Getting systems to simulate flight behavior in Microsoft Flight Simulator 2024 | Sebastian Wloch interview

For Microsoft Flight Simulator 2024, the ambitious flight sim game that debuts today, the physics and environmental systems are far more sophisticated than the 2020 version.

Part of the reason is that there is just a lot more horsepower available from the internet-connected cloud data centers to solve tough computing simulations. Back in 2020, with the previous flight simulator game, Microsoft only handled part of the processing through the cloud as it added more ground data from faraway data centers to supplement the processing on a player’s own computer.

But now the entire game can be processed in real time in the cloud, with minimal storage required on the players computer (around 50 gigabytes). Now Microsoft says the simulation is like a full digital twin of the Earth.

And as a result, the flight sim is able to simulate many different types of aircraft and many different kinds of situations — gliders, hot air balloons, 747s, supersonic fighter jets — than ever before. In fact, the detail on the ground is 4,000 times greater than what was possible in 2020.

Back in September, I joined the Microsoft Flight Simulator 2024 leaders at a briefing at the Grand Canyon, where we flew a jet and played a preview of the game, which comes out on November 19, 2024.

As part of the presentation. I was paired with game journalist Samuel Stone of Den of Geeks for an interview with Sebastian Wloch, CEO at Asobo, Studio, which is based in Bordeaux, France. Wloch’s team was a big part of something like 800 people who worked on the project over as much as four years.

Here’s an edited transcript of our interview.

Sebastian Wloch: I’m a CEO at Asobo Studio. I’m from France, in Bordeaux. I can talk about technology, rendering, flight model physics.

GamesBeat: I’ve been writing a lot about digital twins because I cover Nvidia, their OmniVerse and all that. I didn’t realize that was something you adopted here as well. It sounds like you adopted it pretty heavily years ago. I wondered about the background to making a digital twin of the earth, of all kinds of airplanes. How did you come to the notion that digital twins were the way to proceed for the whole game?

Wloch: In the flight simulation space it’s always been about being as accurate as possible. Whether it’s the plane physics or the plane visuals. If you’ve seen the video of the history, every step is a bit closer to reality. The technology of today – the satellite images, scanning the planes – pretty much anyone can scan something now with only a camera. The difficulty is just the scale of the planet. It’s huge. Gigantic amounts of data. That’s why we moved pretty much everything to the cloud. It was already 99.9% in the cloud in 2020, but now it’s everything.

The power of the machines of today, whether on the GPU side, which we try to use as much as possible, or CPUs now with 32 cores, enables us to get closer and closer to reality around the globe. We talked about the details we improved, to make it look like the real earth not just from the sky, but even when you land. We’ve had good luck that all these innovations are here now. AI helps us look at images from the top down – is this rock, or is this a field? All these technologies allow us to do this now.

GamesBeat: So many parties want to do digital twins. Nvidia’s intention was to predict climate change in decades to come. They have the same aim that you have of rebuilding the earth in a digital form. There must be a lot of parties with the same intention.

Wloch: But maybe the use is not the same. We specialize in the experience of flight simulation. When it comes to climate, we try to have the most accurate representation of the air in an instant. If there’s an updraft in the afternoon, something like that. We don’t try to predict climate over time. You usually get the best result if you specialize in something. You do something that’s maybe not as big and do it well. By focusing something like the climate simulation on flight, we simulate drafts that could happen as a result of a single house, or a tree.

You tried the helicopter tour, right? I did it this morning at six. There was barely any movement. In the morning the atmosphere is calm. But in the afternoon you probably had some bumps. We simulate that as much as we can, very accurately. It’s a different issue, a different type of problem-solving to have an accurate simulation of turbulence around the aircraft, compared to predicting climate around the planet. We’re lucky today that we have all these technologies. Everyone specializes in their field. We’re able to move forward very quickly. It’s a great time to be doing these things.

GamesBeat: Nvidia was looking for meter-level accuracy in their earth simulation. You’re all looking at the same physics and details at this very small scale. But you use it in different ways and get different products. Is all of their data as useful to you as the data you acquire? Or do you have different intentions, different data, and different applications in mind?

Wloch: The best person to talk to about exact data sources is Jorg. But we try to find the best possible data sources for our needs: a realistic planet that looks great from the sky in the ground, good weather, all these things. There are more and more data sources today. It’s more and more accessible. I don’t know what sources other parties are using, though. Jorg can give you more exact details on what, when, where. Around the planet, it won’t be the same source everywhere you go. It depends on where you go. Different companies specialize in different countries, areas, and systems.

Samuel Stone: Talking about physics, gravity is gravity, but how do you refine the physics going from single jet planes to hot air balloons, zeppelins, gliders? How do you make sure you get that refined experience as you hone the physics engine for 2024?

Wloch: A big improvement we did in 2024 is that we’re now fully equipped with instruments to measure the flight dynamics of real-world aircraft. Acceleration forces in the aircraft. We measure the position of surfaces. We put measuring devices on the outside to measure air speed and so on. We have extremely exact data on how these aircraft behave in the real world, and we’ve implemented the exact same instruments in the simulation.

The first finding we had was that our simulation was pretty accurate, but it’s all based on what data we have to describe the aircraft. We were able to improve that a lot in 2024. Many aircraft have been updated with much better flight models, much more true to life. We obviously improved the physics engine, but that was more to add more detail, smaller details, more realistic effects, like weight balance. I had that a few times when I flew. When you do a 360, sometimes the bump is you. It’s not the air or the heat.

