The Vertical Space

#96 Chad Sweet, ModalAI: Small, smarter, safer UAS - made in USA

Luka T Episode 96

In this episode, Chad Sweet, co-founder and CEO of ModalAI, joins us to explore the evolving drone landscape and what’s driving the next wave of innovation. Chad shares his perspective on the rising importance of FPV (first-person view) technology - especially in military and public safety applications - and why user experience, including ease of flight and high-quality video streaming, will be a key competitive edge going forward.

We also dive into ModalAI’s strategic partnership with Qualcomm and the company’s focus on developing NDAA-compliant, AI-powered components that enable autonomy and advanced perception in small UAS. Chad offers insight into ModalAI’s product development philosophy and how the drone industry has matured over the past decade.

Chad:

I think user experience is going to be paramount if we wanna be competitive. And then the enablers of that will be flying well, flying easily, streaming great video. High quality video and easy to fly are gonna drive that user experience. And if it's new technologies, what new could pop out of all this in the next five or so years that isn't quite there today? I think the fusion of FPV, virtual reality, and photogrammetry. Something's gonna happen there where you're putting goggles on for FPV video, very similar to the goggles and maybe even identical to the goggles that people are putting on for virtual reality. A lot of this data captures going in. You've got Gaussian splatting and you've got a lot of these new techniques that are making really compelling 3D models and renders based on the data being collected. So flying FPV, collecting data, generating 3D models, re-rendering them back into goggles that are already on your head. There's something there, I don't exactly know what it is, but I think that's like the far flung use cases that are gonna happen at some point. There's so much going on in those three areas that overlap. The Venn diagram has a lot of overlap of similar capabilities.

Jim:

Hey, welcome back to The Vertical Space and our conversation with Chad Sweet, co-founder and CEO of ModalAI. He's an impressive industry leader and I'm sure you'll love the conversation, and thanks for joining us, Chad. So listen, based on your recommendations to keep our introductions shorter so you can get right to our guest conversation. We're gonna keep our introductions a lot shorter allowing you to get to Chad's conversation even faster. We really appreciate your feedback. If you have any other ideas about the podcast, don't hesitate to email me at jim@theverticalspace.net. So let me tell you a little bit more about Chad and then we'll get to our conversation. As I mentioned, Chad is the co-founder and CEO of modal ai. It's a San Diego based leader in NDAA compliant Blue UAS, and AI powered VOXL ecosystem with over 20 years at Qualcomm, where he held 15 patents and spearheaded computer vision and wireless communications projects. Chad led the development of Snapdragon flight, the onboard computer that powered NASA's ingenuity helicopter on Mars and his guidance modal AI has secured more than$18 million in government contracts and forged partnerships across defense and commercial markets. All in service of delivering smaller, smarter, and safer UAS solutions proudly manufactured in the USA.

Luka:

Chad, welcome to The Vertical Space.

Chad:

Thank you very much. I'm really excited to be here.

Luka:

We, are as well. We start with, is there anything that very few in the industry agree with you on?

Chad:

We'll see if this is a good take, but the FAA was correct to be slow to approach, UAS integration and maybe they haven't been too slow at all. When we first started looking at UAS in 2014, we thought we were six months away from drones dominating the skies, and, drones flying everywhere. And it turned out to be all much more difficult than anybody would've thought. I think we're ready now. I think absolutely the technology is ready now, and I hope we don't pursue, or I hope the FAA doesn't pursue traditional airline approaches because this definitely needs a different and likely smaller, lighter and more nimble approach to integration. But we definitely weren't ready 10 years ago that original schedule I think the FAA put out was like a five year plan for these test events. And we thought, oh my God, we're gonna be, we're going live in six months. And that was absolutely not accurate.

Luka:

Okay. That's an interesting take. Is that primarily rooted in the immaturity of the companies that were interacting at requesting, approvals from the FAA or the maturity of the technology itself or something else?

Chad:

I think it was maturity of the technology and the lack of experience that basically everybody had. I think there was this rush and there still is to some degree, a lot, what's next after the smartphone. That was certainly, Intel and Qualcomm. I was at Qualcomm for a long time. That was the approach is like this is the next, one of the next big things after smartphones and how do we, we know everything now. All we gotta do is put some propellers on it and go, and, this stuff is complex and hard and takes some real knowledge in a disciplined approach to making it successful.

Luka:

So how do you assess the industry now

Chad:

now I think a lot of that is happening. I think companies are looking at this as, they're small aircraft and their approach is much more robust testing and, reliability in terms of manufacturing. They're not just slapping it together like a smartphone. They're treating it like it has to be reliable and robust and putting it through its paces and make, lots, thousands and thousands more flight hours. I'm probably, tens or hundreds of thousands of more flight hours have happened since then. And people understand where the failure points are and have addressed a lot of those that were just unknown at the time. The big problems back in 2015 were batteries coming loose. that's not good.

Luka:

So as Peter would say, where is my part 1 0 8?

Chad:

Yeah. Yeah, exactly. I hope it now, I, now I do think we're ready now. I think we're ready. I,

Peter:

When are we going to see Part 108.

Chad:

Oh boy. it just depends. There is definitely gonna be a balance of, certainly there has, you have to lean into it. In the end, you can only be so ready until you actually do it. A lot of this comes from the need to do the work. Really actually go fly and do these automated missions and automated management and understand the failures. And so certainly, the hope would be that they enable it to start happening. I think there's ways that they could do it that are so low risk though, over bodies of water and so forth that really are really low risk. And then I think the argument would be, they've also done a lot of part 1 0 7 waivers to enable such, so those things too. So, there's probably this transition period, I think everybody still would hope that it would go faster. And, now that there's so much technical readiness, it probably is time to get going. But I mean, it's indirect answer to your question as to what'll actually be in there, but hopefully it allows for the next big wave of advancement. I think part 1 0 7 did there are now whatever, a couple hundred thousand drone pilots that fly professionally and are doing so lawfully and that certainly came through because of part 1 0 7. So I think that next, there'll be a next big jump and big growth opportunity when part 1 0 8 is released.

Peter:

Well, it's been written for a while, so you know, it's gone through I think a lot of a lot of thinking and a lot of iteration. We just wanna see it.

Chad:

Yeah, I think there's a lot of people with you on that.

Luka:

Yeah. No kidding. All right, Chad, well, let's maybe start by chatting a bit more high level on the industry. What have you seen lately in terms of industry developments that caught your attention that you're particularly excited about?

