Subscribe
Share
Share
Embed
The Drone Network explores how drones are reshaping the world. Hosted by Bryce Bladon, the podcast documents the tech, economics and people piloting the world's largest standardized drone imagery network.
The Man Building America's Autonomous Systems Stack
===
Bryce Bladon: [00:00:00] Hello, and welcome to The Drone Network, the only podcast in the air and on the airwaves. I am your host, Bryce Bladon, and on this show, we document the tech, people, and projects expanding the drone industry as we know it. Today's topic: the US is restricting foreign drone tech, but are domestic alternatives actually ready to fly?
Today's guest, Mike Horton, CEO of HyFix. Mike Horton is the CEO of HyFix [00:00:30] Spatial Intelligence and the project creator of GEODNET. His career in autonomous systems spans decades. He co-founded Crossbow Technology in '95 while at UC Berkeley, sold it to Moog Aerospace in 2011, uh, served as CTO at, and I'm gonna say this wrong, Asinea, also CTO at Anello Photonics, and now working on HyFix.
A pleasure to have you on the show, Mike.
Mike Horton: Great to be here. Thanks for having me.
Bryce Bladon: From self-driving [00:01:00] tractors to drones and robotics, walk us through your journey.
Mike Horton: Yeah, so I started out at UC Berkeley working on MEMS-based sensors, accelerometers, and gyros, and we had the idea to use these accelerometers and gyros that were being made in, in silicon chip scale, accelerometers and gyros for virtual reality, and this was back in 1995, a little before virtual reality was ready for prime time.
The compute displays weren't quite there, but what we found with that sensor technology was that we could apply [00:01:30] it to things like unmanned systems for, uh, autopilot and guidance, and I built up my first company, Crossbow, which did, uh, integrated inertial navigation systems for drones and for early autonomous systems, and I sold that, as you mentioned, to Moog Aerospace.
So that's kind of what kicked off my career in the field of, uh, navigation and localization and control systems, and the application that's, I think, the most exciting for all that kind of tech these days, and has been for a while, [00:02:00] is, is autonomy, and that spans robotics, self-driving cars, and of course, drones.
And I think drones is one of the premier applications, actually, of autonomy and robotics and the impact it has had across so many different industries.
Bryce Bladon: It's truly undeniable, but I'm curious, what drew you to autonomous systems so early? As you mentioned, you were already in VR in '95. How'd you get to autonomous systems, and what's changed since then?
Mike Horton: Well, I think the key thing that's making these systems [00:02:30] more viable now is the sensor cost, actually. So the sensor quality and cost has come down, I, I'd say much faster than Moore's Law. So if you were to look back when Google first started putting self-driving cars on the road, the cost to, you know, do, um, say LIDAR, for example, at that time was like 50, $60,000 for a LIDAR.
And now production vehicles are rolling off the line with LIDARs that are under $200. And even DJI drones are inc- including [00:03:00] small line count LIDARs in consumer products. And the same thing can be true for the sort of whole image processing pipeline. Same thing can be said for the precise positioning pipeline.
When I started my career, RTK was barely just a sort of a research topic. How can you get to centimeter accuracy with GPS? And now you've got a global network like GEODNET providing you RTK corrections almost anywhere you are on the planet, and you've got chipsets at well under $50 that can implement, uh, high precision centimeter accurate localization.
So these, the [00:03:30] sensor costs coming way down at the same time as compute has obviously improved and improved, and at the same time as like AI has become, uh, something that really works and is really usable. That combination is what's made physical AI and, and, you know, autonomy, embodied AI, robotics, all the different things you can call it, name it, actually really super practical, um, right now.
And when the cost is low for the hardware and the functionality is high for the level of autonomy delivered, you get ROI. And that [00:04:00] means that a customer can buy these systems and, you know, very quickly, uh, experience a benefit and a return on investment on, on, on it, whether that's, you know, shooting drone photography for fun or for real estate Or doing something really advanced, like a Waymo self-driving car.
You can, you can make these systems economically viable, and that's the key to really getting them to scale.
Bryce Bladon: Well, speaking of scale, speaking of making things viable, let's talk about the current state of the US drone industry. Um, specifically, [00:04:30] something that's really active in, in the listener community right now, um, which is, as of December...
You know what? I'm gonna let you talk about it. You're far more educated on the issue. But let's put it simply. What are the US' current restrictions on foreign drone technology?
