Orest Pilskalns, CEO of Skyfish, on building autonomous drone infrastructure

With drones from China’s DJI already banned for federal procurement and Congress discussing a full import ban, we spoke with Orest Pilskalns, co-founder & CEO of Skyfish, to learn more about how U.S. drone companies are seizing the opportunity.

Key points from our conversation via Sacra AI:

Let’s start out with just a thumbnail bio of Skyfish, its place in the market, and how it came about.

I founded Skyfish with John Livingston in May 2014 when we decided to get into the drone business. My background was as a professor at Washington State where I had started a venture with some grad students in mapping technology, and John helped out. We had worked together before. John had a successful exit from Absolute Software John was CEO and co-founder of Absolute Software until 2013, when he retired and began working with emerging tech founders. Later in 2023, Absolute sold for around a billion US dollars. As far as entrepreneurs go, somebody who can go from two guys at a desk to a billion dollar exit is kind of a unicorn.

My background is computer science. I'm not necessarily a hardware engineer, although having led Skyfish since 2014 I have a lot of knowledge now in hardware, electrical engineering, and mechanical engineering.

At that time, we met with Jim Reavis who founded the Cloud Security Alliance in 2009\. Jim was one of the first to recognize the importance of the cloud computing space and that it would be the future of data security and storage. Jim also believed that drones were going to be a critical part of the bridge between the digital and physical world. He saw that Skyfish’s mapping technology and navigation software technology could become very important in this space. He encouraged us to run with our vision of leveraging drones to create precision data for inspection and management of critical infrastructure.

In part because of that conversation we saw there was a future in drones. We were interested in drones primarily from a data perspective. As a computer scientist, I wanted to focus on 3D modeling and inspection and what you can get from extreme photogrammetry and high-resolution 3D models. I'd been talking to customers, essentially future customers, and they kept saying: “if we could only measure something within a photograph.”

So we wanted to build the best photogrammetric 3D models in the world. At the time, we had neural nets and machine learning, but it soon became evident that we needed better sensor technology to really be able to model what was there, in order to measure it.

But first we needed a better hardware platform. We started looking at everything in the marketplace hardware-wise. DJI was a younger company then, and it was a good platform. But we couldn't really find anything suitable that worked with our software. After really fighting it for a couple years, we decided to just build a prototype drone and figure out if we could do this on our own. Not necessarily to enter the market, just to dip our toes into the whole hardware thing.

Before you know it, I had an $80,000 milling machine in my barn behind my house. We were cutting parts to build a drone. So becoming involved in hardware was a natural evolution of needing to get deeper and deeper in the system. The ArduCopter drone hardware was open source, and we were building our own versions with major changes. If it wasn't open source and we couldn't get our fingers in it and modify it, then we didn't want to use it. It had to be open source.

So you started with software and backed into hardware, which is kind of unusual. It usually goes in the other direction. Was this software-hardware integration the key to arriving at a final product?

Well, then we realized it was all about the sensor and supporting the sensor. We started looking for sensors and got involved with Sony around 2016\. We saw their sensors and started playing with them. That is really the secret sauce if you want a good model. If you want good data, you need a good sensor. They were making the best sensors, no matter what camera you were using. Their sensor was clearly the best in the world, and it just kept getting better.

We had been buying a lot of Sony cameras at that time. Around 2019, somebody from Sony Japan reached out and asked why we were buying so many cameras and what were we doing with them. We explained we were making photogrammetric models. They said since we were doing something in the commercial space, not consumer, they wanted to come out and see us.

We were just eight people in a startup incubator outside the University of Montana. The Sony camera product team finally showed up in January 2020 when it was very cold in Montana. They came into our offices, and we showed them our drone and how we were interfacing with their camera. Then we went over to the computer and showed Sony the 3D models we were making, and they were blown away.

They said they knew people were using their cameras for photogrammetry, but the output looked nothing like our models. How were we doing that? Well by then we had been working in this space for a while, and we had 50 different trade secrets and 10 patents.

Sony wanted to deepen the partnership and we said their camera API needed a lot of changes. The amazing thing is we were in contact with the head of the program at Sony. We would request a change, and within 24 to 48 hours, our contact over there would have new API changes rolled out into firmware that we could flash onto the camera. It is a great relationship. They have always been very responsive.

