Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
Ingmar Jungnickel on AiRO and Scalable Aerodynamic Testing
527 Plays1 day ago
Sports aerodynamicist Ingmar Jungnickel joins the show to discuss AiRO, a new CFD-based aero analysis tool built for bike fitters and coaches. He outlines why traditional wind-tunnel and field-testing approaches struggle with cost, complexity, and repeatability, and explains how recent advances in computing and AI now make high-fidelity CFD practical at the fit-studio level. The discussion highlights AiRO’s emphasis on breadth of exploration—rapidly testing wide parameter ranges—rather than pursuing ever-higher fidelity in isolated conditions.
Key Points
- Ingmar’s background spans wind-tunnel work, velodrome testing, CFD development, and elite projects with Specialized, the German Cycling Federation, and U.S. Speed Skating.
- Conventional aero tools—wind tunnels, track testing, field testing, PIV, and ultrasonic tomography—are limited by scalability, environmental control, rider repeatability, and cost.
- AiRO’s core components include:
a parametric human model fitted from simple photographs and basic anthropometrics,
a digital-twin rider posture controlled through ~18 intuitive sliders,
fast, cloud-based CFD simulations. - AiRO’s strategic focus is breadth over depth: enabling broad parameter sweeps, positional permutations, and large-scale exploration that would be infeasible in a tunnel or on a track.
- Digital repeatability eliminates rider-movement noise, one of the main sources of error in physical aero testing.
- Current limitations include the absence of textured-surface modelling (aero socks, advanced skinsuits) and single-yaw simulations, though both are technically feasible as compute economics improve.
Links & Resources
- AiRO: https://airo.app
- Contact Ingmar Jungnickel: ingmar@airo.app
- AiRO Instagram: https://www.instagram.com/airo_app/
Recommended
Transcript
Achieving Sub 1.5 CDA by LA 2028
00:00:00
Speaker
A world-class CDA by LA o Olympics 2028 is sub 1.5. what on What? In what discipline? Just, okay, track or like world-class time trialist. Even like road time trial is 0.15?
00:00:13
Speaker
Yeah. Holy smokes. And um this is just of a linear interpolation of the trend of the last decade. So the question you can ask is if your goal is to get to sub one five, how many bolt holes do you need to tape over to get there? You need to drill more holes so you can tape them over. Hi
00:00:38
Speaker
hi everyone. I'm Andrew.
Introduction of Hosts and Guest
00:00:40
Speaker
And I'm Michael. And you're listening to the Endurance Innovation Podcast.
00:00:57
Speaker
everyone, and welcome back to Endurance Innovation. Joining me today, feels like I have two guests on today. ouch One of them you just heard a second ago. That's Andrew is with me again, and I'm very excited to have him back as co-host today.
00:01:11
Speaker
um But the real guest of the show is ah Ingmar Youngnickel. And Ingmar is a co-founder of a technology that I am very interested to learn more about, and that is Arrow.
00:01:26
Speaker
But before we jump into what Arrow does, Ingmar, I want to say thank you very much for taking the time and joining us on the show. And
Ingmar's Journey in Sports Aerodynamics
00:01:32
Speaker
ah introduce yourself. Give us a little bit of a background of how you got to where you are today. Well, first of all, thanks for having me. This is really fun to be on this show.
00:01:41
Speaker
And yeah, I am a sports aerodynamicist. Sports aerodynamics has been my dream since high school. High school is now 18 years ago. So it's been a while, but yeah, I started as a triathlete in, uh, when I was 16, doing all kinds of sports science stuff, really getting into physics, looking into all of this.
00:02:03
Speaker
And then I was realizing that the number one resistance triathletes face on the bike is actually aerodynamic drag. not weight, and all the bike companies were talking about was the low weight of their product.
00:02:15
Speaker
Especially 16 years ago. Yes, definitely. 16 years ago. Yeah. Not so much these days anymore, but 16 years ago. And I felt I had figured out this great secret in the universe that you know so few people knew, and it was this big opportunity to see that through.
00:02:29
Speaker
And i hit up all the people even then that were working on this. Key people were Mark Cody at Specialized Back then, Josh Pertner, they're probably all names you all know as well. And just said, okay, how do I get your job? And they said, don't do it.
00:02:43
Speaker
Terrible idea. There's like five of us in the entire industry and there's not enough work to go around. But i hit them up year after year after year and then went to university in Germany, studied mechanical engineering, but really built my degree around low-speed aerodynamics, bluff body aerodynamics,
00:02:59
Speaker
with the goal to get into cycling aerodynamics. Amazing. It sounds like you were able to turn that passion into a career. So tell us a little bit about where you've been, who you've worked with in the past 16 years since high school. And I would like to take notes that I can figure out how to do this myself.
00:03:14
Speaker
this is I think this is probably the the dream for both Andrew and myself as well. Yeah. So um this ties actually a lot to what I'm doing now. And it's crazy how things sometimes go full circle and on like 16 year radius circles.
00:03:30
Speaker
Right. But yeah, so in university, was doing a bunch of different stuff, racing bikes or racing triathlon, working as a research assistant at the local wind tunnel, at the university wind tunnel.
00:03:43
Speaker
And i had a side project with a friend that aimed to do outdoor aero testing. There was a paper by Professor Robert Chung. You guys probably know that name as well, had read that paper. The famous Chung method, yes. I lived in a Bay Area later, got to meet him, still one of my big influences in my life. So cool if he listens, Robert, thank you for everything that you've done by just writing this paper. um Yeah, I have basically actually learned programming implementing this paper. So this was probably the first code that I've ever written as well, trying to make that work and doing extensions. And I was working with that with a friend for a few years and then Alpha Mantis came out with their project.
00:04:21
Speaker
And yeah at that point, we realized that what we had built was very similar. At this
Career Highlights and Aero Testing
00:04:26
Speaker
point, we had a pitot tube with infrared tire temperature, with accelerometers. We'd gone a little bit off the deep end with all of this and suddenly you realized, hey, other people are doing this as a business. This could be a business. um Yeah, I was doing that.
00:04:39
Speaker
And the tunnel that I worked at, the German national team was testing at. And Germany has this government agency called FES. That's this engineering group that does work for the German Olympic teams.
00:04:52
Speaker
And I showed what I was doing to them and they thought this was super cool. They wanted access to this. I ended up half like camping at their Olympic training center for six months doing a ton of validation studies and ended up selling the tech to them.
00:05:06
Speaker
And they brought me on to do error testing with them with the German track team. Track cycling? Track cycling. Yes. Okay. Okay. Cool. And actually another good anecdote, we might get back to all of this and all of this.
00:05:19
Speaker
I was enough of an error nerd to have committed all of this time to it and learn all of this, but I had no means to measure my own CDA in my own position, right? I read all the forum stuff. And then at some point I had my first chance to measure it.
00:05:32
Speaker
And my CDA, for those of you that know enough about what these values lie, was 0.26. Mm-hmm.
00:05:39
Speaker
Was this on a tri bike or a TT bike? On a tri bike, poor college student equipment, round tube frame, no sleeved skin suit, road helmet, I believe from memory.
00:05:52
Speaker
But 0.26 was not what I thought this was going to be and was pretty upset about it. But you know then a lot happened within the next year, working with the team, really getting more the attention of Specialized from that side.
00:06:06
Speaker
Then I actually ended up bringing the technology to Specialized, was working with theirs racing division. So then from the German Federation, moved on to working with the Specialized Tour de France teams. oh wow At the time, Boltz-Dolmans, Astana, Tinkoff, Quickstep were the teams that I worked with.
00:06:26
Speaker
went to all the races, did all the velodrome testing. Was it Richmond for this one? Lizzie Armitstedt and Peter Sagan won Worlds. Those were fun things. Was in the team car for Boltz Dolmens.
00:06:37
Speaker
And within all of that, got a shifty tee, got a nice skin suit, got all of that stuff. So within that year, my CD8 dropped from 0 to 6 to 018. Nice.
Specialized and Speed Skating Aerodynamics
00:06:47
Speaker
which is a pretty substantial jump in how far you can go when people talk about error testing as always a fraction of a second or the last percent. That's like a third reduction in drag. Yeah, that's a massive difference. And to give people idea of how big gains are possible.
00:07:02
Speaker
Yeah, and this is the time where you're you're talking about these teams. This was before, you know, aero was really a big deal where everyone is is looking for the the marginal gains in the aerodynamics. Like the world that we live in is a very different world than than the one that you're describing. Yeah, I'd say so.
00:07:19
Speaker
My role was performance specialist, which I think this is the role that's now being called race engineer. And teams are starting to have race engineers. This was kind of what I did at that point. So specialist was pretty ahead there.
00:07:31
Speaker
12 years ago, having a dedicated tour team race engineer doing core simulations and physics models and all of this side of things. Really sounds like something straight out of Formula One.
00:07:43
Speaker
Yes. Formula One was the big influence. One of our side hobbies was building Formula One pit waltz, where you went through like three, four pit waltz because we felt we wanted to be a Formula One team and we carried pit waltz all across Europe.
00:07:55
Speaker
I ended up placing my laptop on top of it and just using my laptop because you could carry it to the hotel room. But yeah, so d yeah from there, S-Racing, super cool experience, traveled 200 days that year.
00:08:10
Speaker
That was at the same time as my last year of college. So I basically didn't attend any class, but I had my dream job. Still ah really high point, one of the highest points in my career, I got to say.
00:08:22
Speaker
um From there, i moved to California, moved to the US, s got my work visa and started working at the specialized wind tunnel. And I had keys to my own tunnel. Amazing. And that was pretty exciting for me as well.