Mike Horton: Well, as you said, as of December 22nd, 2025, the FCC, uh, came through with some things that had been talked about for a long time, um, as of, you know, technology that would be restricted, and it's called the Covered List.
And the [00:05:00] FCC actually went broader with it than I think people were expecting in their December 22nd, um, action, and that is to, uh, ban all foreign-made drones and critical components from receiving FCC approval for use in the United States. So as you know, any electronic device that has a radio in it, and a drone has a radio in it, needs to apply for FCC approval, um, and get a FCC ID number.
And so that had been the mechanism that kind [00:05:30] of drones were being approved under. And, uh, now with that, uh, Covered List, that means that, uh, new models of DJI and other foreign-made drones, um, will not be given approval. Um, so that means that the foreign drone companies, um, are gonna have a lot harder time accessing the US market.
Now, they can potentially come to the US and set up production and work with the FCC and meet, meet the requirements that they want, but others will probably have, some of the Chinese makers will probably have a pretty [00:06:00] tough time doing that because the ex- the regulation kind of explicitly came from different committees working on Capitol Hill that were concerned about national security, um, and specifically the development and advancement of, of drones, um, made by the People's Republic of China.
And so I don't think that it's very likely that Chinese makers will, will be, uh, easily able to get a FCC ID, but perhaps other countries that are more allied nations will be able to set up shop in the US, make drones in the US, and, [00:06:30] and help, you know, fill the gap that's been created. But it was a broad, um, the Covered List was broader and wider than I think most in the industry had expected.
It had been talked about, I'd say, for a good two plus years, and people didn't... speculated would it happen, would it not happen, a lot of back and forth, and then, and then it dropped on. So it's a, it's the rule of the law, what are they, the law of the land now.
Bryce Bladon: Absolutely. You, you mentioned a bit of a gap there.
So this, uh, legislation came down December 2025. There's a [00:07:00] restriction on new foreign-made drones. Are there domestic alternatives that are ready? And if not, like, what are, what are the risks right now?
Mike Horton: Well, I think there's, there's domestic alternatives that are ready, and then there's domestic alternatives that are being developed.
And then at the same time, the ruling does not restrict the sale of existing models that obtained FCC certification. So models of, like DJI Mini 4, you can still buy, although there can be some challenges getting ahold of it from other actions that the government has taken. But in [00:07:30] general, you can purchase those in the United States.
But DJI Mini 5, which is the newest edition, is not available for sale in, in the United States. So there is a supply, I think, of, of drones that are, are pretty capable, um, that are still available for purchase. Um, but going forward, the US industry needs to innovate and needs to kinda catch up with the state of the art.
And the idea with this, uh, legislation is it's gonna kinda give the, the breathing room for the US industry to do that without facing [00:08:00] what some in the industry had accused of the, of these products being sorta dumped on the US market at below cost and making it very, very difficult for US manufacturers to really be able to get into the game.
Bryce Bladon: You, you hit on something there, and that is sorta getting on, getting in on the game. At the same time- We, we're at a period where supply chain vulnerabilities are becoming clear. Can you explain how fragmented drone systems create security and supply chain vulnerabilities? Like, what is, what is the solution at the end of this challenge?[00:08:30]
Mike Horton: Well, we think the solution at the end of the challenge is to build a system on a chip. That's what my company HyFix is doing, and it's a system on a chip that consolidates the major functions of a drone, including positioning, flight control, secure communications, and AI and spatial computing capability.
And the reason a system on a chip is helpful to the supply chain vulnerability is that, uh, consolidates so many parts. And if you look at today's modern integrated circuits, modern chips, as we call them, most of them [00:09:00] are not made in the United States. In fact, the vast, vast majority are made in Taiwan, Korea, or even factories in Europe.
But really a very small percentage of chips today, um, especially the ones that are in, in drones, which are popular in drones like the STM32 microcontroller, um, uh, Qualcomm, s- uh, Snapdragon, a CPU, GPU, not typically, uh, made in the United States. I don't think any of the STM32s, for example, are made in the United States.
So you look through [00:09:30] that supply chain, first of all, you have a lot of parts, and none of them made in the United States, and that is a supply chain vulnerability, especially around a critical technology like drones. Because when it does come to a conflict or some situation where there's geopolitical turmoil, you're not really sure who's gonna be available to, to get, you know, parts to you and who's not, where the restrictions are gonna be, or where even as we've seen within in the Iran conflict recently, like naval passages can get blocked.