We said that ultimately Sony needed to build a different camera. They probably heard it from several people. A lot of people probably asked for something like the Sony LR1. We worked very closely with them, and they said we were the lead North American partner on the LR1 program. Sony started showing up in Missoula, Montana with 12 to 15 engineers from Tokyo and we deepened the relationship from there.

At that point we realized we had become a full-stack drone company. We're building airframes. We're deep in the firmware. We're rolling our own motherboards. We're building our own battery assembly modules. We're doing everything on our own, for better or for worse. It was probably for the worse, to be honest, just because it was expensive R\&D, and we weren't really commercializing it yet because we were small.

What was DJI’s role in the market then? Was it already dominant?

DJI was definitely the 800-pound gorilla just clobbering everybody because they had seen such a large injection of cash from the CCP, and they were building a good product at low cost. In some respects, the dominance was artificial, given all the early Chinese government support they were receiving behind the scenes.

But the US government realized at some point that the situation wasn’t secure. They said: “We need to build our own drone program,” and they blocked DJI, first in the Navy. Before the ban even got to the level where it is now, people started saying we need to start looking at US players. That really helped us and a lot of other companies.

But we’re different in that there aren't many in the drone space that are full stack.There are a lot of players who say they are. They buy somebody else's hardware, throw it inside an airframe they build, and call themselves “full stack.” It's not easy, but there are some great strategic benefits once you’ve achieved it..

What specific market advantages result from owning the whole tech stack for drones?

For example we were at an internal drone competition in 2022 at this large government agency, and they had all the major players there. I was looking at the other companies’ controllers or ground stations, and they were all phone or tablet-based, every single one. Some of the major drone buyers in the US have told me that they can't work with tablets anymore. They overheat. They lay them out in the sun, and they shut down.

We had a tablet-based controller for a while, which was the Panasonic toughbook, and it was ruggedized, one of the only ones that was ruggedized. But it was expensive. There were supply chain issues. They really were not ideal, even though they're Japanese.

So again: on remote controllers, we decided to roll our own. Today we have our own complete remote controller motherboard. We're using NVIDIA’s Jetson Orin Nano, not only in the drone but also in the controller, and it's an ARM-based chip. As far as I know, we're the first to use a Linux ARM-based system in a controller.

Today, our screen is divorced from all of the components, and you have complete heat management and total control over the system. To really be a full stack player, you need that. There are Department of Defense-type hardware contractors that have their own controller, but you're talking an order of magnitude difference between the cost of our drones and those drones.

And on the drones themselves, we were also able to go through this natural evolution. We started with a four-rotor model aka quadcopter, and then we realized our customers want redundancy, and we were able to move to six rotors. That redundancy on the rotors actually helped us land an early government agency customer. We have since rolled out a second-generation platform incorporating all the learnings and improvements we’ve made over the past few years. This is the Osprey drone which I will talk to in a minute.

Can you tell me about that government agency relationship and how that developed?

It started at AUVSI annual event organized by the Association for Uncrewed Vehicle Systems International 2022. We can't reveal the customer, but a government agency came by and said they were having a competition. They really liked our drone, the six-rotor drone, and the specs on it. They liked the redundancy. But they needed a certain sensor. I said, “that's a great sensor, but we can't get our hands on it, and the sensor company is not that responsive to us.”

Twenty minutes later, the head of the drone program at this agency brought over the CEO of that sensor company. He said: “Here's the sensor. Can you show up in six weeks and do a demo?” We said we'd give it a try.

Since we were full stack, we could do it. We could interface with the new sensor. We controlled the firmware and we could make the data flow. It wasn't that hard, since we had the end-to-end control.

When we showed up to do the demo, every contractor—there were 10 of them—had half a day near Washington DC for this contract competition. It was Friday morning, and I asked the head of the drone program if he had seen anything good yet. He said no and said he hoped we brought the secret sauce, the good stuff. I said we'd give it a shot.

At the end of the demo, he said we knocked it out of the park and we were going to get an order from them. We sold an initial order of six or seven drones, around $500K. Within six months, they bought a fleet of 50 and that was for several million dollars.

Probably the biggest thing is that they said they liked how rugged our drone is. They've got thousands of hours on it.