00:08:33
Speaker
What a story of ah like right out of right out of school to get involved in this kind of stuff. yeah um I definitely echo what Andrew said earlier that that you're making me more than a little bit jealous. Fair enough. Dream come true for me. Absolutely.
00:08:47
Speaker
and there's There's got to be you know a touch of luck in there. like You're the right person, but and in the right place at the right time as well. so Having the right connections and meeting the right people. it's It's amazing how sometimes things just come together like that. Yeah, absolutely. Also, truth be told, in hindsight, these plans look a lot more linear. And was like, yeah, this was I think there were many different paths this could have taken, which in hindsight, I would also tell that story just as linear. And like, of course, this was the next step. And of course, this happened.
00:09:15
Speaker
But yeah, since Specialized built a wind tunnel, they became my clear number one dream job anywhere in the world. And getting that was really a dream come true for me. Makes sense. So when we when we spoke last week, in mar you gave me an idea of some of the the novel sort of experiments or technologies that you played around with um over and above the work or beyond the work, I should say, that you were doing with Specialized.
00:09:43
Speaker
Tell us a little bit about those. Yeah. So... When I moved to California and started getting really involved in product development, I really enjoyed that work, but I also recognized that I missed working with high-end Olympic level athletes in the day-to-day.
00:09:59
Speaker
And I got connected with US speed skating. They had Lockheed Martin as their dynamics partner a while back. Oh, yes. Do you remember that? Sorry. I think everyone probably remembers that. It was a bit of a disaster.
00:10:12
Speaker
Exactly. And the backstory there, of course, the Lockheed folks, amazing engineers, but not as used to talking to athletes and explaining things in a way where athletes understood. So Under Armour and speed skating was looking for people that have more sports aerodynamics expertise, and they came to Specialized.
00:10:31
Speaker
And if you're an aerodynamicist and you get to replace Lockheed Martin at something, you say yes to that. And that's how we got the introduction to speed skating, Shane Domer, VP of performance there. He's become a great friend of mine and mentor and partner in all of this.
00:10:45
Speaker
And he asked me if I wanted to chair the sports science commission and work on other projects. Since speed skating is non-competitive to specialized, I was able to do this and was able to work with them on the site and dream up some crazy project ideas.
00:10:59
Speaker
and new technologies with them that fall outside of the domain of cycling. That's awesome. um ah Do you want to talk about some of those? Because it's it's always interesting. i mean, I draw the parallel in what you're telling me in, um let's say, sport development. You see a lot of very, very high level athletes ah who didn't specialize at their amazing sport, they're their top sport, until they're they're quite a bit older. So they they were more generalist as kids. They played all sorts of different sports. so um I think that your experience with speed skating and speed skating aerodynamics is super relevant to what you're doing now. So I'm curious to learn more about it. Yeah, and it went both directions.
00:11:38
Speaker
what i One part of the story I want to touch on from earlier, because we'll come back around to this, is once I had this arrow testing software around this and then I brought it to Specialized, this was my own company, but I ended up not pursuing this any further.
00:11:53
Speaker
And there were two reasons I didn't pursue it any further. One of them was that Specialized was my dream job, so I felt like, hey, goal achieved. Mm-hmm. But the other reason was that we did show this to bike fitters. We tried to get people to use this technology and we concluded that the challenge predominantly was not a technical challenge.
00:12:11
Speaker
People could just not make the business model work. Yes. Okay. All this like, and then you get up at sunrise and then you find an empty stretch of road and then you take your client there and you take all your equipment and long, then you had lost most of the fitters and then you have to run multiple repeats and you have to do this and then sometimes the data still doesn't look good and then you have to run more repeats and all of these kinds of things error testing, ultra testing, amazing technology um but I at that point had concluded that it's just not the technology that's going to work for bike fitters and it's going to work for wide scale adoption not from a tech perspective yeah.
00:12:47
Speaker
Yeah, it's interesting. And I think having spoken to quite a number of of players in that's in that space, I think they've started to get a little bit of traction now, but you're right. it's it's It's a famously high noise to signal sort of business. And it to to do it right, to your point, is it it takes a lot of control ah controls that you have to get right.
00:13:11
Speaker
And it is not an easy an easy operation, despite of what, you know, maybe some of those wind speed measurement companies slash aero sensor companies have claimed. It's it's still not ah not at all a straightforward process if you want to do a good job of it.
00:13:27
Speaker
Yes, many have made it amazing
Advances in CFD and Testing Methods
00:13:29
Speaker
strides and it's become a lot simpler. And I think they're great tools in the toolbox. I don't want to discourage anybody from using them. If you're into this kind of testing, it is some of the most accurate testing that you can do and it's very cost effective.
00:13:42
Speaker
Yeah, no, no, I totally agree. And I don't want to poo-poo this technology either. it's It's kind of the only, to be honest, it's the only technology I've ever used. I've never been in a tunnel and I've, well, no, I've used a CFD. I've used Andrew's stack virtual wind tunnel as well.
00:13:55
Speaker
But I suppose in terms of like doing it myself, this is the only, or doing it with athletes that I've coached or fitted. Yeah, outdoor testing is the only thing I have experience with, but it is it's still not at the level where you could give it to most you know end users and say, here you go get a good test and expect them to get a good result on their own without quite a bit of guidance. Yes.
00:14:18
Speaker
And I think that's the the big hurdle to overcome. Just you you need an aerodynamicist there, and it's not really scalable to to a practical level. So I totally understand what you're saying. Like the business model was a big challenge, and with Virtual Wind Tunnel, I ran into the same thing.
00:14:35
Speaker
um so And I think we'll probably touch on more of the parallels between what I had attempted to do and what you're doing now. Yes, I'd love that. And I think the reason I wanted to mention this, and I wanted to foreshadow that a little bit, is to say that many of the design decisions that we had to make later down the road came to, it has to be easy to use, it has to be fast to use. And that the thing that holds people back is not whether any of these technologies are accurate or whether they can be higher fidelity, is that they're too complicated. And that that was really one of our key things.
00:15:10
Speaker
When I worked at Specialized, there was a time when I had maybe one of these error sensor companies per month reach out to me, know telling me that this great new idea that nobody has ever had before. And they're gonna revolutionize error testing.
00:15:23
Speaker
And I generally gave people flat out that feedback around that. I think they have great technology. Many of them have really cool new innovations and they all have their own secret sauce.
00:15:34
Speaker
But the key thing is really the complexity in the day-to-day. And some yeah, that was has always been my stance on this based. Well, not always. i had to learn that painfully myself. But um yeah.
00:15:46
Speaker
Anyway, um going back to all of this, though, error testing on the Velodrome, we then um I was working with speed skating, as you can imagine, working with speed skating, trying to make them faster.
00:16:00
Speaker
Cool side points, sliding friction coefficients are you know a little bit higher than in cycling. CDAs are roughly comparable. Their posture is not as aerodynamic, but there's no bike.
00:16:13
Speaker
and So the general physics and then the speeds are roughly the same too. So the general physics between speed skating and cycling actually very similar, and you can do a lot of technology transfer interesting from from cycling to speed skating.
00:16:27
Speaker
And as a general point, I forget the exact statistic that I've heard, but it's ah those two sports, cycling and and speed skating, are the closest in terms of the speed you can hold for something like one kilometer, um so in terms of human performance anyway. So they're they're kind of at the limit, but it's it's interesting how closely correlated they are.
00:16:46
Speaker
Exactly. Yeah. It's been great learning for me. now own speed skates. I've done maybe a total of 30 laps around this. I'm still not a good speed skater. It's funny. Actually one of the projects won me a coaching award for a speed skating coach. Essentially have still not ever coached an athlete or know all that much about speed skating other than my specific physics and sharing the commission there for now number of years.
00:17:11
Speaker
and So ah for the speed skating project, were you measuring them in the field or were they in the tunnel? Well, that was the challenge, right? is the We've done tunnel testing for skin suit development. None of the current
Optimization and Broad Search in Aerodynamics
00:17:23
Speaker
approaches that we used in speed skating are super compelling.
00:17:27
Speaker
So um Germany and the Dutch, to my knowledge, have power measuring skates. The US does not have power measuring skates. Without power measuring skates, all this kind of testing I used to do in cycling is out the window.
00:17:40
Speaker
Yeah, I was going to say without, yeah, that was that's where I was going with that question. It's hard to do field tests if you can't measure power. Yeah. Next option was wind tunnel all testing, and this is what we've been doing, but you can't actually statically hold a speed skating position.
00:17:54
Speaker
The way speed skating works, you're essentially falling from one leg to the other all the time. You can't freeze in any of those positions. So repeatability with real athletes is poor. The movement movement is very dynamic.
00:18:05
Speaker
You can do it for suits and mannequins, but for posture, tunnel testing is kind of unsatisfying for me as an approach as well. So now we'll get to one of these other fun offshoots, which is very relevant now with some of the article that we discussed earlier coming out.
00:18:23
Speaker
Another approach in all of this, in aerodynamics and in physics in general, there's this concept of conservation of energy and conservation of impulse. So if you put force into the ice or into the ground, that has to go somewhere.
00:18:35
Speaker
And it turns out that the amount of draft you generate, so the air that you accelerate behind your body, is directly correlated to how much drag you have. So this is called wake deficit. So if you integrate or you you sum up the air behind the body, you do some math with it, it's another way to measure a drag than directly through a force.
00:18:58
Speaker
Okay. so I'm going to pretend I know what you're talking about, but yeah. Okay. Yes. That's why we have Andrew here. Yeah. Andrew, go ahead. Yeah. No, no. It it makes sense. it's It's something that I've i've looked at before with um with general aerodynamic principles. I remember in in class integrating velocity profiles, something like that, to come up with the amount of drag you'd have.