And so then you can be [00:10:00] restricted from having access to the technology and parts you need to build drones, and that obviously has strategic implications. So I think the goal with a system on a chip is to, one, you know, have something native, uh, domestically produced, and the simplification of the supply chain.
Both of those are big factors in what makes that a great approach to, um, you know- Solve some of the supply chain challenges
Bryce Bladon: Fantastically said. Let's take it back just a little bit though, 'cause you talked quite a bit about how you've combined all these things. What, what is this called? Is this the, [00:10:30] the H1P?
Mike Horton: Yeah, so we have the H1 is the core silicon- Okay ... and we're calling it the autonomous systems chip 'cause the application of this chip is the kind of first dedicated system-on-a-chip really targeted for autonomous systems. Now certainly, a, a GPU platform combined with a, you know, MIPI interface and, and a separate radio and can do all these things.
But really building one all-in-one package and doing it with an appropriate level of compute where you can do some [00:11:00] AI, you can do for vehicle control, but not so much compute that it becomes, you know, you need a huge battery pack or a very large vehicle, that you can take this all the way down to, say, a sub-250 gram drone.
So that's the, um, r- reason we created the H1 was to sort of hit small drones and small robots with something that was performant for AI, um, communications, positioning, and so forth, but not, you know, over power or over weight or over cost, uh, relative [00:11:30] to the, to the needs in that marketplace. And we're taking that H1 core chip and we're putting it into two modules.
The H1P is the first one to come out, and that module is targeted at, uh, flight control And positioning, and it has our, uh, uh, positioning stack in it. It takes, ties into GEODNET. It ca- it runs PX4. Um, and it has a free CPU in there, 64-bit, that can run Linux, can do other tasks that the customer wanted to do.
And then we have the H1D that's coming out that couples that module [00:12:00] with additional, um, DDR and camera interface and is really... And a, and radio, RF radio, uh, link. So that is targeted to be an entire drone module, and that will come out in September.
Bryce Bladon: Oh, wow. Okay, that's incredible. So I've gotta ask, having what can only be described as a community of passionate drone pilots who are now wondering where they're gonna get their next kinda consumer-grade drones, let me flip the question around before I ask the big one, which is, what's been the hardest part of building this in the US [00:12:30] at this level?
Mike Horton: Well, I think the supply chain in the US is complicated, so the packaging and testing. There are foundries here, so GlobalFoundries in Malta, New York, um, Samsung in Austin, Texas, Intel in Arizona, TSMC is building. So there are fabs that are available to run the front end, which is, you know, making the wafers, but then you also have the back end consideration of packaging and tests.
And then when you start to assemble up, you know, from the chip level into boards and [00:13:00] modules and things, there's, there is supply. It is at a price premium relative to, um, some of the international competitors, be it China or the Philippines or Malaysia, and so that's, that's, I think, the challenge. Also, lead times tend to be a lot longer.
So you take a, a part you wanna make, a custom part, in, in China you might be able to get that turned around in a week or something. In the US it might be four weeks. So these are things. Now, there's a lot of community around reshoring technology, um, a lot of investment going in, so I [00:13:30] expect these things will get better over time.
But at the current state, you're paying a, a, you know, a higher price and it's taking, taking longer to get things done. That's the bad news. On the good news, I think on the semiconductor side there are some, you know, really, there's, there's, there's some good fabs that can, that can produce chips in the US.
And with, like, Intel 18A, for example, that's gonna be Catching up to where TSMC's at, it's a 1.8 nanometer process node, and it's gonna offer, I think, incredible compute capability [00:14:00] for everything, for AI, data centers, and all this stuff. So, um, it's, it's a work in process, I'd say, uh, to get sort of all the manufacturing and electronics capability c- sort of more back on shore, but it's not, I don't think it's out of reach.
Bryce Bladon: Very well said. Just to sort of summarize that, HyFix is developing a, a chip that integrates GNSS positioning, flight control, secure communications, on-core compute into a single American-made package. So when Mike says that's difficult, I hope some of that [00:14:30] context explains why. I... How should I put this? I've never actually met somebody who's been in autonomous systems as long as you have.
Uh, so I'm really curious whether it's something you've historically seen or something that's more forward-facing. What is something people get wrong when it comes to the future of autonomy?