The instrumentation on the aircraft, getting them to fly–we measure them on the ground and in the air. We measure everything. That helped to get a whole other level of realism to the aircraft.

Stone: In terms of the digital twin, one of the beautiful things about MSFS, it’s dynamic. It’s matching what something is doing in real time. How is it feeding that into the system? Right now it’s monsoon season in Korea, so this is how the weather would interact. Or it’s really cold now at this altitude in Canada. How is it taking all these different environmental factors in real time and working them in?

Wloch: I made a list of systems. That’s system four. It goes very wide in terms of scale. The parameters we have are things like the position of the sun, for example. We know how much energy the sun releases. It’s 1,400 watts per square meter, something like that. But then the air will absorb some of it. We know about atmospheric pressure, low or high. We know the cloud cover. We know how much makes it to the ground. Often it’s just 600 or 700 depending on the altitude, sometimes 800. We know the angle. Then we compute how much heat this creates.

We’ll compute this for the whole area around the user, how much the ground is heating everywhere. Maybe a parking lot in direct sun will get very hot. We compute that surface temperature, which is transmitted to the air. The hotter air will want to go up. Colder air wants to go down. This creates movements in the air. When the air is humid, it will also go up and cool down, which creates clouds. We simulate all this. We have the air temperature from the weather surface. Usually they give you the air at a two-meter height, and then by layers. We can correct when we know, for example, that the sun has been hitting a parking lot all day. That might be very hot, but then the air outside is only 15 degrees. Maybe there’s a wind from the north and cooler air is coming in. Maybe there’s not much of an updraft anymore.

If you fly in the afternoon in the summer, when the sun is high, you get something exactly like you had today. A lot of bumps. But we flew in the morning on the Vision Jet. It was very smooth. I had one single bump, just this little bump. I have the videos. Nothing moving. But it was six in the morning. That’s how different it is. Summer, winter, morning, afternoon, there are huge differences. Especially in the Grand Canyon. All day long, the sun is in the south there, but at angles like this. It will only hit one side of the canyon. The other side isn’t heated. All day long it will go like this. When you fly over this area, you get a big bump.

Stone: There are more than 15 million unique users playing 2020. What were some of the things you noticed in user and community feedback about physics and how the planes handle that you were mindful of going into 2024?

Wloch: People gave us a lot of feedback. We try to listen to what everyone thinks, and to reality. We built all these systems to measure the planes. The simulation is always based on reality, but the problem is, what do people feel? That’s based on feedback. We can still impact the feel without changing the realism. For example, the way we move the camera. Maybe you did the F-18 challenge in the Grand Canyon. The camera does all sorts of things. We try to make you feel all the bumps and weight impacts in the yoke. If you flew the Vision Jet, it’s very heavy. It resists a lot.

GamesBeat: I was fighting the stick, yeah.

Wloch: Right. There’s a way to map what you do with the yoke to make it feel a bit heavier. It’s exactly the same physics, but we try to impact the feel. That’s why we get a lot of feedback to improve the systems so it feels more realistic, even though the actual simulation is just based on real parameters.

GamesBeat: Is that a matter of haptics?

Wloch: In the real world, the elevators go up and down. The force required to do that when you’re flying slowly is much lower than what’s required when you’re flying fast. What we map is not the angle, but the force. When you need a lot of force, we make you move it more. In the real aircraft, you barely move, but it’s very heavy. What we do with the yoke, when it’s supposed to be heavier, we make you move more. If you have force feedback then that’s a different thing, but most people don’t. We try to convert that resistance into amplitude of movement. If you have to do a lot of movement, it’s still a bit more tiring. You feel like it’s resisting, so you need to move more. That creates the same feel. Plus the camera, the feeling of the bumps, you can get the same effect.

GamesBeat: You had a very big improvement in cloud technology between 2020 and 2024. Is it simply a matter of data centers getting bigger, or did other things lead into that?

Wloch: It’s a lot of GPU. We rewrote the GPU code for the clouds. It’s very complex. Photons come from the sun, hit the atmosphere, and change color. It goes into the clouds, bounces off here and there. In 2020 this was very simplified. We couldn’t trace, for example, the sun coming through the atmosphere. We didn’t know if it was going high in the atmosphere or low in the atmosphere before it hit a cloud. It was approximately the same everywhere, and so all the clouds had the same color.

GamesBeat: What I meant was the cloud computing. You had only partial cloud computing in 2020, but now it’s completely cloud computing. You use a lot less storage.

Wloch: Yes, we moved the storage almost entirely to the cloud. A few critical files are local, but it went down from 150 to 20 gigs, maybe 30? We keep adding more computing. We don’t cloud compute in real time, but we pre-process stuff. The data is already ready and you just download it. We do the real time computing locally and then pre-process cloud computing every time we ingest new data. We run however many thousands of machines. Sometimes it’s AI and sometimes it’s just processing, but a lot of it is AI.

GamesBeat: The quality of cloud computing just gets better?

Wloch: Yes, better and better. We also make an effort–sometimes you have to rewrite a system entirely to make it cloud-based rather than running locally. The pipeline of data and what happens is not the same.

Disclosure: Microsoft paid my way to the Grand Canyon for the purpose of this story.