00:19:22
Speaker
In aerospace, this is actually very common if you measure the drag of an airfoil of a 2D section. You actually take a little wind speed sensor behind it or a pressure sensor and you step it through the profile.
00:19:35
Speaker
You get the the wake profile and you can essentially do some math with that and it gives you drag. It's actually the accepted way of doing it. Is that what um Swiss side does? They have this big rig that they have behind. Exactly. Yeah. With a bunch of pitos, like it's a matrix of pitot tubes, I think, or some kind of track measurement tool. Yeah. Okay.
00:19:53
Speaker
Now we're getting into all the different measurements you could do if you need to measure wind speed behind a person. Okay. One option is you take one speed sensor and you just move it around in in the wake and try to sample every different spot.
00:20:08
Speaker
Downside takes forever. Next side is you take a ton of those and we just you build an array. and i looked into a bunch of this stuff also with low cost pressure sensors. They've become extremely good, extremely miniaturized and super cheap.
00:20:23
Speaker
So that was something we were also exploring was this wake rake, how that can be called. That's what he called it. Yes, that's right. Yeah, yeah exactly. um Another technique, now we're getting into the bit more esoteric stuff, is what's called particle image velocimetry. Yes.
00:20:39
Speaker
And the way it works, I have a little three-year-old upstairs, he loves playing with bubbles. We have a bunch of these bubble machines. Brings me back to these days. I used BIV and university a little bit.
00:20:50
Speaker
And you have an industrial size bubble machine that makes a boatload of bubbles. And then essentially you have a camera system and a motion tracking system and you track those bubbles. The trick there is you can make this a lot easier if they go through a laser so they get illuminated. then you have these really cool images of these these bubbles going through a laser.
00:21:12
Speaker
At the time we were recording this, the Escape Collective podcast just came out. Yes, I'm fresh off listening to it I actually haven't finished listening to it. as then And folks, this is a you know an unpaid plug to the Escape Collective, which you should absolutely subscribe to because they do really good journalism.
00:21:26
Speaker
So this is Rona McLaughlin's former guest of this show as well, ah episode with Dan Biggum and the cycling spy on PIV.com. that i I heard about. And then before we started recording, i was like, Ingmar, have you heard about this new technology? And it's like, oh yeah, yeah, we yeah this is something I've played around with a little while ago. so you know, back to you. That was just my little interview. And now also Escape Collective doing amazing work there. And then Dan, of course, is always on on all the really good cutting edge stuff. one hundred Really good seeing Dan using it.
00:21:57
Speaker
And yeah, so this is another way. And there's actually a Dutch paper that came out, I think originally either for cycling or speed skating that they called the Ring of Fire. And the idea there was to use this PIV technology for drag measurements in sports where it's harder to do other things. Like you could do running aerodynamics or speed skating aerodynamics.
00:22:19
Speaker
And there's a few papers. If you Google Ring of Fire, you can find them and you can read them. Thing is, use a super high powered laser the thing in the article in my own experience, we left the optics off at once and it burned a hole into the ground.
00:22:32
Speaker
um Don't do that. I don't know if When I started my company here in the US, I needed to get liability insurance. If I would have said I'm pointing lasers at Olympians, I think I would just not have gotten that liability insurance in the US.
00:22:47
Speaker
So we felt that technology is honestly just out. I talked it through with a friend here as a professor of sports science, ah Jim Martin, who I think you've also had on the podcast. He was, he was our cadence guy. Yeah. He was, he was the crank length guy. yes Yes. His episode is one of the ones that gets cited actually probably in top three or four of our, of all of our episodes. Yes. Awesome. Cadence, crank length, see tight. He's done a lot of the big things, even like 10, 15 years ago. Right.
00:23:14
Speaker
And we just talked recently how, yeah, that's been ah now picked up and how it's been 10, 15 years after the paper came out. But anyway, so we talked it through, I was like, yeah, no way is this ever going to happen in the US. You're just not going to get it legal.
00:23:28
Speaker
Then there's another that technique paper from, I think, Dubai University to just use his phone cameras and colorful lights to do this. um And we just kept looking into all of this, but then we came up with a different approach in all of this and start looking if this at all has any basis in papers in terms of what it does. Mm-hmm.
00:23:48
Speaker
And that is using tomographic ultrasonic velocimetry to do this. Mouthful of words to say. So I think I need to explain what that means. So the ultrasonic, I'm i'm very curious about. the ah yeah So actually, one of our previous guests, Chris Morton, who's a friend of mine, he had done some some of his PhD research was based around tomographic PIV.
00:24:13
Speaker
And he would, the the amount of data you generate, the raw data you generate with this is astronomical. Like it's, he would fill up hard drives in seconds basically. um So it was just incredible how it's, ah how data hungry it is.
00:24:29
Speaker
Yes. Tomographic. And Chris Morton of AeroLab, right? that Or formerly of AeroLab. I'm not sure if there's but still around. Oh, perfect. Yeah. Lots of names I've heard. Some of them, you know, same people we've talked to. Some of them I haven't talked to yet. Love to meet everybody in this industry.
00:24:44
Speaker
So many great people. Yeah. And ah the main thing, though, this this works a little bit different. And I think I got to explain a few of these things. Tell me if we're going too deep into the nerd deep dive here, but I'll just get going and you guys stop me. From what I understand, sometimes it's hard to get a whole like a full understanding is we have some very smart people listening the show. So those of you who are smart and are still listening, thank you.
00:25:07
Speaker
So please, yeah, explain it as if we are. You don't need to explain it as if we're five. Explain it as if we have, you know, college level physics. Yeah, cool. I think we only have to put a few things together and then this whole thing will make sense.
00:25:23
Speaker
The first thing is that a tool that exists that's entirely common are ultrasonic wind speed sensors. It's actually the things that you see on top of sailboats. And the way it works is this idea that speed of sound is constant.
00:25:36
Speaker
So a sound wave travels through the air at sea level roughly 232 meters per second. And it actually travels through the air with that speed plus the speed of the air.
00:25:47
Speaker
Okay. So if the air travels towards you, the sound wave gets to you quicker. If the air goes the opposite direction, the sound wave gets there slower. It's not like the speed of light that gets smushed with with velocity. Okay.
00:25:59
Speaker
Yep. You have that effect too, that you can actually tell the speed of the object from the Doppler shift. um Kind of related, um but actually somewhat separate, is that this is called time of flight.
00:26:11
Speaker
So there are different techniques. You just measure how long this sound wave takes to get there. So you can make a sound and if you know the distance, you can back calculate the speed the sound took there and then you can back calculate the wind speed.
00:26:23
Speaker
So this is how an ultrasonic wind speed sensor works. Entirely very robust accepted technology, especially for super low wind speeds. So basically, if if I'm just to paraphrase it to maybe like you know high school level physics here.
00:26:38
Speaker
Are you taking taking back your initial? ah No, no, I'm not taking it back. I just want to repeat it so that I make sure that I understand it. ah So if you if you are pointing this ultrasonic sensor at let's say a cyclist in, you know, in an airflow um and you know the distance to to the cyclist and you expect the sound to rot to cover that distance, you know, in a certain amount of time because you know the speed of sound. If that time is, ah you know, longer or shorter, you can calculate the velocity of the air between the sensor and the cyclist. Is that is that correct?
00:27:14
Speaker
So, um, there's adjacent technologies to this. There's also, you can use this with an echo. So what you are talking about is like a distance sensor. This is how car parking sensors work, what you're describing. okay And in fact, we actually use these car parking to sensors for this project and they're like two bucks a piece and that made everything. But that's not the technology? Yes.
00:27:35
Speaker
The technology this works is you have the speaker on one end of the airflow and the microphone on the opposite end. So you don't have an echo and it goes back and forth. okay You have them in two different locations and it travels through the air in that location. Think of it as injecting a particle into the airflow and you just measure the time that that particle takes to to get to the event. So that's an easier way to visualize it, I think. Now I'm with you because then, you know, again, i was like, let's say half with it where I said, you know, you know, the distance between the two things. Now you clearly do because the speaker and receiver, obviously some kind of fixed distance.
00:28:08
Speaker
And by measuring, by comparing the expected travel time to the measure travel time, then you can infer something about the air velocity between the the two things. Yeah, exactly.
00:28:20
Speaker
So now we explained an ultrasonic wind speed sensor. The next step, let's explain how a computer tomograph works. And the very simple way to say is that in physics, there's different ways to describe the same thing. And, um you know, you can think of the world as like...
00:28:39
Speaker
ah Once in zeros, you can think of the world as sine waves. This is what's called a transform. There's also another way to think through the world. And that is imagine you have an object, you shoot a bunch of lasers or lines through it from different angles. okay And you look at the absorption rates from this, from all kinds of different angles. Okay.
00:28:59
Speaker
And that's all you get if you think of a ah computer tomograph that we know of. It tends to be an X-ray CT, right? So it actually just shoots a bunch of X-rays from different angles through a body.
00:29:12
Speaker
And then you do a bunch of math with it and out comes a 3D object. And if you put the two technologies together, what you actually get is you get many speakers. So in our case, I think was 96 speakers and you get many microphones and they are arranged in two circles that you skate through or bike through or do anything, go through.
00:29:31
Speaker
And each speaker makes a sound and then every microphone picks it up. And then the next speaker makes a sound at each microphone picks it up and then each speaker makes another sound. And so you have all these sound waves traveling, you know, crisscross through the area behind the athlete.
00:29:46
Speaker
And from that, if you put this into a reasonably powerful computer, you can back calculate the airflow in 3D behind the person. That's very cool. That was like my 2019 project or something like this. I was working with that with two friends. Then 2020 rolled around. Major life events happened. I think we're all familiar with that. um yeah This required being there in person. Other life elements just took precedence and that kind of killed that project.