Chad:

So, we're certainly focused on Western components and us based specifically it, it's all happening. what's exciting is it's happening now. The, I'll probably end up speaking out of both sides of my mouth a bunch on this topic because, it's all happening. But is it, we're not there yet either. There's, still lots more work to be done, but now if you track, DIU blue UAS framework, so, efforts, which we were a part of since I think we were basically the first it to be a part of that. So that's a, vetted NDA compliant or u I don't think it has to be US made, but NDA compliant components for UAS, if I looked on that page this week and there's, I think, hundreds of components on there. And a year ago, 18 months ago, it was five, I think. So significant growth in that. And now is that enough to build a, super cheap UAS? Probably not yet, but there's enough there to get going. And you have, once you start putting it all together, the volume, the demand, those, the flywheel begins and costs come down and capabilities increase and reliability increases. And so that's really starting now. it's really happening.

Luka:

What about on the end user side of things and the use cases, which use cases do you see, ramping up or, do you expect to ramp up in the next, call it 12 months?

Chad:

So FPV, that's a big one. And that can mean a lot of things. Some companies think of FPV purely as autonomous target recognition. Some people think of it as building clearance, inspection. Some are doing, say dangerous, coal mines and dangerous construction sites and disaster site, inspection or clearance or surveillance, and those are all super relevant. I think that's the big thrust. That's certainly what we've been putting a ton of time in for the past few years and where I really think the big growth is going to be. It's fundamentally what DJI enabled actually. I mean, yeah, they really enabled the big acceptance of it with both their systems just having great video that you could fly with, a lot of people just sta stare at the screens and fly. People started putting goggles on DJI built some FPV kits, DJI built some radio units that are individual components for great, FPV radio links. And so I think it's caught on holy cow, this is really useful and maybe, this is what we need, exactly what we need. Now we need to do that in the US and we certainly think we're a big part of that

Peter:

You're making a bet on FPV, taking a bigger footprint of drone missions. Right? And so let's look at what's happened to FPV in the last couple of years, principally in Europe. How do you assess that and what does that portend for the future? how are you gonna play that with your strategy as you enter it? Where do you see FPV sitting in a few years from now as an industry and as a technology platform?

Chad:

Certainly the momentum is towards it being just table stakes and just a big part of everything. There'll be an FPV drone everywhere. Now, is it, in, the Ukraine, Russian conflicts, it has become somewhat of the capability for say, trench warfare, where they're like literally lobbing them over you, lobbing them over the battlefield. I'm not a military strategist, so it's tough to say that's what everybody's gonna do everywhere. And where the, I think every conflict in every situation is different, but this at the core, watching the video from the UAS in a low latency reliable manner is absolutely going to be the crux of how people use, UAS and drones.

Peter:

Yeah. Well, it has turned into, in the last 24 months, the most disruptive technology on the battlefield. And it's grown. I mean, in 2023, I think we were looking at roughly 10,000 drones per month on each side being, produced and used. And now we're approaching, at least on the Ukraine side, I understand, close to 350,000 drones per month being produced and put into the fight. And so, the rate of innovation is, like on timescales measured in weeks. Whether it's in the electronic warfare or it's in tactics or it's in all of these other ways that it's being employed, it's just such a fast innovating space now. It's doing it out of necessity and it's doing it on a short term time horizon. But out of that following this conflict is, I think a really formidable platform. The question then is, well, what does it need to look like, not for Ukraine, but for the rest of the world because the rest of the world is going to adopt it. What does it need to look like as it is applied in public safety, and in other missions outside of the tactical environment? Right. And. How does it serve as a vehicle to high volume, to a robust supply chain that is forming right now, especially in Europe, that can stand alongside DJI in the commercial and enterprise market? Because I think up until FPV, DJI had a position in the market and a supply chain that was just really hard to compete with. It was really hard to see any avenue by which a new entrant could come up and compete with them. But now here's this other thing that, that isn't really in DJ i's strong suit and that a lot of the customers, one of their first requirements is that it not be, Chinese content. And so it's, it is a window for this whole new segment of the industry to develop. And here you are. Looking at it and making a play in it. So how do you assess that and how do you look at, alright, well, where are you gonna be in three or four years in the midst of all of this happening?

Chad:

Yeah, so one thing I would say is DJI does make really good FPV radio links. That is absolutely in their strong suit and that is the gold standard. They're now pushing 15 kilometers, 30, 40 millisecond, glass to glass latency. they do it at a very high bit rate and high, quality fidelity. The benefit they had there was similar to, what's going on in the Ukraine is the volume from the consumer market that really helped them refine and, drove demand for the technology, helped'em refine the technology feedback. What is working, what isn't working, when does it work, when does it does not work and so forth. The other thing that I think, and this, is going to tremendously benefit, organizations that don't have the time for the training required to fly acro mode, is that making it much easier to fly is going to be a big part of that. And that's the other facet of what we've focused on. Flies, well flies, easy streams great video. Those are the cornerstones for what we're really focused on our r and d efforts. And because, so commonly you need to pick up a simulator and fly for 40, 60, 80 hours in the simulator, then you need to go out and fly live in this acro mode it's really high precision and, high reaction time of course, that actually drives demand on the video link as well, because that video link needs to be super responsive. You actually can't achieve that flight. And so now in Ukraine, obviously that's been a necessity. So they just keep training people and they keep going after it. but it's not clear that's exactly what'll be needed all over the world all the time. Especially say law enforcement, if you're gonna pick it up out of the trunk of your car every two or three weeks or, you're a disaster recovery specialist and now it's one of the tools you have with do you really need, do you really wanna be on the sim 12, 15 hours a week honing your skills? Or do you want, my kids fly DJ I drones? There's no issue. Right? They don't struggle with, they're up, they're just playing a video game. They're just up instantaneously. And I think really to scale that is what is going to be needed in the FPV space where it's a tool in the toolbox, and not the only tool, not the only thing you're focused on.

Jim:

Chad, could you explain a little bit about, the role of FPV on the military side and why it's had such a big impact, and then why you believe on the commercial side it'll have an impact as well.

Chad:

So there's just, well, if, I don't know, maybe law enforcement isn't exactly commercial, but certainly it's non-military. We pushed out a product a couple weeks ago called Stinger, and our, we've know we're pretty nerdy building kind of development platforms and build, trying to enable our customers and so forth. And we just had this flooded inbox of law enforcement. We're desperate for this. We, how do we get one? Can we borrow one? Can we look at one? And, we're, we weren't set up. We're not set up for that at all. full disclosure, we're, very technical company focused really on very sophisticated systems and, really rely on our customers to often, bring those capabilities to market. And we have a number of, DFR, customers using our systems. And so this was just a bit surprising. So there's absolutely, we know there's now demand in law enforcement. I'd say anything, I then I think is, it gets more broadly. That same people love DJI because the video looks great and it's easy to fly. Yeah. The form factor's different. If you're doing construction surveillance, inspection, so forth, you want a nice gimbal and so forth, which is different. you think of FPV, maybe you don't think of the gimbal, but that's about the only difference, Otherwise, that core radio link, the core video, link, the flight modes, that's all the same. That's all identical to what I think is really going to be the end state for scalable FPV outside of, I think in the Ukraine situation as. Obviously nobody wants that, but it's a very, it's somewhat of a niche, there are two countries butted up to each other and you could literally, be on one side lobbying FPV drones into the other country. I don't know how prevalent, certainly in the United States is how prevalent that use case exactly is going to be. But you know, again, I'm not a military strategist, so I don't know. But I think that technology, if you go one layer deeper, having great video, having great flight modes, great flight performance, that is gonna translate to basically anything that, where a pilot is involved on the stake.