Mike Horton: Well, I think that, that, uh, the amount of kind of field work you need to do to really perfect a product Kinda can get people, catch people off guard.
The [00:15:00] prototype phase of building the hardware and getting some software to do initial stage autonomy oftentimes goes pretty quick, but the number of different things that can go wrong in the field is just very varied, and I think this is a place where AI is still kinda catching up. I'll give an example unrelated to drones, but, you know, recently Waymo had to kinda cut back on operations in some of the s- southeast states where they had just started out kinda running because of flooding, and I, the, the, [00:15:30] the, their, their models and their training data just had not seen enough flooded roads to really handle that safely.
So these are the kinds of edge cases, and they've been at that for well over a decade. And, and so that wringing out edge cases around these kinds of products is a hard, hard-fought battle, and a lot, I think, can get a lot of customers, can get a lot of companies kind of off guard and, and end up with unhappy customers, no revenue, difficult [00:16:00] conversation with investors, and you know where that goes.
Bryce Bladon: Absolutely. Something that is really remarkable about, um, one of the last projects you created, GEODNET, which I am, uh, intimately familiar with, partially from that point you were bringing up, which is when autonomous systems meet real world applications, to put it simply, you encounter frictions. I think it's just a, a physical reality, particularly as a lot of people get very used to things being digitally frictionless.
When they take their ideas into the real world, you see all the, all the things that cause friction. But something [00:16:30] that is extremely remarkable about GEODNET, it's decentralized, there are tens of thousands of ground reference stations, um, and it enables more resilient and accurate positioning than GPS alone, which is relevant to our discussion, easily jammed and spoofed.
It's in over 145 countries, the largest RTK network globally, generates millions in annual revenue, and this is the part that really matters to me. It's a DePIN project, which is jargon that means decentralized physical infrastructure, which is also, in the [00:17:00] world of crypto and blockchain, probably what I would call the biggest proving ground for is this a real idea?
I, I had a, a project manager from Spexi on the other day, and as they put it, the single best way to determine if a crypto project is real is to look at their revenue. Is it a real business? Is it active in other countries? Do they have enterprise partners? And for the record, GEODNET does have enterprise partners, like DroneDeploy and Questel and, and Frotobots.
I might have said that second one incorrectly. These new company names are killing me. Nice. Anyways, what I'm getting at is [00:17:30] building an American-made drone seems like one of the hardest challenges somebody could take on, but frankly, so does building a decentralized, uh, ground reference station network, uh, to build more accurate positioning than GPS, and you've done that.
And I would also add, and this is the part that I am so excited about, is how HyFix seems to build on your experience and, and GEODNET's foundation. I'm just really excited about what your experience implies about what's coming on- Because, I mean, let me just ask, where do you see the biggest opportunities in the next three to [00:18:00] five years?
Mike Horton: Well, I think drones and robotics, physical AI is going to be, uh, absolutely Tremendous business opportunity. Clearly right now, LLMs and generative AI is, is really where the action is at. But I don't think you have to look too far forward to see something like Waymo scaling to millions of vehicles, and that really driving a whole new type of economy where physical work can be highly automated.
And given physical work [00:18:30] is like 70% of the total GDP on the planet, I think it, it could be incredibly transformational and really free up people's time, um, kinda completely. And so I, I'm super pumped up about that. Drones is a big part of that. Um, drones is, I think, an early win, and it's already happening at pretty amazing scale.
Um, there are, you know, 20 odd million drones produced every year. Um, some of those for consumers, some of those for [00:19:00] commercial, some of those for agriculture, some of those, a small number right now, for drone delivery, but I'm sure that will grow. And, you know, some of those obviously used in, in the military.
So it's an example of, like, how effective physical AI can be, but I think there'll be more and more physical AI to come. Ground robots, humanoids, all that stuff, it's, it's just a, it's just right really around the corner. And I come back to why. It's, it's really the cost of the hardware is tractable to build the equipment, and [00:19:30] the AI and the compute is there to make it work well.
So while the tuning cycles may be a little bit longer and, you know, take a bit more time to get through 'cause there's safety considerations and practical things like waterproofing and stuff like that to work through, um, that you wouldn't have in, say, a generative AI model, on the other hand, there's more...