00:30:15
Speaker
It was amazing time working with two good friends on this. um It's also fun. One's a NASA arrow guy. at The other one is a high-performance compute guy. was a what's fun times. yeah Amazing.
00:30:27
Speaker
So what are the limitations of this? I mean, it sounds like it's it's pretty robust. you can do you you I imagine you can have people move through it, like speed you were saying speed skaters who are very dynamic. What are the limitations of this technology?
00:30:39
Speaker
So it's like two large gates that you have to go through. And um that's the first part. With size comes cost, right? I don't think it has to be crazy expensive because we I mentioned we use $2 car parking sensors for all of this.
00:30:55
Speaker
But still, you actually run into the same issue as you run with the wind tunnels and all of the other stuff. So it's computationally hard. There's a lot of math. You've got to pay these people at some point good money to do all of this.
00:31:05
Speaker
um Then you have one or two of those across the country. People don't want to travel to it. How you going to recoup the money? Like where all of these projects fall down. I've talked to, I've been involved in like wind tunnel projects, building wind tunnels, all of this stuff.
00:31:19
Speaker
Where all these projects fall down is how do you get the money back of what the whole thing cost. And this is just true for all technologies in all these cycling aerospace, because we are such a niche sport, unless you can scale it to many people.
00:31:33
Speaker
So this goes back to the fundamentals. And this is one of the biggest issues. Technically, it's super sensitive to the distance between the two things. um So you have to build this actually out of super rigid metal, which costs money. You have to be super high in your tolerances.
00:31:49
Speaker
And if you have these other elements, um getting random echoes from all of this stuff, so sound distillation, those were other elements. that we had to deal with. We never made it to like production technology. It was still a kind in the speculative phase, but we had prototypes. Fair enough. Yeah.
00:32:05
Speaker
Yeah. You had to control your environment really well with this. So um this is an amazing walkthrough. touched on, we didn't really talk about wind tunnels because I think everyone knows about wind tunnels. so we We talked a little bit about a field testing.
00:32:18
Speaker
ah You talked about a couple of I would say super novel technologies for, for getting aero done outside of wind tunnels and field testing. I think this is the, is a great time to start to introduce what you are working on now and arrow and um probably the best place to start. Cause we've been talking about all these other technologies is where, what were the biggest gaps in, in what's currently available that made you want to, um you know, come up with something better?
00:32:46
Speaker
Yeah, so um after we stopped pursuing this tomographic ultrasonic idea, it was still the issue, okay, we need skaters and cyclists. um We need to get their drag.
00:32:58
Speaker
On bikes, we were pretty happy with it on like a corporate specialized large company kind of direction. okay But on the local bike fitter dimension, I felt none of the tools work economically. So it's really the cost complexity to outcome in all of this.
00:33:16
Speaker
And then in speed skating, DSU was that none of these other technologies work as well. right It's the tunnel doesn't work, power meter not available, got it and so on. And then there was another option in all of this, and that is CFD. CFD is computational fluid dynamics, which essentially means you simulate the air flowing around the body, but you do it in a physically correct way or something that at least approximates in a physically correct way and then is able to get get the drag that way.
00:33:47
Speaker
And um in industry, does this is essentially how most product is being developed that way. The bikes I worked at at Specialized, they were very heavily designed through CFD. In fact, the first bike I worked on, the first time we put it to the tunnel, it passed all the tests and we were done. We did a single tunnel iteration in all of this, but we did something like 3,000 CFD iterations or something like that.
00:34:10
Speaker
Okay. So we've ah before we move on, ah we've talked about CFD on the show in the past, but it's probably been a couple of years since we've touched upon it. So let's give us a ah wow high school level, university level physics primer on what CFD is and how it works.
00:34:28
Speaker
Yeah. So turns out the main concept is reasonably simple. In physics, there's these two big things that are called conservation of energy and conservation of impulse.
00:34:39
Speaker
And conservation of energy just means the energy that you put into a system must go somewhere. And it all has to add up to the same amount. It can't get lost. Right. Impulse for people that are not as much into physics is kind of similar to energy, just how fast is like a certain mass going.
00:34:55
Speaker
And then it can be going in different directions as well. So there's some more to it, but it helps to say it's a related concept too to conservation of energy.
00:35:06
Speaker
I think North American folks who've gone through North American schools may ah may know it as a conservation of momentum. Okay, perfect. Yeah. ah From Germany originally. Yeah, for sure. No, was just, um ah yeah, when you said impulse force, like, I'm not sure i' I'm familiar with this concept, but then you described it. I was like, okay, yeah, we just used to call it something a little bit different.
00:35:26
Speaker
I appreciate it. Yeah, thank you. um Yeah, and it turns out, essentially, if you just say, okay you have a person going through this, here's the speed on the surface of the body, energy cannot be lost, do all the math, how much track is being lost?
00:35:41
Speaker
You can take a computer and you can solve that math equation just like you can solve a ball falling to the ground. How long does it take for the ball to get there? So in a way, it's pretty simple physics.
00:35:53
Speaker
The unfortunate thing in this specific case is that it turns out there's no great solution that people have found, no direct solution to solve this physics, so it has to be approximated.
00:36:05
Speaker
And that approximation, um basically until the whole AI thing came up, was the most compute incentive stuff that you could do with computers. And so this does not run on your phone. This does not run on your home computer.
00:36:19
Speaker
um You need some pretty strong computers, but also computers have gotten a lot better. And so these abilities keep getting better and better. But generally, it is we understand the physics of the airflow.
00:36:30
Speaker
We've turned that into math and then we solved a math equation. And out of this, we get dragged. And this is how Andrew's virtual wind tunnel technology worked as well. Yeah, yeah, it's the exact same technology. the the The solution of the the drag problem was really done, I think, in exactly the same way, probably using the same toolbox, I would imagine. Yes, open phone, right? Yes, exactly.
00:36:55
Speaker
So it was just how we get to that point is really the difference in approaches here. um So what I had previously done, and this was, again, the the problem that you brought up earlier, the scalability, what I was looking at is putting...
00:37:09
Speaker
low cost scanners in people's hands um to try and directly capture the the form. And that was a nightmare. yeah There were variations in the ability of people to scan. Like we we ended up having some pretty good partners who were able to consistently capture scan.
00:37:25
Speaker
But there was no guarantee of quality and the feedback time. um it might be say 24 hours by the time someone scans and then gets the the feedback of like, Oh, your scan didn't turn out well. Can you get your athlete to come back in? Oh no, they live, you know, 50 kilometers, hundred kilometers away. So it's really awkward to, to ask them to do that. So that was, that was the really big scalability challenge that we had. And, uh,
00:37:49
Speaker
you know we We tried a bunch of different things, but ultimately it was just not practical to get it to to scale out. yeah Yeah, absolutely. So I looked into that a few years prior and I came essentially to the exact same conclusion that this can't be done in a practical way.
00:38:06
Speaker
And there's two developments that changed my mind in all of this. One of them is that CFD, because it's based on computers, improves just at an insanely faster rate than anything that's not software.
00:38:21
Speaker
So last time I did a lot of CFD at Specialized is roughly five years ago. Turns out CFD compared to then the same stuff I looked it up is roughly 25 times faster and cheaper wilder than it was five years ago.
00:38:34
Speaker
And that would probably be Moore's law scaling. Yeah, exactly. It's Moore's law and then it's on top of that algorithmic improvements. And now we have AI enhanced stuff. In fact, this is the other economic thing.
00:38:46
Speaker
You still could make our software business work. if it weren't for AI coding, because we can we run this whole thing with two people and it would have taken us way too long to get this done as well. This is little little offshoot of the other thing why this would have been hard to do five years ago economically.
00:39:03
Speaker
And so computers, last time I looked, it would have taken whatever, four hours to run an accurate simulation and just simulation time alone would have probably cost $100 or more.
00:39:15
Speaker
If you then say, okay, the company wants to make some money, then you say, okay, maybe the bike fitter has to charge the person $400 per position or $300 per position. If you want to have a good business, nobody's willing to pay that and the whole thing falls down economically again.
00:39:30
Speaker
right Yeah, you can go you can almost go to the tunnel for that much money. Exactly. And in that case, you know each has its pros and cons, but in that case, I think people are more excited about the tunnel. The tunnel is also a sexier process, right? like oh ah There's a kind of like cachet about, oh, I went to the wind tunnel and I did this testing and now a little bit faster. So i I feel like I have to say this, but there's a saying that I love bringing up.
00:39:53
Speaker
um It's ah related to numerical and experimental work, but everyone trusts the experimental work except the person who did it, and no one trusts numerical work except the person who did it. so Yes, perfect. I've used that quad two before with alignment.
00:40:09
Speaker
And we'll get back to validation. Does it work? What are the limitations? That's definitely part of the conversation we should have as well. um Yeah, but the main point here is, okay, it now can be done in minutes and it can be done for single digit dollars.
00:40:24
Speaker
around it. That was one of the that simulation d analysi yeah actual The actual CFD simulation. And to give you an idea, still our goal was five minutes per simulation.
00:40:35
Speaker
We pulled out all the stops. We optimized this pretty heavily. We're at seven and a half minutes right now. So we're close, but not quite there yet. And this runs on, ah for those of you that are really into computers, this runs on 198 core one terabyte RAM kind of machine. So we just that was the other solution is we just found the biggest computer money can buy.
00:40:57
Speaker
But if you rent them for seven minutes, it's still only only a couple of dollars in rent to get that whole thing to work. You make up for it with the speed. Yeah, exactly. And so that's the first part. The second part is exactly, Andrew, what you found is we started with 3D scanning, but we found that the process was not reproducible and too labor intensive and typically took manual touch-ups.