Luka:

Chad, on the industrial use cases side of things, how do you see the balance between a human operator with FPVs as opposed to an automated workflow of, machines inspecting, collecting, analyzing without the human being involved at all?

Chad:

Yeah, so we thought about the warehouse use case as something that we had fairly unique positioning and so I think in 20 20, 20 21 we really started to build out, we actually a warehouse mapping navigation capability, and we now have, a number of customers using that. We have, a really great public partnership with a company called Gather AI in Pittsburgh And they're using our core and we've, helped enable them on a full drone that just, it's on a cart, operator just hits go, it goes, comes back and does the full, shelving inspection autonomously, and, sends the data of to their cloud system where it's processed and analyzed for inventory, accounts, checks, and so forth.

Luka:

And do you see that this is a use case where FPVs might be perhaps a better option or, I guess where I'm trying to s get your thoughts on is, what really is the upside of operators with FPV goggles in inspection use cases, for instance, or enterprise use cases?

Chad:

So. I think it's somewhat of the classic use case where today, where it's. People are using DJI and that is the, they're trying to not use DJI for power line inspection and construction site inspection. and maybe not even'cause they know why, but just simply because they hear, okay, either DJI might be banned. That's a huge risk. If your company, their whole role is flying DJI UAS to go perform inspections. And all the operators are trained on that and the, their software stack is built around that. And you're hearing, I mean, it could have happened last week kinda thing where, you know, the stroke of a pen and they're gone. Or at least no new DJI, that's a real business risk. So, back to, what is this video useful for? it's all the, it's all the use cases. And right now there aren't many options in the non-Chinese world for great video links and great FPV, just in the context of maybe not goggles on, but even just with a controller, great first person flight to go perform all the inspections that now we have, referencing the couple hundred thousand part 1 0 7 pilots, they're not flying something like probably 95% of the time they're flying DJI. So now all those jobs are at risk. If they don't, you get on something quick. So having the right components into this market is going to be very important.

Luka:

And DJI aside, or just, OEM discussion aside, when you talk to end users in some of these inspection use cases, do they prefer an FPV like experience where they are in control and they view, what the drone is inspecting through a first person, view? Or would they rather want an automated system that just flags pictures of things that need addressing?

Chad:

So one of the capabilities we have on our most recent platform, we released Stinger. We call it Backtrack. So it's a visual, autonomous bread crumbing system that you fly FPV, say you're flying into a building and you're looking around and you're, say it's for, disaster surveillance or so forth. But, buildings are tough for radios because there's sometimes metal and there's lots of obstructions. So. If the radio link starts to go, you can press a button and it'll replay its path autonomously, radio link comes back up, you can go get take back control. And so those are the types of things we hear where what is autonomy trying to achieve in these scenarios. It's really, the pilot is still the smartest agent to control the drone. And they need to be in the loop, but they also, don't wanna fly acro, don't wanna spend, most of their life on a simulator trying to stay proficient. And they want the system to be smart and recover and help them recover when they need it and help'em, there's a, there is a trade off too in obstacle avoidance, with DJI launched that, that became like the biggest thing and everybody was super focused on it. But there's definitely, especially for confined spaces and tight spaces the users really want that shut off because they want to be able to go through the spaces without the drone stopping'em, the obstacle avoidance system, stopping'em. I think lane keeping and so forth is super valuable, like you find in your car, but maybe not just letting go of this, the hands off completely, where you want the operators, the users, they still wanna be in control of the system.

Peter:

Doesn't Beta flight, include sort of a couple of easy flight modes, and so how would those fit into this? Because that's pretty widely used flight controller firmware, right?

Chad:

Yep, yep. Yeah, absolutely. So I think Beta flight does have what probably what they call stabilized mode, angle mode. And so those are definitely easier, but still require hands on sticks and pretty good operation. On our systems we employ computer vision so you can fully let go of the controller and it will just hold position. And so we even have some modes where you can do that at night in the, pitch black and the UAS will still hover and hold its position so you can recover. Especially if you're on goggles and you wanna, I don't know, you need to take'em off and go look around and move and so forth. It's really good not to, need to be in control of the situation the whole time and let it just hold, stay in flight and so forth. DJI say for instance, is a good proxy, is much, much easier to fly than the most advanced Beta flight modes.

Luka:

So we talked about DJI. What are your thoughts on, on a world where DJI is not able to, or, cannot compete in some of these markets?

Chad:

So, yeah. Yeah. Like I said, I mean there were rumors that could have happened as soon as last week. You, I think personally, I mean that, I think it would be a little disappointing. They provide this really great North Star and gold standard and to not both have that kind of competition and market awareness. And also if they're not pushing it, they do a really good job of telling the industry what to be working on. And so I felt like the government saying, you, the government cannot purchase. I and the government needs to move away from Chinese components, really help create a good dynamic, maybe a carrot versus stick dynamic where, obviously innovation has come up over the last five years since the NDAA 2020 has exploded. There's tons of companies, tons of startups trying to get in, compete, build compelling products that really try to catch up to DJI That I think is wonderful if you completely ban it, then, I don't know. I think some of that I'd hate for the whole industry to just become complacent. Well, this is what you got, this is it. You just take it or leave it. I think there's still a lot more to be done and I think there's, I don't think the DJI system's fully the end state either, but I think there's a lot more to be done in the space that really, both brings in the balance of pilot control and autonomy and, communications brings in all these capabilities, pushes'em further and further. And I think having DJ around is good for that.

Luka:

I see. Great. Let's talk about modal AI and the relationship with Qualcomm.'em and how that has evolved over the years.