You know, there can be very, very direct tangible payback for these things. I believe, like farming, for example, what an important and huge industry that is just right [00:20:00] around the corner, I think, to be completely automated. And that's gonna allow for, you know, food prices to go down and, you know, much more effective use of just the land and water and gas and resources.
So that's an area where you could really see full autonomy at the robotics level just making an enormous societal impact.
Bryce Bladon: Absolutely. What, what would you say a healthy domestic autonomous system stack actually looks like? And, and you mentioned it's around the corner. How far away are [00:20:30] we from it actually?
Mike Horton: Yeah, I think we're there. I think it's, a lot of it's still just putting the pieces together. The-- There's some great building blocks. So for, like, the H1 chip, we focus on two kinda core open source projects as, uh, kinda baseline infrastructure. One is the PX4 autopilot, and a huge shout-out to the creators and all the supporters of that project over the year to build it into such a nicely architected piece of code with really good baseline autopilot infrastructure, [00:21:00] including advanced Kalman filters and good controlling, control loops, and just all the stuff you need.
And then Robot Operating System, ROS 2, provides you the infrastructure for more advanced vision, um, and, and bringing in models and all the plumbing you need in terms of data logging and kinda sharing of data across different tasks within a autonomous system. So those are two core pieces of software infrastructure.
Sensors are there, um, whether you're [00:21:30] a, a, a camera-only advocate or you like camera plus LiDAR. Obviously for outdoor positioning, RTK is huge. We're really excited we have, uh, a partnership both with GEODNET, um, tracking low Earth orbit satellites, and then in the HyFix chip Uh, using those low Earth orbit satellites with a company called Xona Space Systems, which is launching these satellites.
That gives you a signal that's 100 times stronger than GPS, but kind of acts like GPS. Penetrates deep indoors, gives you precise timing reference, um, allows, when it's full constellation, it [00:22:00] will allow for positioning both indoors and, and much more contested and interfered environments. So whether it's a cell tower that's interfering with GPS, um, or it's a actual active jammer, it's gonna be much harder to corrupt these low Earth orbit signals 'cause they're just way stronger.
So the infrastructure, sensors, uh, compute, these things are there. There's certainly a gap in practical experience. There can be a gap in regulations. There can be over-regulation. I'd ca- I'd argue in the case of drones that some of the things that are holding [00:22:30] back the really massive usage of drone technology has been regulation that's been maybe too conservative in some cases.
I think if we, uh, certainly within the United States, if we look at kind of how quickly China's moved on the low altitude economy, they've, they've moved more aggressively, and that's probably, that's something that I know the current administration is trying to fix, and it's, it's important because we want that to flourish.
That's a huge win. Think about every time you do a drone delivery of food, how much more... The total efficiency of that relative [00:23:00] to driving it to you is just extraordinary. Yeah. The, the cost, the actual physical cost to fly a drone from a supermarket to someone's home in the neighborhood is probably, like, the electric- electricity cost for that's probably at most $1.
And compare that to having somebody get in a car, get on a road, add pollution to the environment, get stuck in traffic just to get two miles from the grocery store to your house. I think you can, we can all say that drone delivery, you know, offers a huge benefit. [00:23:30] So seeing that these kinds of tech expand, that, that ties into regulation.
I think some of the autonomous car technology also ties into regulation. And I get we need to be careful and, and be safe, and we also need to think through privacy around these systems. But at the same time, they offer, I think, a real way to move society forward.
Bryce Bladon: Exceptionally well said, Mike. I notice, uh, we're just at time.
So first of all, I wanna say thank you, but second of all, where can people follow you and learn more about what you're doing with Hyfix?
Mike Horton: Yeah. So our website is hyfix.ai, [00:24:00] and we are active also on X and LinkedIn, so I recommend give us a follow on whichever of those two you like, and look forward to keeping up.
Bryce Bladon: Absolutely. I will make sure there are links to all those things in the show notes of this episode. But once again, Mike, thank you so much for joining me today.
Mike Horton: Thank you, Bryce, for having me.
Bryce Bladon: And thank you all for listening Thanks for being a part of the Drone Network. Subscribe wherever fine podcasts are served to get a new episode every week, and remember to leave us a five-star review on your podcast app of choice.
It helps a lot. [00:24:30] Today's show was sponsored by Spexi Geospatial and LayerDrone. Learn more about standardized drone imagery built for global scale at Spexi.com. That's S-P-E-X-I.com and LayerDrone.org. Thanks again for listening