But it's big and heavy, and they need a smaller drone now. That’s why we've been building the Osprey drone, which we just launched as you’ve seen. We’ve been building it for the last couple years. To do so, we’ve been listening not just to the government agencies but engineering companies and customers that have photogrammetry use cases. Customers told us they wanted a smaller 10 pound drone, maybe 12 pounds, that’s super portable and fits in a backpack. That's when we came up with the Osprey and the new controller.

That agency has now bought another fleet of drones. A repeat order is one thing. A repeat fleet order from a major government agency is great news for the company. It makes us feel we’re headed in the right direction.

So this agency was not DoD, given that Skyfish is in the process to get on the “Green List” of vendors cleared for compliant government and commercial drone purchases, but not on the more restrictive “Blue List” for defense purchasing, is that right?

Actually, we've sold into DoD entities, but they had to do additional paperwork to get the purchase approved. Hopefully, that whole thing gets ironed out with the Green List, the Blue List, as it has created confusion.

To be honest, we haven't focused on DoD. We can and we will, down the road. We have some plans that would be interesting for that sector. But we've been mostly focused on government agencies that need ISR Intelligence, Surveillance, Reconnaissance use cases, and then the commercial engineering use cases, including critical infrastructure inspection.

How much of a tailwind have you seen from this whole Blue List and Green List effort to move away from China-manufactured drones? What about the market as a whole?

It's huge right now. It will be for another year, year-and-a-half. It's simply an open window for people to really just pound through that hole that was made, and come out the other end.

We're in a pretty interesting spot right now. For us, betting on full stack was risky, but now it's like, okay, we’re one of only a couple of differentiated players, and we’re the only one that hasn't taken a huge check yet from outside investors.

We've done it lean and mean. We don't have any extra frills. We didn't waste our money. We conserved every last dime. Everybody working here wants to be at Skyfish. Essentially, the employees are all owners—they all have significant chunks of the company. Skyfish did a lot of things right, maybe the hard way. But there are tailwinds right now for Skyfish as a result.

Roughly, what is the breakdown in terms of your customers between the major types, government and commercial?

It's about half and half right now. It's going to be more engineering down the road. Since AUVSI, we’ve seen a flood of inbounds. There's a ton of interest in the Osprey with the LR1 sensor. It's an amazing combination for engineering considering the size of the platform. There are only so many ways you can arrange a battery, a drone's body, and the sensor. The way we did it where we can collapse the whole thing and put it in a backpack and deploy in seconds is really a powerful package for anybody doing engineering use cases—and it also does a great job protecting the payload.

Can you give me some idea of what the customer breakdown looks like within the commercial use cases? What are the main two or three biggest customer types?

We're really focused on utilities, telecom. We started in that area selling into the communication tower industry. It just made sense. Obviously, the work is out in the open, great for photogrammetry. They need precision measurements. I can say we are making the best models of towers with our platform. We partner with Bentley Systems, and they always look at our models and ask how we do that with their software. It's the same sort of story as Sony.

The power industry is the other one. That's a huge one, and you can get lost in that industry forever. We're focusing on those two. There's a lot of business there for right now. Could we get involved in a lot of others? Sure. But those are going to keep us busy for quite a while.

We've got some top secret projects internally that we think are interesting moving forward. A lot of people are interested in drone nest technology, and we're interested in building that technology, but the right way. We've seen a lot of designs, and we've got a few patents pending in the area. Having worked in the drone industry as long as I have, it takes a while to get it right. We're building internal prototypes, and we have something really special at some point in the not too distant future, but I can't say a lot about it.

If you control the stack, you can build the drone that's needed to make that vision a reality. If you're rolling your own motherboards, you can change form factors and sizes and all sorts of capabilities.

Before we jumped in with the Osprey we looked at market leaders. We looked at the DJI 350 and now the 400\. We aimed for the same class as far as size and capabilities. They were in the crosshairs. We knew people were going to be looking for a replacement. We wanted to build something that people would feel comfortable with as far as handling—the controller, the drone itself—and fitting within that DJI spec because we respect the fact that they've done a lot of research in this area. They know this is an optimal size for these customers. We know it too from our own research. We've said directly to the press that this is a DJI replacement for a lot of people.

In terms of “full stack,” that doesn't include the sensor, correct? Can you walk me through how a platform like yours thinks about sensors? Wouldn't it also be the case that customers would want to swap out sensors and use different sensors for different use cases?