Chad:

Yeah, so I led robotics r and d at Qualcomm. And a big part, I'd say 95% of what we were working on were drones. That's where really the integration found in a smartphone had the strongest value proposition was with fighting gravity. All the, in incredible amounts of technology that would be packed into the little supercomputer that's in your pocket is very relevant when every gram or fraction of a gram counts. And so Qualcomm always led in that they, would market it as MIPS per watt which technically meaning how many instructions you can run in a second per unit of power. And that is very relevant to UAS. And so we were really pushing the tip of the spear in iot. There's a lot of skepticism as to whether you could actually use these smartphone components to do anything other than control a smartphone. What does Android, what is your Android apps? Does that have to do with robots or, even cameras too, even, just, IP cameras and so forth. And so we were really one of the first projects to look at that. And so we built complete systems and complete reference designs. One of those products or reference designs was called Snapdragon Flight fun side note is that it was up on Mars on the helicopter, the Ingenuity helicopter, and helped that was their visual navigation processor for the helicopter up there. And so. Now what ended up happening of this is, more Qualcomm lore in history, but the, they were being sued by the FTC as a monopoly Apple for their, this is all public, widely known, 5G patent stuff with Apple. And then the Broadcom was trying to acquire them. So Qualcomm was looking at creative ways to become more efficient and they, we, it was almost mutual proposed, spinning out what we were doing into a separate business because we had a, they Qualcomm, the go to market is reels of chips. That's what they sell is just a reel of a thousand chips or 2000 or 5000 chips. And that's what you get in the mail. This is what we get in the mail today from Qualcomm are reels of chips. And so, the drone space wasn't. There were maybe one or two, maybe DJI at that time, maybe it's not even clear, they were taking chips of that sophistication and integrating at that time. And, we're really ready for that level of, work. they wanted, companies want boards that, that are integrated and systems that are kind of pre-integrated and give you a good running start. And so that was our strategy. That was the modal AI strategy was, okay, we're gonna make this stuff available and help, drone integrators and OEMs get their systems up and running and built. And, and so that's, it was really a mutual endeavor. And they were super supportive. They wrote off a bunch of stuff and helped us start our lab, our engineering lab. And, they were actually our first paying customer as well.

Jim:

Let's say we're 10 years ago and we were in Qualcomm's headquarters. whoever was planning their interface with the drones and they mapped out the next 10 years of the future of Qualcomm with drones. How did reality end up being different from what that was envisioned 10 years ago?

Chad:

So that was like today a period of rapid innovation and rapid information coming about as to what was gonna happen. So I remember first proposing it and I was asked, okay, prove that it's more efficient than an automobile. Tell me how many acres of farmland there are in the country and how many drones you would need to survey the farmland. And then we got a little help. If you remember Jeff Bezos talked about Amazon Prime Air on 60 Minutes. And it was like, oh, okay, well if they're talking about it, maybe this is a real thing. That was like a nice tailwind to, push us forward a little bit. And then, there was, I'd say even some hesitancy on toys. What maybe DJI was seeing maybe is kind of a toy. it's fun to fly at the time. I think that was kind of what it was. Go take pictures of your family at the beach and that's, that's a fun use case. But then that volume really started going, and there were a number of engagements with big companies, that were saying, we wanna go get into this space. Really big companies like top five size companies in the world. And I have one in my office of an old reference design that was like, not even a reference design, it was a product ready to go to market that was canceled right at the last minute. And it was canceled then because the speed was quick. It went from should we be doing this at all to, okay, there's, millions of these being sold a year. Maybe there's gonna be many millions, many, maybe this is the next smartphone flying cameras. That's the thing. We all wanna be in this flying camera business. And, all of Qualcomm's big customers were kicking around drone designs and then. DJI just won like that quickly. They just it was, I don't think anybody, I, it still even surprising the winner take all nature of the consumer drone market. It still, it's still perplexing how they got so good so fast and really created that user experience that everybody still uses today. That was, and they really did, I mean, in a three year span or something, it's just they were, every three or four months was a new Phantom, a new, and then the Mavic comes out and everything's going and it was very impressive, very impressive.

Jim:

One of our guests said that one of the advantages they had is they just completely immersed themselves in the user experience and what has to be done to be able to have a great experience. Where some of the US companies were more focused on what were their capabilities and how were their capabilities is going to maybe different themselves as a company to attract capital, but we're not as focused on the user experience itself. What are your thoughts?

Chad:

Yeah, that's, I think 100% correct. And I'd say we were a part of that as well. The, that, because that's what the market, so that's what existed as a market was DJI got the consumer segment quickly and effectively. And they even put all the Chinese companies, I mean, Chinese companies out a, a way as well. I mean, I think maybe Autel is a competitor and that's about it. But there was Yuneec and, a number of different folks that were, competing with d or trying to compete with DJI. And so what was left, if you took this pie? And you thought, okay, consumers, 95% of it, there's a small amount of pie left to go divide up. And that's gonna all require, I take it, I'd say, what DJI achieved in the consumer space was FPV, just not what you're thinking of. Goggles and cheap and goggles. But, first person flying around, recording video's pretty, you can use it for sharing videos of your friends. I film, I help a friend film their engagement with a drone, right? that's, that was the great use case. And so companies, all that was left were niche. And that's, like I said, we went after the warehouse space. We thought that was okay. That's a pretty, we think that's gonna grow. We think that's going to push the pie. That part of the enterprise pie larger is these things become integrated and there's a lot of warehouses in the us there's a lot of warehouses in the world. We could, if you can pull people or help people not need to go up on cherry pickers in the middle of the day to go find out if they have more that size jeans or whatever it is, because the inventory system is refreshed every day and they know, or every couple hours, whatever it is, that, and they know where everything is. We thought that was like, that was a really great use case. And that is, there's a lot of technology that has to go in there. We've focused, I've been working on visual inertial odometry for GPS denied navigation for 12 or so plus years and that, we're still working on it and that's still a core piece of these systems and all of that technology has to go in there to go enable that. It's not an operator necessarily flying around, although maybe you can make an argument that if an operator would go fly up there and look, then that's pretty good too. But the panacea is a full automation and a digital system that the manager could just look up and type in a number, type in a thing and go, yep, they're on shelf 23 on, IL four, and find, they find the product. And so I. Now though that's, this is where, the government's definitely helped with NDA 20 and later. Every year. They actually, in 2018, I think they said, you can't use DJI if you're in the military and then in 2020 you can't use any Chinese components. And they've continued to tighten that as those regulations have tightened, it's created, it's shifted that pie of what is available now. There's, oh wait, we don't have flying cameras for government use. They don't exist. So, and everybody, everybody's working on their niche stuff goes, okay, we gotta get into the flying camera business. And so. Now, obviously that's happening. It's hard though. It takes years of effort and lots of iteration. I mean, DJI is on their 25th 30th revision of a flying camera flying, video camera. And that's very hard to accelerate. Peter mentioned you see that in Ukraine on the FPV side and, there's a forced acceleration and it goes, but it's hard. if I saw, I believe a congressman last week, say the US government purchased 50,000 drones in 2024. Well, that's probably not enough to really drive those cycles of innovation and that are needed to go be that competitive. But the market's opening the. There are components available for sure. There are UAS coming online that do a lot of it. The now because that market's opening and I think there'll definitely be, the costs are gonna come down with that. As the volumes go up, the costs are gonna come down.