00:41:23
Speaker
And the manual touch-ups make everything super slow, super expensive, that economically goes out the window again. Yeah, that was exactly my point. was There was a lot of manual intervention. We couldn't automate the process, um or we we tried and were unsuccessful automating the process. So it was um it was something that just didn't, like the technology wasn't quite there. And maybe...
00:41:43
Speaker
Maybe in five or 10 years, the scanning will be sufficient. And there's been a lot of developments in terms of the cost of scanners coming down and the algorithms improving, but it's still it's still a problematic thing, especially scanning something like a bike. Something skinny and reflective is like the the bane of the existence of people who focus on 3D scanning.
00:42:05
Speaker
Exactly. So really hard process. We chose to not go down that route. Wise choice. Then, yeah, or we tried going down that route and then abandoned it pretty quick. I should rather say. Next option was to 3D scan a person in upright pre position and then use gaming technology to digitally bend them in a different position.
00:42:24
Speaker
Cool. um Which has the nice effect that after you've built this, like bend people in different positions, you can use this bending technology to um do your variations around it.
00:42:35
Speaker
So that was a part that we really liked about this approach because it keeps the 3D scan reproducible, right? You use the exact same 3D scan for both of it. And it means you can, you know, make changes in seconds instead of having to rescan another person.
00:42:52
Speaker
Right. So that was one of the insights in all of this that we liked a lot in this. And through some past projects, I have a good friend who comes from the gaming industry. So he's well versed in all this animation technologies, building user interfaces.
00:43:10
Speaker
That was actually another part for us, quite um building this user interface to make it intuitive for people to make these adjustments. The really cool thing that this opens up is optimization, like true optimization, where you can you can do a hill climbing algorithm or something like that too to optimize just what position is the best for a given person for a given set of geometry.
00:43:31
Speaker
Yes, exactly. Optimization and just parameter studies. um Cutting ahead, we've done 1,800 simulations so far in the app. Give you an idea. I tried to do math how many track tests and wind tunnel tests I've done in the eight years prior.
00:43:47
Speaker
That's roughly the same order of magnitude. So in and in the three months we've had this, really been using it, we've done that. A little bit more scalable then, yeah? Yes, exactly. um So, you know, a question I always get is, oh, can you build correlation studies on all of this? Can you build optimization? And you can guess that that's on a roadmap and you can guess that that's things we're working on.
00:44:09
Speaker
Nice. um Yeah. And then the other key part in all of this that finally made this work is that we found this technology to use ah what's called a parametric human model. And the way to think about this is um it's this university technology.
00:44:26
Speaker
It's always a person, it's always a human, but there's 4,000 things you can adjust about this human. So you can make each arm longer, each arm shorter, you can change biceps diameter, you can change quad diameter, you know, you can change 4,000 things about it.
00:44:43
Speaker
And then it comes with an AI tool that allows you to fit this model to just a couple of images and a few measurements and do that in a super robust, repeatable and automatic way. That's the really, the special sauce, I think here, the real divergence between the virtual wind tunnel and and what you guys are doing. I think that's where, that's where it, it comes down to.
00:45:03
Speaker
Absolutely, yeah. And I think, um i I don't know if this existed at the time that I was looking at the virtual wind tunnel. So ah it ah again, timing and luck is a big part of everything.
00:45:16
Speaker
Oh, absolutely, yeah. I think we built on your shoulders, truth be told. And I mean, I saw what you guys were doing. i've Actually, it's great to meet you in that regard and have this interaction, because I saw this like when it came out and was like, man, this is cool.
00:45:33
Speaker
um I wonder if it can work. and I was doing CFD with like human models at this point, but I do think one simulation took me around four hours. Yeah, that was about the timeframe. I wasn't throwing a ton of resources because we weren't we weren't focused on the seven-minute turnaround, like this real-time feedback, if you want to call it.
00:45:51
Speaker
quite I know that's the goal, but um but seven minutes definitely gets you in the realm of having someone in the bike fitters studio. and not leave. But four hours, asking someone to hang around for four hours while they get results is a little bit, it's a big ask.
00:46:07
Speaker
Yeah. And I want to put a pin in that conversation. Sorry, let's put a pin in the conversation of like the the real time stuff, because i I think that's really important. But i I want to get to it and give it enough, enough, app you know, oxygen when we talk about how this actually works and how folks are going to use it.
00:46:23
Speaker
um So we were you were talking about um ah ah scanning an upright human. Is that where the technology is now? So you're you're scanning upright humans? how does How do you actually interact with it? So workflow, this goes back to the key thing we had to solve was it still needs to be accurate, right? we We don't want to give inaccurate results.
00:46:41
Speaker
So that hasn't changed. But beyond that, it has to be easy to use. It has to be fast to use. And it has to be so affordable that people can use it at scale. nice Like all the other ones are simple, affordable, quick.
00:46:54
Speaker
And this is where I feel all the other approaches that I've worked on fell down in the past. Yeah, pick two of three, or maybe, probably actually in aero testing, pick one of three. Yeah. um Yeah, I don't know, affordable, Tetsu, go out the window and everything else, yes.
00:47:11
Speaker
Not in the mix. Yeah, exactly. Um, which is, uh, there's a tangent there. We can go, okay, what are people currently doing? Like copying people or looking at frontal area, happy to talk through those as well. So there are the simple ones, but they always fall down on the accurate dimension.
00:47:27
Speaker
For sure. There's this the slow Twitch wind tunnel, right? Like where you post your position, a slow Twitch and ask people to comment on it. Exactly. Yeah. um Yeah, so our combination was really just parametric human model plus a user interface that is very simple, easy to use, plus fast, accurate CFD. Like those three things all needed to happen to have this be a viable product.
00:47:50
Speaker
And the way it now works is you take a photo of your athlete standing with your arms off to the side, no helmet or shoes. You need height, weight, gender. You plug this in the computer. 12 minutes later, you get that 3D model of the person.
00:48:04
Speaker
just standing, not yet in the bike specific pose. it The next step is to recreate the baseline pose. So you take a photo of the person standing and a photo of the person, sorry, a side photo of the person on the bike and the front photo of the person on the bike.
00:48:21
Speaker
Okay. Then we run some other math on it that's called pose estimation. So we have to computer guess what this posture looks like and take an initial stab at all of this. This gets us quite close to already replicating the pose.
00:48:34
Speaker
Right now we still need some manual labor from the bike fitter to fine tune this pose through, I think at this point, 18 sliders. So there's 18 things we can adjust about the posture.
00:48:46
Speaker
And we try to find a trade-off again between being easy enough to use that bike fitters don't get overwhelmed by throwing a wall of sliders at them and comprehensive enough that we can test everything a bike fitter could care about. So you don't have 4,000 sliders on this. That's.
00:49:01
Speaker
Exactly. Yeah. egmar you showed me the yeah this is more mostly for the the benefit of the listeners but you showed me the sliders and as a bike fiter myself they made intuitive sense you know you were talking about you know ah pad stack and and pad width and and hip angle and things like head head angle and neck angle and shrug.
00:49:22
Speaker
So these are all things that are to a bike fitter are not foreign sort of concepts. You know, you weren't talking about ah really anatomical terms that you needed a ah you know, maybe a medical degree to understand. This was a lot more approachable.
00:49:38
Speaker
Exactly. As I said, easy to use is one of our key things in all of that. We knew that that is one of the main hurdles, other technologies I felt hand met. And I think very technical people tend to assume that that hurdle is is lower than it actually is.
00:49:54
Speaker
If you want to scale this um the bar, if you work on something yourself, everything seems easier to you than to a new person. And especially in software, and nobody has the attention span.
00:50:05
Speaker
that people have in other environments. So the bar to make this really easy to use is actually quite high. And if if I were to say some of the stuff that we were quite hard on is exactly this down to like which words we use to describe stuff yeah and, you know, where we place buttons and what we show to them and all of these kinds of For sure. You got to really understand your audience in order to to go down that path. Yeah.
00:50:28
Speaker
Yeah. Okay. Exactly. Cool. And yeah, so after you recreate the baseline post, the next thing, and I think that's another very compelling thing is we 3D scanned a bunch of the helmets and we're currently 3D scanning all the relevant time trial helmets on the market.
00:50:42
Speaker
And you can virtually put a helmet onto that person. yeah So you don't actually have to have a bring your own helmet or um you actually have the ability to test helmets virtually that you don't have in the real world.
00:50:54
Speaker
Are you able to adjust how the helmet fits? Because I know there's always going to be some preference and there's some variability and you know one person puts a helmet on a certain way and another person might have it sitting further down over their forehead.
00:51:06
Speaker
Exactly. Yes, we do. um That was important. The current helmets we only have with visor, the scans we're doing now are like with and without visor. So you can even decide if you want a visor on or a visor off.
00:51:17
Speaker
And I have to give you kudos for the helmet choices because of the five helmets that you had in there, I own three of them. So I agree with your helmet choices. Okay, perfect. um There is plenty that are missing right now, but before the year is up, we should have all of them.
00:51:31
Speaker
So we'll have the top 20 helmets. We can have the listener guess to what the top 20 helmets are. That might be its own good Christmas bingo game. A giveaway for of a free scan to somebody who guesses all the 20. I mean, yeah you can always just go down the ah you know the the Kona bike count that Triathlon publishes and see what helmets were on the heads of the the folks at Kona.
00:51:52
Speaker
Exactly. And yeah, that's kind of the main part. Then you hit simulate and you can um it spins up at this high performance cloud instance, run CFD on it.
00:52:03
Speaker
The other really nice thing on all about this is we don't only have one of these machines because it runs in the cloud. We have, I don't want to say unlimited, but we have a bunch of them. And the moment you send off the first one, there's a button that's called create variation.