It depends. When you get into the government agencies, for right now, what we can see is that they're very focused on one mission and a particular payload or sensor. They don't even want to swap sensors because they don't have the time. They're going to need day-night ISR. That's their payload.

If you get into the engineering side, absolutely. You're going to want to swap. There are some people supporting all sorts of different things, but there's not a whole lot that we've seen that we're focused on. There aren’t many large customers that are going to be swapping out a ton of payloads, especially within a certain form factor.

We've tried to support other sensors besides the LR1. We're going to do some sort of thermal solution in the next six months, and it'll be hot swappable. We'll start supporting new ones after that as we see demand.

The people that are needing five different sensors, they're probably not the people that are going to be buying fleets of drones. They're more going to be the mod shops that are like, okay, I've got this mission today, and tomorrow I need this sensor, and I need some small lighter package.

Most of the large engineering customers, your telecoms and power companies, have their specific mission, and they need maybe one or two sensors. That's what we're focused on supporting: just a handful of sensors that are needed for the industries we're focused on.

I've heard that with current sensors, there's still limitations for certain use cases. We’ve heard that in a typical deployment in the utility sector or with an oil pipeline, drones might flag something like corrosion or leak. But if there is an issue, a team has to then go and confirm it. How far can you get with a drone in these inspection use cases?

We probably have the most experience in the tower industry specifically, but it's the same sort of story elsewhere—thickness of steel, for example. We have a precision engineering-grade system, so we can measure it. There are special sensors that if you're on-site, you need to put on the tower itself. Maybe it's a guyed tower, and you have to do some sort of cable analysis, and you have to put something physical on it to measure the tension in the wire and the sag. Some of this can be done with a drone, but you're right. There are some physical interactions that can't be replicated by a drone.

Maybe down the road, there will be sensors for these use cases, and in that case, we'll support it.

That's why the full stack's important. A lot of times, changes in the sensor cascade into changes within the drone, not just the physical, but the software side, the firmware. We have an Ethernet ecosystem and actually open ports on the drone. People keep asking us, can you mount this or that on your drone? We can if it makes sense. If something new and significant pops up, we're ready for it.

For example, we might have sensors that have a physical interaction with something, essentially flying, landing on something and measuring it, with devices that stick out from the drone. It’s going to come along, but robotic physical interactions are not really something at the top of our list right now. Doesn't mean we couldn't get into it.

You're right. A drone inspection is not going to solve all your problems, but it's going to maybe solve 95%. And the tech is improving all the time: Corrosion, for example, there's some new sensor and hyperspectral imaging for that. We've supported that on our M6 drone, which was also designed for LiDAR, infrared, and 3D modeling. It detects all sorts of things in agricultural type use cases.

Then you have sensor fusion. You might fly with one sensor and then with a second sensor. You want the data all co-aligned. One of the things we're really good at is we spend a lot of time making sure we geotag really well. Doing it really well is different from just okay. Knowing exactly where the center of your sensor is in space and the exact timing, supporting geotagging and doing it down to the millisecond and not being off ever—it’s a big deal. Doing that right and putting the circuits in the software and everything together to do that is tough.

We call it frame-synchronous data. Doing that really well ensures that you can do really good photogrammetry, which we've done, but sensor fusion down the road will open up all sorts of new use cases where—when you do get that second sensor—that is life changing, when you combine it with really other good data and you can co-align it, and you're not spending a lot of time fighting to get it right. That's where we're going to shine. Probably not a lot of people have thought about it. I don't even think some of the biggest players have thought about it too much, but it is a real issue.

So to paraphrase: in owning the whole stack, the firmware and software that runs on the machine, the components down to the metal, you can run all the crucial processes precisely enough that then when the downstream need arises, the data is aligned, tagged, and ready to be fused in any way needed. Is that accurate?

That's exactly right.

There’s a company called Gecko Robotics: They stick robots on infrastructure to perform deeper scanning. Does the future involve specialized robots like Gecko’s working with drones, each leaning into their strengths?

I do think it's going to be something like that, and it's also going to be drones dropping off robots onto infrastructure potentially, with nest technology, what some people call “drone in a box,” which allows for autonomous docking. Figuring it all out is going to take some time, but it's coming.