Luka:

Chad, let me take you back a couple years. When Qualcomm was, and the rest of the western industry was observing this rapid rise in DJI, as it gained market share, most of the West, kind of adopted this narrative of hardware will become a commodity. Let's move into the software layer where there's, greater value quote, for better or worse, but arguably because it was impossible to compete with the manufacturing prowess of DJI and their rapid innovation and the fact that they had a lot more margin to play with because they were so vertically integrated. Do you think in hindsight, and I'm presuming from our discussion, that Qualcomm in, in light of that, pumped the brakes on its involvement in drones and shifted to, perhaps other parts of the stack. Do you think in retrospect that this was a smart move by the western industry?

Chad:

Yeah, that's a, was it a smart move? I, well, there probably was the only move really at the time. Look at our, the opportunities that created modal AI were because of that, the. Market fragmented. The government had to step in, because DJI became so dominant, I What would the, you know, I certainly, well companies, big companies overreact, I mean, always there's a, there's, you know, the, when they jump back in, they're gonna go, they're gonna go way past, and when they get out, they're gonna go all the way out and start back over again later. Whether it's a, was it a smart move? is tough. I think certainly a lot of companies probably regret not maintaining some percentage of their hardware manufacturing. Domestically, if not outside of China. They're spending a lot of energy right now getting out of China. I mean, obviously Apple's well discussed in the news on how they're moving to India and Vietnam and so forth, and that, I think maybe the le the big lesson learned is diversity of supply chain is incredibly important. And I think 10 years ago, eight years ago, people would've said, oh yeah, it's important, but we have two CMS in China that are handling this for us. And that wasn't good enough. That so well, there was the smart, it was probably the only move, but. This is largely because it's still, a relatively small industry. it had, the bigger question is on the smartphone side is should everybody have gotten just full stop reliant on China for everything?'cause that's what drones come out of is the, Chris Anderson said it's a derivative of the smartphone wars and that's absolutely true. And so the, we leveraged all those components and so we, so I think basically everybody just followed what was going on in the, and they were looking to fill as the smartphone growth waned. They were also, those factories were looking to fill it with stuff. And I mean, Xiaomi, well, that was a company we worked with and they, they had a drone spinoff, that was doing drones to help because, oh, it's just a flying smartphone kind of thing.

Luka:

I know you're not part of Qualcomm anymore, obviously, but do you know, what is Qualcomm's, participation in the drone industry today? How do they see it today? What are the products, that they're serving the industry with? Is it through modal ai? Or, other ways and how do they sit relative to the competition?

Chad:

So modal AI certainly helps Qualcomm scale. I mean, they would refer to us as a scaling partner. So the, and often in a lot of their pro chip set product development, there's a lead partner and a scaling partner and we would serve and we do serve really well as a scaling partner, how to get as many companies onto their chip sets and technology as possible. And so their core business is historically in the chip set space where the processors and then the modems and of course they still feel very strongly about, and we still work with them today on cellular connected UAS and that was something that we worked on significantly when we were at Qualcomm as well, was really proving, for one, proving that it was possible. There was a lot of skeptic. There still is some skeptic, oh, the antennas are pointed down and so forth, but it works. It's, you're up there, it's free space propagation. You actually, the part of the bigger problem is that you have too much coverage and you're actually, one drone is transmitting to all the different drones or all the different base stations in the city. And so, we put out a white paper help dissuade that. And so that's still something Qualcomm's super active in, is participating in the standard, I'll probably muck up the acronyms, but like a TSM and other, standards bodies. Qualcomm is a very large standards organization and they dedicate time towards, pushing forward cellular connectivity in a lot of areas. And they actually, part of 5G low latency, aspects were really designed for autonomous systems, like drones and flying and driving and so forth, and having connectivity, built in and reliable and robust. And so beyond that, I think they're, constantly, like all big companies kinda looking at what can they put together presents them as unique and special. Their chips are, the chips are great for drones, cellular connectivity, they have the best. So they're well positioned on a lot of those areas today.

Luka:

And is it fair to say that Qualcomm's interest in the drone space has been shaped by the perception of the relevance of cellular connectivity in the industry? And if that's true, how does that role of cellular evolved over the years? You mentioned that you've done some early studies. I believe, reading through some of these early reports many years ago, about coverage and quality of data links with, with cellular, how have you seen that arc of evolution?

Chad:

I'd say if cellular connected drones were ubiquitous and everywhere all the time, that Qualcomm would be very interested. That, that, I think is fair to say. But it, they've, there's definitely been a consistent drumbeat of activity and and promotion on their behalf. There was, they were big supporters and at least when I was there, certainly big supporters of the network connected remote id. but the remote ID is today in today's implementations tough to see the value. And I think if you had tiered it and said, okay, if you're. I don't know if you're doing something, drone delivery or, carrying, how about just say if you're carrying something, you're, you need to be connected. You need to be connected to a cellular network and broadcasting your location to the FAA or publicly or however you wanna broadcast it. I think that's just a no brainer. All of the, if you go to any of the drone delivery companies, I promise you they have at least one cellular connection, on each drone. And they are monitoring that drone at their, at their central location. So it's there, it's already there. I think it's just a huge miss that they, FAA isn't getting that data as well. I don't even think the companies would object. I think they go, oh yeah, give us the API or whatever, and we'll publish the data and move on. And so that now it's there. So maybe from a Qualcomm perspective, it's there and all these drone delivery companies are cellular connected and so maybe that's the good state, but they're definitely should be more next generation ATC needs to be absolutely cellular connected. It's by far the most ubiquitous connectivity layer, and if you're doing anything else, you're basically reinventing cellular. So if you have a bunch of UAS flying at 200 feet and you need to connect them to a broader network, it is cellular. Otherwise you were, yeah, absolutely. Just reinventing a cellular network.

Luka:

Great. Let's talk about the technologies that you are developing at modal.

Chad:

Yeah. So we look at our role in the world as getting our customers 90% or so of the way there, and they go build on top of it what makes their platform special? What makes it unique and special? Jim brought up user experience and, they, these, that is a big part of it, and maybe I didn't quite finish the thought, but that is really important, so you need great technology to go build those next, generation user experiences on. We obviously, as mentioned a few times are pretty biased towards the Qualcomm processors. They, generally compared to an equivalent NVIDIA system, we get it's one 10th the weight and half the power. And so for fighting gravity, really important. And so you can build much smaller, safer, smarter drones using our capabilities. And we continue to push forward that ecosystem around that. I'd say, I don't know, maybe it was like hubris coming from a big company or so, but we thought, well, we'd bring the computer and maybe a computer and flight controller and the industry would build all these components around it and we will just be off and running. But we realized, okay, so there's some, these modems, but nobody quite knows how to connect them, so let's build some adapters to connect some modems then. Okay. Well, everybody's using Chinese cameras and so if they're using Chinese cameras and they pair with, our system is so not NDA compliant and there's no US made cameras, so, okay. So we should build cameras too. And so then, and then now we're, the FPV links are. they're not low latency enough, so we should go, we're gonna need to go do that. So we've just continued to expand and integrate around this central computing platform. speed, electronic speed controllers. Everything was Chinese. and so, we went and built this stuff around it. We build reference UAS around that. And so we have a quick start. Those drones, we have a number of the Stingers, our latest FPV drone that's fully open and, accessible for software development. Our codes at code do modal ai.com, and Starling has been used by, I mean, I think we've shipped thousands of Starling drones to different universities, different government labs, and they go build niche applications with those, there's a ready to go platform that is, really capable with computer vision and reference software to go build a lot of different things. And so that's the crux of the business is really getting people going with sophisticated autonomous capabilities quickly and hopefully reduce their time to market in a very measurable way.