00:52:18
Speaker
And what that does is it loads exactly the same thing in that you ran the last time. And now you can make a very targeted change to all of this and can hit simulate again. Super cool. This change realistically takes 15 seconds, 30 seconds, and you can do many of these in parallel. So typically we tell people you can set up 10 to 15 of those in the seven minutes it takes for the first one to be done.
00:52:42
Speaker
And some turning that around super quick is really useful because you know nobody wants to stare at the wall at a fit studio. And then the other part in this is that the number one source of inaccuracy of wind tunnel testing and field testing is people's inability to hold their position repeatable.
00:53:00
Speaker
For sure. And our digital twin holds a position the exact same way every time. So this is actually in accuracy one of our big ah opportunities. And because we bend the person, we don't have any scanning artifacts where there's like the calf diameter is different on the one scan than on the other one.
00:53:18
Speaker
We have the exact same model, just having to have one centimeter lower if you want to have to be one centimeter lower. I love this because I'm listening and and going down a trip down memory lane thinking, like these are all the arguments I used where it's exactly the same model every time. And the difference that you're seeing in the results isn't the difference on how someone gets on a bike. It's it's the difference in the actual performance of the position.
00:53:41
Speaker
um Now, whether they can replicate that in the field later, that's a problem for that person. But yeah. But that's a double-edged sword because at the end of the day, what we care about, it like as an end user, what you would care about is, does this work in real life? does this is is this Is this representative? So I think it's a conversation that's worth having. And I know, Ingmar, you and I spent a bit of time talking about this when we first spoke, but I think it's a conversation worth having.
00:54:04
Speaker
And that is how ah reproducible are the results that we we see using a product like yours to raced it. You know, like I keep saying I'm in, I don't know what, five days I'm going to be racing in, uh, in Arizona.
00:54:21
Speaker
I did some field testing. i did a bunch of runs. It was pretty, well, kind of a long day, but not a long day by aero testing standards. Um, you know, ah definitely not five hours long. How would you address that issue of, you know, fatigue sort of degrading position, um but also other variables that you would encounter in the field that may not be captured in a fit studio?
00:54:43
Speaker
Perfect. Yeah. I think there were two questions here. One is on the what's the repeatability of the tool itself. And then what are all these other elements of this? To speak to the second question first, initially we rolled this out for empty users.
00:54:57
Speaker
And I'm always a person to say, okay, let's just get this out to everybody. This again, thinking, oh, this is super easy to use. Everybody should be able to use this. Democratize error testing. Exactly. We've heard that before. It's great.
00:55:09
Speaker
That was the plan. That's great. And we can actually see which positions people run. um So with pro teams, for those that worry, I talked to some of the two different teams that are always concerned.
00:55:21
Speaker
We can make sure we don't look at the kind of data. So and we generally do not share this with other people. sure But if we need to look, we can see what people run. And that gave us a little bit pause for some of the people.
00:55:34
Speaker
And I remember actually, you know, the first time I was in the wind tunnel, the first time other people in the wind tunnel. It is very common, even for a bike fitter, the first time they're in the tunnel to build positions that nobody in their right mind can write.
00:55:49
Speaker
And people get... Yeah, the CDA is amazing, right? They're not they're actually not fast if you can't pedal or if you just flat don't hold the position. But the CDA a in the tunnel is very low. yeah um And this is a common path to get wrong.
00:56:06
Speaker
And so we decided that this does need training. This does need education. And it does need people that are more still focused on the holistic side of you got to be able to look up the road. You don't want to get injured.
00:56:18
Speaker
You want to have the range of motion sites. And so we changed our path. And so we're only selling the technology to end full time bike fitters and coaches nice at this point.
00:56:29
Speaker
And it comes with a four hour training. I just did the four training, give the class today. um One of my college thesis was actually the bike fit guidelines for the German Cycling Federation. so I've done this training before.
00:56:42
Speaker
um So, you know, we really know that getting the fitters trained on what are the best things to try, the best things to test, what to do, what not. It's actually almost as critical as the app itself.
00:56:53
Speaker
And we feel a lot better about this partnership than just putting this out in the wild. And just as ah just as a side note, I mean, i mentioned before and folks who have listened to the show know that I've i've spoken to a lot of folks in this space.
00:57:06
Speaker
And to my knowledge, none of the providers service providers of technologies that do... Similar things, not the same as yours, obviously, but similar things have that sort of training, right? So like, okay, you can test, you can go out there and maybe if you're very good at it, go out in the field and get some CDA numbers, but then like, so what?
00:57:26
Speaker
So then what do you do? How do you increase your likelihood that whatever you change is going to improve the end result? Like that is still very much, you know, ah almost like the art of of of bike fitting, aero bike fitting.
00:57:41
Speaker
Yeah. And we recognize that what our end customers want, why they come to a bike fitter is not the tool. They want a faster position on race day.
00:57:52
Speaker
And they want a better position on race day. So in a way, how we provide a faster and better position on race day isn't really the super important part, but that needs to happen. And that turns out is a great combination between education and just, as I said, 15 years of experience in this across wind tunnel track and everything.
00:58:11
Speaker
And some academic studies on bike fit, bike fit there as well. And the app, there are some simple rules that tend to work pretty reliably. And then there is a bunch of rules of like things worth testing. cool And of course, combining things worth testing with a testing tool is it is a good combination. You kind of need both of those elements in order to be successful.
00:58:32
Speaker
Yeah, exactly. Okay. So, um so you have this tool that is aimed at fitters and coaches. um It looks really easy to use based on what kind of what I've seen. um It doesn't require anything super special. Like it doesn't require a scanner, like what Andrew and I were doing back in the day, just your, I guess your, your iPad or, or similar phone to take the photos and to move the sliders around.
00:58:57
Speaker
Um, And you mentioned that that you can do multiple simulations. Now, my question is, can those do those simulations run sequentially or are they run in parallel?
00:59:08
Speaker
They run in parallel. So when you're saying seven and a half minutes per simulation, you can launch you know multiple of them ah concurrently? Yeah. You can launch one every 30 seconds, roughly. Yeah. faster, but I'm saying it takes around 30 seconds to do a variation and send it off.
00:59:24
Speaker
Got it. This sounds very promising. um But as I've said, as I've said before, we've had lots of folks on the show, you know, talking to us about the sliced bread of, of Aero testing.
00:59:35
Speaker
um Let's talk about limitations. Where, yes what are your, um you know, in your experience and then also based on your, your deep knowledge, what can Aero not do or not do well at present?
00:59:48
Speaker
Great point. um Joke always that the C in CFD stands for colorful fluid dynamics. It's very easy to get pretty pictures out of it yeah and pretty pictures that are somewhat right, but getting pretty pictures out of it that are actually physically correct is a lot harder.
01:00:06
Speaker
Andrew's nodding for listeners who are yeah who obviously don't have the video. yet um so Before we talk on the limitations, I want to talk a little bit more just about the process on how we got to the end result, how we know about the limitations and what we know and what we don't know. For sure. sure Showing your work is important.
01:00:24
Speaker
I love Even more in the limitations. First part in this is, okay, um you know I've been doing some CFD for a while and then I reached out to a friend of mine ah Dr. Mitt Williams, former Lead CFD Engineer, McLaren F1.
01:00:39
Speaker
Back from when Specialized had a McLaren partnership, that's where the connection comes from. oh right But um you know from last I heard, Formula 1 teams are pretty good at aerodynamics. McLaren's one of the better teams around.
01:00:51
Speaker
they have ah They have a lot more dollars to ah to invest, I think. Yeah, they've been doing CFD at a bit bigger scale than I have. So he's been a tremendous help and he's actually been a big partner or like really lead in leading the CFD team.
01:01:04
Speaker
algorithm development while i did more of of the bigger picture side. So, first part second part is really we took our starting point in academic literature. There's Professor Bert Blocken at formerly TU Eindhoven, now I think he's in the UK.
01:01:20
Speaker
He released a big papers on can you do CFD on flow around athletes. Happy to share these articles with anybody. If you look block and airflow athletes, if you Google that, I think you'll you'll find those papers.
01:01:33
Speaker
For those of you that want to take a deep dive and see, okay, what's the academic stance on can this be done and how could it be done? And we use that as a starting point for our own approach. And then the third part in all of this is that we've now done two rounds of wind tunnel testing ourselves and have done you know a bunch of position comparison.
01:01:53
Speaker
And then just heard of a partner who did a bunch of validations as well. um Don't have that data yet, but hope to get it soon. So we've done our own wind tunnel validation. So what is the conclusion in all of this?
01:02:06
Speaker
First thing is like, okay, what's our, you have to think about repeatability and accuracy in wind tunnel senses, in CFD sense, different than in a wind tunnel. Okay, pow. The good news in our case is if you run the same thing twice, you will get the exact same result.
01:02:22
Speaker
Well, that's not necessarily true in a wind tunnel.
01:02:27
Speaker
so you know I feel that's a thing that's often not discussed is how much repeatability can you get out of a wind tunnel. all And the number one source of error there tends to be rider movement.
01:02:38
Speaker
And this goes back to what I was saying earlier about everyone trusts a wind tunnel except the person who did it. So there's repeatability issues. um It's relatively small, but it exists.
01:02:49
Speaker
and And the rider is the... and i I always said, like you've got a big squishy person that you're dealing with that's not... you know not a machined part that is going going to hold the exact same position.