We are going to have a lot of automation in the critical infrastructure space, and it's going to improve capabilities and reduce downtime. Right now, we all hear about aging infrastructure. Robotics is going to help a lot with that.

Another company we hear a lot about is DroneDeploy. It seems like their whole value proposition is that we'll take your data from any kind of robot or drone that you have, we'll stitch it together, help you make it valuable, and integrate it into your systems. One point of view is that is where all the value is going—the post-flight data and software layer—not hardware. How do you see it?

It's a good point. It's a combination. It gets back to my thesis at the beginning where I wanted to get involved in this, and at the beginning we wanted to be more “DroneDeploy-ish.” But they live in a good-enough world where hyper precision just doesn't matter. If somebody's going to put their engineering stamp on something, they probably aren't going to use DroneDeploy data to make that engineering-grade mission critical decision.

We were always playing in a different category where good enough isn't good enough. You have to be really good. There's a lot of value in what DroneDeploy do. I'm not saying there isn't. But in order to do what *we* want to do, which is data for really high-end engineering use cases, you have to think about the old saying in computer science, garbage in, garbage out. To a certain degree, AI can correct a lot, but it can only do so much. The problem is that if you have bad data, it might look great after AI gets done with it, but it'll be all wrong. No matter how good your AI is or how good your back-end is, if you don't have good data, then you won't have a good product.

That's why we're involved in the hardware. We think that it's going to be an edge over the competition for the foreseeable future because we're going to collect better data than anybody else, and that data is going to create better models.

We offer comprehensive software called Skyportal for storing the data and the models and presenting the models. There we partner with Bentley\] on the rendering engine. But the presentation layer, we built that out on a gaming engine ourselves. Once a model is created by Bentley it’s really a bunch of polygons stitched together. To understand what they are, we've built our own patented machine learning engine for 3D. First we leveraged existing AI tech that’s really good at understanding 2D space. Once we understand those objects, we build them up into 3D, kind of like an MRI. That’s our Machine Language, which looks at that stack of objects and goes, “okay, this is what it is,” now we can measure it, because we have the model.

There are a ton of use cases for that, and that's really where there's going to be a lot of value because if you get the measurements right, and you can understand what's there in 3D space and it's very accurate, then you can make the engineering decisions without having to go back into the field and verify.

So do you have software revenue as well as hardware revenue?

Yes, we have both reocurring revenue from data processing and recurring revenue from the AI features inside Skyportal, which is a hosted service we offer to customers, but also the dashboard, our interface for 3D modeling and management. You can manage your media assets and do analytics on top of all of that.

What’s an interesting downstream application customers have opened up with AI? Or is it more about speeding up processing time and throughput?

A lot of companies don't tell you exactly what they're doing behind the scenes with AI. What we see is that they ask us for certain things.

What we’re seeing is questions like “Okay, can you put this into Sky Portal? We want models or maps that can support this detection or help us find this.” That’s what AI is enabling. They expect AI to be able to do specific things. If you have high quality data to start with, it makes it easier.

Obviously you have to train it, but given the improvements we’re seeing and what people are now getting used to with AI—it's something of a double-edged sword. Yes you can do some amazing things. But customers expect a lot and think it's easy, and it's still quite a bit of work.

There’s a general limitation anyone will observe with AI and that I see in my daily work life and within the company. If you have a well defined pattern and problem, and it's repetitive, AI does great on it. But if there's greater reasoning that has to be done—not always so good.

We spoke to a company that provides drone services to large utilities like PG\&E. They were expressing frustration that while customers don't want DJI on jobs, the available drones are either way too expensive, or not getting the job done. There's a gap in the market where you can get quality, but not for the price.

That's what we're addressing. We launched the Osprey around AUVSI this year. Not a lot of people know about us yet, and we purposely don't want to be too out there—we want to make sure we can service all our existing customers, and that we're servicing the right customers. We don't want to take the wrong type of jobs for a drone. We want our drones to be successful. We want to place them in good homes, so to speak.

DJI is a remarkable product, and sometimes people expect an exact one-to-one replica of what they were getting price wise. We're maybe a tiny bit more expensive, but it's not crazy. We're in the $25,000 to $30,000 price range for our system, with the whole RTK kit. By the time you put all the bells and whistles on a DJI, it's maybe up close to $30K as well. I don't think the price is going to be an issue for a lot of buyers if you’re in that range.