Luka:

What is the remaining 10% that the end users then develop on top of your components?

Chad:

Oh, we have, there's so many. So we have a company in the Netherlands that builds greenhouse inspection drones. They fly around up and down crop aisles and take pictures of the crops and put those into a cloud database for AI analysis. We mentioned gather ai, who we work with, similar things for warehouse aisles. The Dutch company's corvus drones. really cool stuff. And that's one of the great, feel good. Okay. You're just improving crop efficiency. we've worked with coal mine inspection companies that are like, there's a, an issue with the mine and somebody's trapped, or they wanna see if they're trapped. And you can fly a drone down there quickly, get some video and come back. there's obvious government, military applications. There's, we've worked a ton with, Nasa, JPL. We are still slated. Our voxel is slated to be on some rovers that go on the moon. That'll drive around autonomously on the moon. It was supposed to be this year though. Hopefully it's still this year. lots of, it's just, just a number of, it's really broad. And so that's. We can't go tackle all of the, as a single company, we can't go tackle all those use cases. So it's really exciting to us that folks go and then go build on top of that. And sometimes it's autonomous. Sometimes it's FPV, sometimes it's just, a rover for the moon. It's all diverse and sometimes comes across as fairly random. Most random, I think is somebody used our system to build a glove for a VR system. So the glove, they could track because we have this visual inertial endometry capability, they can go track with a glove is in 3D space.

Jim:

This is really interesting. So when your company was introduced? What was the problem you were solving and what is it that you did uniquely and do uniquely?

Chad:

So really the Qualcomm capability is, or the Qualcomm SOCs are really capable for unmanned aerial or any aerial system, quite frankly, including riding along on a rocket to get somewhere. And it's also really, it is not even a next level. It's two levels up of electronics capability. There's no, even for us, we've got the same team that we had inside of Qualcomm designing this stuff. And I'd roughly pencil in a brand new design with a brand new Qualcomm chip set at a year, nine months maybe if you're moving really quick to get to market inside of Qualcomm. We did it in nine months for this first Snapdragon Flight. and that was like a heroic effort to get it to market. And so we build that capability and make it available in a very low friction environment. So if you're gonna go to a company like Qualcomm or even Amberella or Nvidia and try and work directly with their chips, they're of course gonna have some volume expectations at a minimum and constraints. And then you're gonna need an engineering team that can do it and go actually build these high speed designs where the layout, just the layout alone will be 12, 16 weeks, in a dark room with, precise traces. And then you've gotta bring it up. And then you have to enable all the peripherals and you've gotta go make the system reliable and make sure it's patched away, security patches and so forth. And so we look at that, if you go work with modal ai, and that was the original purpose of Snapdragon flight too. It's if you go work with these systems, you're gonna get flying really quick. And now we have, we have full, complete drones that you can go start your software development on while you're performing your drone design. And we have a lot of customers that take our designs and like just modify it to go meet their need. And then, and so we provide the CAD files and they can go modify it to meet their need. And it's really low friction access to getting the best technology so you can get to market in efficient timeline.

Luka:

And so what is all comprised in these reference designs that are coming from modal? Is it the, obviously the processing, you mentioned the, the communications, chips, goggles now most recently speed controllers. What else are you developing on the component side?

Chad:

On our reference UAS, almost everything. I think sometimes we include a wifi dongle just because it's FCC certified and we haven't gone through that process on the radio side yet. But yeah, cameras, we build all the cameras in the us, the, take the Sony chips or the on or, yeah, on semi chips and place'em on the boards. Mount the lenses, focus'em, test them. Cables. Lots of cables. The cables are the lubricants of the drone industry. There's no, there's cables everywhere. we have almost, 180 unique cable designs on our website to go build these systems and tie it all together. But yeah, speed controllers, we don't build motors. We don't build propellers. Trying to think what we don't do. Motors and propellers and antennas. Batteries. Yeah. Yeah. We're not doing batteries, propellers, motors, but pretty much everything else.

Luka:

What are you most proud of in terms of the specific technologies? Is it on the perception side, processing or radios?

Chad:

I think I'm most proud of, a little bit more holistically that we are building this competitive capable technology in the United States. I think in the end, what I'd love us to be known for is really pushing advanced deep technology in the US with no compromises, including cost.

Luka:

Can you talk about the perception part of, the stack and what your approach is? You mentioned, Visual-Inertial Odometry, but can you talk more about it and how different players are going after this problem and what your view of the different approaches and trade-offs is.

Chad:

So, yeah, so yeah, VIO is certainly core to most autonomous UAS systems. Skydio certainly has built incredible technology. They just also announced a visual anchor feature last week that, that's really cool.

Luka:

and Chad, please describe VIO for the audience as well.

Chad:

Yeah. So Visual-Inertial Odometry takes imagery from digital sensors and the inertial measurements from an accelerometer and gyroscope and fuses them together to create a location or in odometry this kinda a term for location. So in metric space. So if the system is working well, it knows exactly where it has an x, y, z coordinate in the metric system so it knows how many centimeters or meters it has moved since the system started. And this is really powerful. In terms of obviously the higher level term is GPS denied navigation, so flying a predetermined path using only the vision and inertial sensors. But it's also very useful just for just p holding positions stably. So even on GPS, if it's just GPS hold, it'll float around, kind of bob around. If it's vision hold, we can barely look. It looks like it's just hanging on a string from the sky. If you didn't hear it, you wouldn't even know that it's flying. It's just holding so still in the sky or in the building and so forth. And so this also helps for that same kind of performance, what's called position control or position hold inside of buildings such as warehouses or high rises or whatever the, OR stadiums, drones can fly through stadiums and go do, interesting inspection tasks and so forth, and capture video and so forth. And so that's the crux of, that's a big part of what modal AI works on. And it's a lot of software, but it's a lot of hardware. It's also a lot of software that's interacting at a very low level with hardware. The timing is very important, the precision to which these ac measurements are being made mechanically. The structures need to be designed in such a way as to not create too much vibration and the systems, are built robustly and can survive crashes and so forth. And so that Visual-Inertial Odometry is a core part of what modal AI does. It's a, it's certainly a big part of what a lot of, DJI skydio those systems, they're all employing some form of visual inertial endometry to go help build out these robust, more advanced flight modes.