01:03:01
Speaker
Exactly. And the amount of people in the industry that don't even run repeats, so they wouldn't even know what their repeatability is, is little bit mind boggling. And hopefully that'll change over time. So general rule of thumb, other people might have different values. Those are my values, right? This is not not super rigorous. But from my end, I say that the absolute best in the world, very experienced time trial test writers, good time trialists, world champion pursuiters,
01:03:25
Speaker
And so on can hit repeatably half a percent error in a wind tunnel. your Your standard issue world tour time trial list, so it still a pretty good time trial list, um you know probably far beyond the average person, or you're very dedicated error nerd that has run very many repeats and all of this stuff has done the error testing, I think can typically hit around 1% repeatability in the tunnel.
01:03:49
Speaker
And then from there it goes downhill. So I think one and a half, 2% for like a novel person that's never been to a tunnel isn't uncommon. If you have, the more you go into um amateur entry level,
01:04:03
Speaker
Triathletes are people that never ride their TT bike unless it gets low to them. It's not uncommon to get two, or three, four, 5% repeatability in the tunnel. Unfortunately, once you have 5%, the value of the tunnel session goes pretty close to zero. yeah And i've I've had athletes in the tunnel where we had to say, look, we can't tell anything apart. We're really sorry here.
01:04:24
Speaker
um But this is to say which value you have um depends a lot on on the person you're dealing with. And in that case, to cut a little bit, our mean error, so our error of IRO specifically, to our knowledge is around 1.5%. Okay.
01:04:44
Speaker
Between what and what? Because if you're saying that you're, you know, if you're modeling the same thing, you're always going to get the same results. So i imagine the error there is zero or the variability is zero. so then Oh, yes. The mean error between the wind tunnel and CFD from our wind tunnel correlation studies.
01:05:00
Speaker
And that is accounting for the wind tunnel error as well. So there are some stats you have to do and say, okay, if my total error is this much and the wind tunnel error is one, one and a half percent. What's my right?
01:05:10
Speaker
Because you can't assume, oh, i ran a wind tunnel simulation. I ran this. The two don't match. Default is 100% on. CFD or default is 100% of the tunnel. You have to do some stats to figure out, okay, which is which. that makes But our mean error is around one and a half percent.
01:05:27
Speaker
So for our very high end partners, like Olympic teams, World2 teams, it is absolutely true wind tunnel is still more accurate. And in that case, we also actually recommend Iroh more as a pre-screening tool to test many positions and then validate the final And that was going to be my comment. You just beat me to it. Mine as well. yeah yeah Yeah, you can optimize. you can You can do a parameter scan or parameter sweep beforehand, and then you have more valuable time in the tunnel as a result.
01:05:54
Speaker
Absolutely. Probably too long for today, but another topic I can talk for a long time is one of the things I do is innovation strategy consulting with the US Olympic team and with some... um of the national governing bodies in the US.
01:06:10
Speaker
And we're talking about what we call our Pareto gains philosophy in contrast to the marginal gains philosophy. And big part there is like a goal on experimentation on search and searching much broader and in weird spots and in spots where you didn't look before.
01:06:25
Speaker
So I like tools that with very little cost can allow you to explore things. And i'm my tool selection in the past from my work has actually generally not been to pursue higher fidelity, but bigger variation studies, more parameters or broader search spaces.
01:06:43
Speaker
That's very cool. I feel many people in Arrow are towards like deeper and deeper taping bolt holes, looking at things. I'm more into like broader and broader looking in weird places, trying different things. i'm The reason I'm laughing is because, again, listeners, you don't see, there's ah there's a tri bike in the background here and and I'm getting it ready for for Arizona and I was taping bolt holes earlier today.
01:07:06
Speaker
Okay. so So yeah yeah you've ah thank you very much for that call out, Igmar. But um yeah, I love where you're going with this and sort of where this ah makes me excited is the fact that we keep seeing people go faster, right? And becoming more aerodynamic. So there's still...
01:07:27
Speaker
there's still a lot of, so you know, optimization to be had. And we're not necessarily talking about marginal gains. Like there's so many things in cycling and in triathlon that has hit the, you know, the level of marginal gains where the the increments are just very, very small, but you're still seeing fairly substantive changes in position, for example, and in in equipment. Yeah.
01:07:49
Speaker
Yeah. From one year to the next, even now. So a tool that allows you breadth versus depth, I think, is is really interesting because, heck, we might find something really, really totally new or maybe somewhat new and somewhat exciting from this stuff kind of stuff. Well, a good example of something that's kind of outside the normal environment.
01:08:08
Speaker
ideas for bike fit is the Superman pose, the the classic Graham Obrey pose that he used. And that was super fast, but that obviously with a normal optimization, you wouldn't likely get to that point.
01:08:22
Speaker
That's, that requires some external inspiration. Yeah. Yeah. And um I want to get back to the accuracy validation imitation thing in a moment, but just to throw this out here as a thought piece of why I think breath is so important in this.
01:08:36
Speaker
A world-class CDA by LA o Olympics 2028 is sub 1.5. what on What? In what discipline? Just, okay, track or like world-class time trialist. Even like road time trialist 0.15?
01:08:50
Speaker
Yeah. Holy smokes. And um this is just of a linear interpolation of the trend of the last decade. So that's wild the question you can ask is if your goal is to get to sub one five, how many bolt holes do you need to type over to get there? You need to drill more holes so you can tape them over. like Exactly.
01:09:10
Speaker
Like the marginal gains just don't add up to those amounts of gains that we get. yeah So we have to look in places we haven't looked before and we have to be more creative. And I guarantee you there are still things.
01:09:22
Speaker
If I know where they are, I would have already done them. but And some of them we have figured out and we're about to do them or will do them. Very cool. But by and large, there's a focus on search more than driving key fidelity. And that's one of the other things in design decisions that we've made.
01:09:37
Speaker
Going back to limitations though, one of the big limitations we have to speak through is simulating texture. And this is still um paper came out again, Professor Blucken, that this can be done. I've doubled in this in the past.
01:09:51
Speaker
currently not ah included. So if you wear Aero socks, if you wear one of the high-end textured skin suits, this is something that we're currently not accounting for. So you're wearing a skin tight, smooth skin suit in IRO, you're not wearing one of the high-end texture ones, essentially.
01:10:09
Speaker
Probably one of the current largest simplifications. Yeah, and that's something that even for academic studies is quite difficult to capture. Mm-hmm exactly. Truth be told, a high-end skin suit can almost cut the upper arm drag in half.
01:10:23
Speaker
So of course this shifts um this shifts the importance of the upper arms in CFD, which skin suit you're wearing. But the amount of athletes I've worked on that change their position based on which skin suit they're wearing and skin suit position interaction puts you again into the very pointy field of people. Totally. This is this is the Dan Biggum stuff.
01:10:43
Speaker
Yeah. Dan was one of the people I was thinking in I remember so he was he was on the show. He's been on the show a couple of times, but I remember talking to him about like, oh, will your choice of helmet dictate the optimal skin suit you have?
01:10:57
Speaker
he goes, yeah, probably, but it's not a practical question because for almost everybody who is going to be out there, except for the you know handful of people for whom they have the resources to do all this testing, it doesn't matter that much.
01:11:11
Speaker
Yeah. um My personal view is if you look forward on 2D, if you want to do the, to the France level or Olympic stuff, this interaction will become more and more important. I believe it. But yeah, for, for the people that we are like, goes back, let's remember the big picture thing. If you haven't worked on your position at all, you haven't done any error testing yet. This is not something you need work.
01:11:34
Speaker
but we want to be transparent about what their limitations Well, I think there's yeah there's two groups of people you're going to have. The people who pay for the app and pay for the the service, and then the people who get it for free because they're they're at such a high level.
01:11:48
Speaker
um And even if they were paying, realistically, like 90% of your population that you're dealing with is the the people who will see these big gains anyway. So I think it's ah maybe not as as relevant to to worry about that.
AI in Aerodynamics Testing
01:12:04
Speaker
Yeah, well, and as I said, even on the high end, and I work with some of the more high end people out there, um even there having a broad search tool in addition to a high fidelity tool is, I think, a really good mix of tools. And if you have a $30,000 a year wind tunnel budget, then keep going to a wind tunnel, i'll totally keep using it, and then run a thousand simulations. had one client where we ran 150 simulations per person for it Oh, wow.
01:12:32
Speaker
And um we, this is kind of what Iroh enables on the high end is, you know, before you tested one bar width, we just test every bar width. We test every bar width, elbow in, elbow out. We test every head position.
01:12:46
Speaker
We test every helmet in every head position. If you wanted to do that for helmet testing, you'd go pretty crazy doing that in a tunnel. And so just this whole thing that... Yes, fidelity and accuracy also matters.
01:12:59
Speaker
But if we just see around, okay, in a helmet test case, is it more important to drive up the accuracy of this test? Or is it actually more important to recognize that the helmet position you tested the helmet in is actually not the helmet position you're riding in.
01:13:15
Speaker
And you may want to test the helmets in more than one helmet position. um which again, you start hitting feasibility limitations in in tunnel testing to test five helmets in four different positions.
01:13:26
Speaker
There are people that have those kinds of budgets, but there are not many. So there's one question that... um I had wanted to pursue, but I didn't have the ability to do it at the time with the virtual wind tunnel. Maybe this is on your roadmap already.
01:13:40
Speaker
um And if it's not, I think it'd be cool to look at. ah But basically, you're you're going to have at some point in the future, this source of a huge amount of data, um different biometric data for people. You'll have different positional data, CFD results.
01:13:57
Speaker
Can this all be used to train people AI, basically, so that you have instantaneous feedback where you say, my biceps are, well, you may not even have to say, you can just take a picture and then, you know, my biceps are this size, my belly is this size, and then you move the sliders, get to the position you're in, and then instantly get, this is your predicted CDA, and maybe seven minutes seven and a half minutes later, you get back a validation of it.