For us, success is how well you understand your customer's use case and how much time you've put in with the drone in that use case in the field. That's why we're successful with these government agencies and ISR. We really understand their use case, and we've done a really good job of executing on it.

People buying our drones specifically for the tower industry or for power or just general photogrammetry and mapping are going to be super happy. If they're doing some GPS denied use-case under a bridge or in a tunnel, they're not going to be so happy because we really haven't investigated that use case. We're upfront about what our drone can and cannot do, and the price is reasonable.

As you grow in a marketplace and get more and more market share, you understand the customer. You might see some growing pains, but it's going to be worth it. Think about DJI as they were building out their product, all the growing pains they had. The question is, can the company roll with it and execute through those growing pains? We've already done it, so I think we understand how to do that.

How should we think about expectations that the FAA will loosen restrictions for beyond visual line of sight or BVLOS drone flights in the US? Will that open up more use cases and demand?

Probably. We're already doing BVLOS for the government agency because they have a waiver. People ask, can your drone do this? We're like, yeah. It has been doing it for the last couple years.

How much is it going to expand the market? I'm not quite sure. It probably will, but I don't see it as being a big game-changer compared to something like the Blue List or just thinking about the government’s DJI ban in general. Those are major changes. The FAA changes are definitely important and will be helpful. But it's going to take people a while to adapt, so maybe it's a long term big change. It's not going to be overnight.

What market needs—maybe in other parts of the value chain or in specific industries—do you think could be met with drones that aren't being met today?

I think people want to use drone nest technology because they want to basically have the drone there on demand, whether they buy it or it’s provided as a service. But today, the cost is not reasonable. The tech is not mature yet, and it doesn't work well. People that offer it basically have cell phone sensors on there.

Nest technology—from the perspective of data collection, ISR, and critical infrastructure—is going to be huge, but it needs to be figured out in a cost-effective way and a safe way. It's not there yet, but it’s going to be the next big change.

You've seen the news on Amazon delivery with drones, there’s more and more of that. Obviously, delivery is going to be a big thing too, but that's going to take longer to work out since it’s more than just a tech problem. Whereas if you get drone technology right, that’s going to move much faster.

Can you give me a quick layperson version of why drone nest technology is important? What business applications are opened up?

Think about power infrastructure and getting people out to monitor lines. Now imagine a drone that can fly five miles in every hop, from nest to nest, on a regular basis. As I work in the power industry, all I hear is that we can't find personnel. Everything is underserviced. It's becoming unreliable. We've got thousands of miles of line.

Now imagine a low-cost solution with high definition that can jump along that infrastructure on a regular basis telling you everything you need to know. Those are the type of use cases that are going to pop up.

In public safety, they're already using this technology in New York and elsewhere—you've seen some of the stories from Skydio. That's going to become more and more common. For example, there’s a traffic accident, and they need to know what to deploy, and they have no idea what it looks like on the ground. With nest technology, the drone's there in seconds. Whether it’s a fire, a traffic accident, whatever. It's like having high-resolution satellite data on demand, and it's going to happen.

But it hasn't yet, and somebody's going to win there big.

If everything goes right, what does the world look like for Skyfish in five years?

Our hope and our goal is to be the American made DJI replacement in the US and any other part of the world that doesn't want DJI. At that point Skyfish becomes a multibillion dollar company. Skyfish wants to be there with products that are similar in quality to DJI, and hopefully better in some ways. For photogrammetry, we are better. That's just a fact. Skyfish’s models are better. Our data's better. Our sensor is better. Everything's better. But like I said, we worked on that use case for a long time.

We also want to be a leader in the nest space. We think that's a multibillion dollar opportunity as well. If we execute, Skyfish is going to be the major US player.

This transcript is for information purposes only and does not constitute advice of any type or trade recommendation and should not form the basis of any investment decision. Sacra accepts no liability for the transcript or for any errors, omissions or inaccuracies in respect of it. The views of the experts expressed in the transcript are those of the experts and they are not endorsed by, nor do they represent the opinion of Sacra. Sacra reserves all copyright, intellectual property rights in the transcript. Any modification, copying, displaying, distributing, transmitting, publishing, licensing, creating derivative works from, or selling any transcript is strictly prohibited.