Luka:

This is a relative positioning method, right? So you need some sort of understanding of your starting position in terms of, GPS coordinates. What if you don't have that initially?

Chad:

Yeah, so we actually have a, we, I mean, we can, you can load up a map with our system and click, this is where I am. And right now I think we require them to point it due north. But otherwise,'cause you also, if you don't have, you might not know what direction you're facing as well, but if you can, if you have those through things, you click on a map, say, this is where I am, and point the drone, do north, and it can go fly with VIO and go do something. Now, our systems tend to work well up to 30, 50 meters of altitude. there's nothing really stopping it from going higher. It's just sometimes you need, tweak, tune the processing, maybe bigger lens, bigger sensors and so forth to go higher. But, there are systems, of course, that folks are, lots of folks are building is that's the dedicated portion of their company to go do high altitude, long range, type flight. Those GPS denied navigation systems really are often, I mean, almost exclusively visual inertial endometry in some way, shape or form. We offer ours just integrated in the system with no extra charge and fees. The code's actually open source for the most part. As a starting point, we leveraged, an open source package and then we've, we spend just tons of time optimizing it on the processing side so they can run efficiently on this edge platform.

Luka:

Now, how do you compare and contrast skydio approach to perception and collision avoidance?

Chad:

So I would, I mean, I don't know the internals of their system, but they're using a visual inertial odometry system. They're, I mean, to get super technical, they like a bunch of sensors pointed in the same direction. We like sensors pointed Orthogonally from each other. But that's, that's probably pretty nuanced. I think either system would work well, switching that around they seem to do a great job. I'm, I have a ton of respect for skydio and the technology they've built, I would say on obstacle avoidance, they've done a great job. What we have found more recently though, that demand has really waned on the obstacle avoidance side. That used to be just a really big topic. and I think, I, when Skydio first came out, that idea of a flying virtual camera that would follow you and avoid trees along the way. I mean, that's just insanely impressive technology. I think in the end, back to maybe FPV and maybe FPV broadly. People want to fly, they wanna have control, but they want the autopilot to assist them and not do anything stupid. And so I think Skydio probably does a really good job at that as well. It's just the, at least that original use case seems to not have been in higher demand, is I think everybody kind of thought, and I know they weren't alone. I remember at Qualcomm multiple big name brand companies saying this is exactly what we want. And I think in the end it turned out people like flying them. It's fun.

Luka:

What else about autonomy or technologies that are enabling autonomy is worthwhile discussing or flagging here? What do people get wrong when it comes to thinking about implementing autonomy in constrained environments?

Chad:

So I, I think maybe that autonomy is an end state. It's what you need to automate something first, figure out what you're actually doing and then automate it. And that's where the value's going to come from. And I think, I mean, I'm sure we, we've done this plenty where we've shown videos of drones flying through buildings, fully autonomous. And that's part of where we sit in the world as a technology enabler and en and enabling, further on and use cases. But that's not the end state. if you're doing inventory surveillance, you need to be accelerating that capability. so if there's one person on a cherry picker, or actually a great one is rollercoasters we're not actively working on this, and I don't know who is, but it's a great use case for autonomy. So they have to inspect every bolt on a rollercoaster, at least once a year. And they have a team of people at these big amusement parks that are dedicated to that. They're dedicated to that job. And it takes them all year to go through the whole amusement park. Now, a drone could like really do that every day. Or you had one drone parked out every day, and maybe now they're doing this, but it's a great use case to just highlight, well now they're just better. There's still a team that's employed doing this, there's somebody looking at the bolts like, yeah, maybe AI could do it, but would you really trust it for sure to just say, yeah, we're 100% you want, you still that same person inspecting. It's gonna be inspecting it, but now they're doing it every day per rollercoaster instead of once a year. There's still gonna be a team actually now more technical, maintaining that drone in a box. Or maybe they're not in a box, they just go out and park closes, they go launch it and go hit a button and it goes, does its thing. There's still this team, the same team is out there now they're more technical now, they're way more efficient. They're doing this way more often and, achieving a higher level of safety. And so I think that's a much better end status that, take some task that takes a while, slow, cumbersome, expensive.'cause now you gotta need an expensive piece of equipment to get somebody up there and somebody's, gas and fuel and the whole thing and make it quick, cheap. Abundance. That's really what autonomy is for. And so I think that's like a great end state of what you wanna achieve.

Luka:

Great. What about radios? We, touched on cellular connected drones. I'm sure you've been tracking a lot the developments in the war in Ukraine and how LTE has been used, to, hide drone activity, and protect from, from some forms of electronic warfare. What are your observations, in thinking around LTE enabled BVLOS operations?

Chad:

So in the commercial space, the, as we talked about before, drone delivery, fully abundant, I think it's absolutely there and I think you need it. And I don't think anybody's not doing it. So in the more government military use case, I think certainly in the law enforcement, good backup, especially if they're, I don't know, drone is first responder, I'm sure. I think you'd be crazy not to have a cellular network on there. Maybe when they're doing DJI and they're rebroadcasting, those the DJI link, they can't have a cellular on the UAS, but absolutely that's another use case. On the military side, obviously the Ukraine Spider Web operation really highlighted the fact that, boy, that's reliable connectivity sitting there that's not being turned off. And I guess what happens next is do, was it a one time thing? Do they keep going after it? Does the other side start to disable their cellular networks? That would be tough. The the population's not gonna be happy about that. Now, probably not happy about the other outcomes as well, but it creates a really tough situation to say. I mean, and it wouldn't even, you couldn't even isolate it unless you had some sort of intelligence that says, oh, they're going to attack this town. Then maybe you might make that choice to shut off the cellular network in that town. But if not, if it could be anywhere. The cellular connectivity is, ubiquitous or in a lot of places that really matter. And so very difficult to disable comprehensively. and you'd have to replace it with something that would be just to provide the same capability so that you're there. And so as a part of, and I don't know the state of base station technology today, but as a part of what we were doing at Qualcomm when we did this study with at and t, was that. The networks could tell if it was a drone, but they only took measurements every 15 minutes. And there was post-processing involved there. Absolutely. At least at that time was no concept of identifying a drone on the network. But, that might be where it goes too, is like there becomes, counter UAS, there's a whole faction of counter UAS companies that are just trying to provide realtime analysis on cellular devices and what they're doing and where they're going. But that's tough'cause you need basically, there's only a couple base station companies and they're gonna gate keep that whole thing, Ericsson, Huawei, Nokia or whatever. That's about it.

Peter:

Now, earlier you talked about creating low latency FPV links, as part of what you're developing, are those using cellular or does cellular introduce too much latency for what you're talking about?