01:14:22
Speaker
um So maybe that instantaneous feedback through AI is is possible. Yeah. yeah Yeah, absolutely. The spelling in our app is not because we just wanted to jump on the whole trend. So wrote AI is kind of highlighted in this.
01:14:38
Speaker
um Of course, the way we use artificial intelligence or statistical methods, if we just want to use all of this, is in how we recreate the baseline pose, the pose estimation. We do a bunch of behind the scenes stuff already.
01:14:52
Speaker
um Next thing is actually that we use AI, how we initialize the flow volume and the initial gas of the CFD simulation. And the main reason there for what you spoke about, Andrew, The AI tool would be great if you can get it to a point that it's accurate enough. right The value of that part, this pure AI guess, is basically not there, not there, not there until it's there. right and We don't know how much data this takes, and these things start costing like $100,000 plus.
01:15:21
Speaker
We're still a small business, so $100,000 is a lot to run on a compute training run to afterwards see that this wasn't worth it. I guess it's the whole chat GPT analogy where it wasn't good enough until yeah all of a sudden it was. And now it's incredible what what these general AI prompts can provide. um So maybe you know once you have enough training data, maybe all of a sudden things will click into place and then you get this super accurate non-CFD based.
01:15:50
Speaker
at the risk of sharing um our strategy a little bit too far. But um the other strategy is that you actually use the AI model to provide an initial guess to the CFD simulation.
01:16:02
Speaker
And the nice part there is um basically any guess is useful. And you can already, even like a 50% error is useful because you have a starting point and the simulation will converge quicker.
01:16:15
Speaker
And essentially in this case, you can just get the simulation to run faster and faster and faster until at some point you don't need the simulation at all anymore. So in this way, we can actually make use of the AI tool even before um it reaches a consumer accuracy because it just provides the initial guess for the CFTC. So if I understand correctly, you're saying that you can get to you may be able to get to a point where just by looking at your position, once you've adjusted all the sliders, you can get an accurate CDA without running the CFD simulation?
01:16:48
Speaker
Yeah, we will definitely get there. That's cool. The question is, do we need 10,000 those? Do we need 100,000 of those? Do we need a million of those? um For reference, our model currently predicts we will have tested 200,000 positions by the end of 2026. That'd be a lot of wind tunnel time.
01:17:06
Speaker
um Which is, I think, maybe more than all wind tunnel tests in the history of aero testing and all of cycling have been done. We will have tested more positions virtually um next year than all humanity has tested together till 2026.
01:17:24
Speaker
So one of the cool things of digital technologies, right? And I think, well you know, this is the other thing, cftd got 25 times better in the last five years.
01:17:35
Speaker
I don't see that slowing down, especially with all the AI push on on chip improvements right now. In five years, this will be 25 times better again. And wind tunnels will be roughly as good as wind tunnels will now. So just ah betting on digital technologies.
01:17:51
Speaker
We have a bunch of additional things already. We have a peddling model. We have a bunch of these other things. It's not that we can't do it. It's, again, that the economics are the limiter. But at the speed at which digital stuff is happening, um the next couple of years will be pretty exciting on this too. Okay. So one other question I have for your roadmap, um are you planning on testing different yaw angles?
01:18:14
Speaker
Is that something that you would like to look at? um Yeah, great question. And this goes back to this economics debate. So we're currently capable of doing this, but it's not a feature we have enabled India.
01:18:27
Speaker
And the main reason is that this is five times more expensive than running just a single yaw angle in this. If there's enough need from our fitters, if we get enough feedback around, hey, we want the Yaw Sweep and we have enough customers to justify this as an additional feature, we can turn that feature on.
01:18:46
Speaker
A little bit my perspective in all of this could be controversial viewpoint. Different people have different perspective there. Even people that I highly respect have a different opinion than mine. and One of my jobs was to drive all the Tour de France racecourses with an instrumented car and get actual yaw measurements and check what the yaw measurements were.
Challenges with Yaw Angles and Crosswinds
01:19:06
Speaker
So I come much more from the road side. I know Ironman Hawaii is a little bit of a different beast.
01:19:12
Speaker
So I might be biased a a little bit more to the cycling side. But generally the most common yaw angle is zero degrees. The next one is five degrees, 10 and 15 outside of Hawaii. Hawaii of course being a pretty important one is exceedingly rare.
01:19:27
Speaker
Five degree yaw performance highly correlated with zero degree yaw performance. um This goes back to Pareto kind of style. I think just zero gets you very far, but we could enable these higher fidelity elements if the interest is there.
01:19:43
Speaker
And the other comment is that the faster you are, the less yaw you're running. Like you're infinitely fast. You would be be at zero degrees all the time. Yeah. And for perspective, my thesis, my master thesis in university was the handling of wheels and crosswinds.
01:20:01
Speaker
and built this like dynamic study around aerodynamics, vehicle dynamics, and human in the loop control. And so I looked a lot on um how how would do deep wheels handle in crosswinds.
01:20:12
Speaker
And the reality that I felt was, okay, at a point where 10 and 15 degree yaw is so common that you get a substantial benefit from a 90 millimeter wheel, most people I've talked to are not willing to ride a 90 millimeter wheel.
01:20:28
Speaker
Yeah, that makes sense. Because I mean, like, yeah, what is what is, who's going to do napkin math for me? 15 degrees EY, if you're going 40 kilometers an hour, is what kind of wind speed if it's directly from the side? I'm sure it's easy enough to.
01:20:41
Speaker
That's Chad TPT, but it's a pretty stiff breeze. Yes. And the other thing in all of this is that you know there's a distribution to the wind and to get 15 degrees yaw 10% of the time or 20% of the time, enough just to be useful, you're like 1% wind gust, you're 1% wind gust is not fun at all.
01:21:03
Speaker
in most gusty conditions. And this was a lot of my work is okay. In handling, you optimize for the 1% of the time wind conditions. And in in drag, you have to optimize for you know the 10, 15, like the the bigger parts of the bulk curve and getting both to align at the same time. Again, people I respect have different perspectives there, but this is my perspective. Even when I worked at Specialized, you can see the rapid wheels were last wheels I worked on.
01:21:31
Speaker
and the design philosophies were pursued there. Very cool.
Conclusion and Future Connections
01:21:35
Speaker
So Ingmar, I think this we've already spent more on this conversation than I kind of have a a bit of an allergy to really long podcasts. Andrew and I have joked about this in the past.
01:21:47
Speaker
So I think ah this is a really good place to wrap up. We're about the hour and a half mark. ah We covered so much ground and I know that i I'm going to say this now. I think we need to have you back on because there's also other stuff that maybe we can touch about after we stop recording and then have a proper recording of of that you're working on or at least thinking about, which is super interesting for us.
01:22:10
Speaker
But um I mean, thank you so much for the the the high school slash college physics lesson. I know I learned a bunch of stuff that even though I've been playing in the sandbox for a while that I didn't know about,
01:22:22
Speaker
um I think you what you have is ah is a super compelling offering, which I kind of wish I was still doing fitting professionally so I could take advantage of. But I'm ah no joke. I'm already like talking to some of my local bike shops. They're like, hey, folks, do you want to ah partner up on something here?
01:22:39
Speaker
Because i know i know a technology that could be super useful to to both of us. um So thank you very much for the time. um And, you know, I wish you all the success in the in the coming months and years.
01:22:52
Speaker
ah If folks want to reach out, if they want to learn more, maybe this is something for them. What's the best way to get a hold of you? yeah So first of all, thank you so much for having me on board here and on the call. It's been a ton of fun.
01:23:05
Speaker
Love these chats. Happy to talk about it. I think you can tell. I can talk about this all day long. If people want to pick my brain, talk about stuff, yeah, happy to talk. The technology is called AIRO, A-I-R-O dot app, A-P-P. That's the website of this.
01:23:23
Speaker
Check it out. We have some cool flow simulations on this, which you run a GPU accelerated smoke simulation in the background of a website. In terms of fun little tech things. um Yeah, check it out. Aero.app. Reach out to us. If you're a customer that's looking for a bike fit, we can connect you to fitters. If you're a fitter, reach out to us.
01:23:42
Speaker
Appreciate it. And yeah, if you want to reach out to me, it's Ingmar, I-N-G-M-A-R. at iro.app. Sent me a message, ideally even better through the form. ah Yeah, the outreach last couple weeks has been incredible. We're honestly booked wall to wall. If we get a little bit slow getting back to you, we apologize, but we want to get back to everyone. So it's been it's been quite the ride. It's fun.
01:24:08
Speaker
So in light of this, thank you, an extra thank you for for making the time for us today. And Andrew, thank you for also making the time for us today because I haven't seen you on a podcast in, i don't know, probably six months now.
01:24:20
Speaker
It's been a little while. I think since the last one we recorded was Texas. Oh, after texas okay. Yeah, it's been while. that would have been like May. Yeah. Well, this this was good catnip to get me back. This is definitely right along the lines of my own passion. So i'm from from a personal note, I just want to say I'm glad to see someone else pursuing this technology. And I didn't have the ability and the ah the technological background to to bring it to fruition. So I'm so glad to see someone else is doing it.
01:24:50
Speaker
Well, I appreciate it. As I said, we build on your shoulders. So thank you for pushing the ball forward. We hopefully pushed it a little bit forward still, and maybe others will push it even further.
01:25:00
Speaker
So, and yeah, love to pick your brain even further in all of this. I'm sure you've had some painful learnings of... Things that are still in our future that we could maybe cut us a little bit. Many painful learnings, yes.
01:25:13
Speaker
That's fair. That's just how this goes. Okay, folks, as always, if you like the show, give us a rating or review, follow on Endurance Innovation, and we'll talk to you soon. Thanks, everyone.