so they have experience, they have done projects with TU Delft. So yeah, so they know it and can easily work on that and extend it for our purposes.

Christina Noren: So Paul, I think we've got everything we need for the session. What do you think?

Paul Boutin: Everyone's going to ask me when do we get to watch it fly?

Stefan Powell: We're definitely working on it and it's going to be pretty soon, probably within the next few months. 

Unlike rockets, once again, this is very incremental development, so the first flights are really just going to be hops off the ground and fly a few circuits within visual line of sight and land again. But over the next 12 months or so we'll be pushing that performance out further and further, and within 12 months we should have a rocket motor on board and then we'll start doing flights that really, you know, look like space, as in we'll be high enough and fast enough that you'll – you know, the sky will be black. You'll see this blue shimmer over the earth and you'll see the curvature of the earth. 

Christina Noren: I assume you'll have cameras on board to shoot these images back.

Stefan Powell: Oh, absolutely. I mean we have to have them for situational awareness and for publicity. 

Paul Boutin: [Laughs] And so all of us Americans who may still be stuck in the house then can watch it. That will be great.

Stefan Powell: Yeah. Absolutely. We're actually thinking of being able to livestream it.

Christina Noren: Okay. Well thank you, you two gentleman, for spending the morning for you and the afternoon for us, talking to us about the software behind space.

Stefan Powell: Yeah, no worries.

Koen Lucassen: Yeah. 

Christina Noren

Follow Christina Noren on Twitter and Linkedin.