Chad:

So all of the work we put into optimizing the Qualcomm camera pipeline for FPV imagery ingress, encoding, decoding is applicable. And it actually, we've done some tests and it looks really good and usable. Connectivity dependence of course, but just in our office, plugging in a T-Mobile sim card and trying it, it looked pretty good. And then we've had the same success by applying these techniques to some of the more prevalent, mesh network radios like Doodle Labs, the where we can just by applying the software techniques that we've invested in on the FPV side, they are relevant. Now there's different performance and radios are tough and there's always different things that are holding them back, but there is stuff that can be done there to make those links usable for FPV as well.

Peter:

Interesting. Okay.

Luka:

What would you say is the achievable, reliable, repeatable latency that one can expect from a cellular connected drone to be able to employ in an FPV manner, across the globe.

Chad:

There's a lot of uncertainty there. now you're talking about, fiber optics under the ocean and so forth too, but it's, to truly be across the globe, but. There's no reason it's not possible, and especially if you throw in a little edge autonomy in there to make the system easier to fly. That's a big part of that as well. The lowest latency is required if you're racing in kind of acro mode and then you get into, say, stabilized modes and so forth. You still need pretty tight control. But if you relieve that burden and it's a little bit easier to fly and not gonna just spiral outta control, if connectivity goes down for a second or so, then it's all a very, it's all very doable. We demonstrated that at Qualcomm and well across the country. We had an office in Philadelphia and we flew from the office in Philadelphia, a drone off the rooftop in San Diego, and, and we're flying around. It's definitely doable.

Peter:

From the publicly available video from Operation Spiderweb, it appeared that the drone operators were dealing with really high latency, just looking at the way that they approached the aircraft and the way that they selected where they were going to land, the drone on the aircraft. I looked at it and it looked like they were flying in a way that was very careful and deliberate and dealing with really high latency. Do you have any other information on that? Is that the case and is that indicative of what we would expect from cellular, or do you see otherwise?

Chad:

Yeah, there's just a lot of, it depends in between to be comprehensive, but I think from what I saw, I think what they've described is Ardupilot GPS waypoint missions, which were triggered by cellular. And so, whether that's autonomous or automated or so forth. I think it looks deliberate and so forth because they were literally flying GPS waypoint missions, which kind of looked like up over, up, down, up left, kinda stuttered missions sometimes. And so I think in that specific case, but it was just triggered by cellular, Hey, health check drone online. Yep, we're good. Okay. Go.

Peter:

You know what I'm referring to is, for instance, the, some of the first video that came out when it showed their attacks on some of the TU 95 Bears, that the pilots were flying over the aircraft, getting right over the part of the wing where they were trained and wanted to strike, and they had the camera pointed directly down, and then they were slowly descending onto that spot. It was a very different flight pattern than what you see. In the typical videos in Ukraine where it's an FPV link and it's very dynamic and they are, flying with, complex maneuvers and a lot of agility. This was completely different. And so, I don't think that was GPS Waypoint based. Maybe the approach to the airport was, but the terminal phase appeared to be a pilot visually selecting the part of the target to land on and doing it in a way that was very deliberate. And that could be explained by the fact that they were dealing with a ton of latency on their channel. I mean, I don't want to, go too deep down a rabbit hole on this, but it is an interesting question,

Chad:

Totally fair who even knows what the network latency is just from handing off between Ukraine to Russia, there could be some switch in the middle that's just slow and causing problems. So there would need to be experimentation required and maybe they built it in such a way that they didn't know and they didn't know what it was. So they were really trying to just be robust and which is possible as well.

Peter:

But the takeaway is that you and your team have, gathered real flight data, operating video links on cellular, and seeing really great low latency performance, just, it's a different environment that you're flying in, but that's what you see,

Chad:

Yes, but again, all conditioned on a lot of different things. But you can, there's absolutely techniques that can be done to, and processing helps, like latency is every step of the way. So, I mean, we spent time in building our FPV links. We spent time on how long it takes just to get the data from the sensor into the image signal processor, and then how long we spend in. In fact, we didn't use the smartphone image signal processor. We put everything in the GPU just for latency, purely to cut, milliseconds of latency out of the stack. And so I. All of it matters end to end. And I don't know enough about that. The system architecture on spiderweb, I thought I saw maybe raspberry pies and Ardupilot and so yeah, so our system's better than Raspberry Pie in terms of processing video so much faster. much, probably, 10 x faster, something like that. So, I would expect ours to have better latency performance just from that computing. So it's hard to say where that latency and we'd really have to analyze it and look at, and we do this on our systems. We look at every step of the way and measure the full timeline and document the timeline and then work on where we think we can optimize that timeline further.'Cause it really does every single and on the same on shoot, we, some of the, sometimes people are like, oh, the latency's terrible. And it's just because the software they're using on the ground station side, buffers a second of data before rendering to the screen. And so that now it's a second to latency just added because it's buffering. And so the FPV stuff is really complicated. Digital FPV, there, there's a lot of steps along the way that, and the rendering side matters just as much as the, as the capture side. So, and the radio matters just as much as everything else. So, you know what the, I think if you were to just start doing ping tests over cellular, you'd see pretty quick round trip to some, we could ping a server in Ukraine or something and from your phone, and you'd probably get a sense for what that round trip latency was. And so the rest of the system matters as well. And how you handle the data, the protocols used and so forth.

Luka:

Interesting. All right. Well, as we wrap up, maybe one parting question. If you had to bet on one area of innovation in the drone space that will define the next decade, what would that be?

Chad:

Oh, there's a lot. So maybe back to Jim's user experience. I don't think we're there yet. Outside of DJI, I think DJI is still the only one with really a great user experience and so we're all technical, tend to be technical and, engineers and really wanna build the next great technologies. I think user experience is going to be paramount if we wanna be competitive. And I think. Then the enablers of that will be flying, well, flying easily streaming great video, high quality video and easy to fly are gonna drive that user experience. And if it's new technologies looking for, what new could pop out of all this in the next five or so years that isn't quite there today. I think, there's the fusion of FPV, virtual reality and photogrammetry. Something's gonna happen there where you're putting goggles on for FPV video, very similar to the goggles and maybe even identical to the goggles that people are putting on for virtual reality. A lot of this data captures going in. You've got Gaussian Splatting and you've got a lot of this, these new techniques that are making really compelling, 3D models and renders based on the data being collected. So flying FPV, collecting data, generating 3D models, re-rendering them back into, goggles that are already on your head. There's something there. I don't exactly know what it is, but I think that's like the far flung use cases that are, gonna happen at some point. There's so much going on in those three areas that the overlap, the Venn diagram has a lot of overlap of similar capabilities.

Luka:

That's an interesting idea. Great. Well, Chad we really enjoyed this. Thanks so much for your time.

Chad:

Yeah. I hopefully I didn't get, sometimes I go too technical,

Luka:

No such thing on this channel.

Chad:

Well good. Yeah, it was great. Absolutely.

Luka:

Thank you very much.