Cybersecurity, faster product development are key benefits that Skydio achieves with a U.S. manufacturing base.
By Mark Shortt
When the U.S. drone manufacturer Skydio started as a company in 2014, its co-founders had been working on autonomous flight technology for several years already. Abraham Bachrach, Adam Bry, and Matt Donohoe met in the mid-2000s as grad students at MIT, where their lab was one of the first in the world to start putting computers and sensors on drones and trying to get them to do “smart stuff,” according to Adam Bry, Skydio’s CEO.
“The motivation for starting the company in 2014 was looking out and seeing all the amazing things that people wanted to do with drones, but feeling like none of it was really going to work the way people wanted unless you could trust the drone to fly itself,” Bry said in an interview with D2P. “It’s a very simple idea, but it’s technology that we knew a lot about and loved working on from our backgrounds, and it just seemed like it had enormous potential in the market.”
Bry and his fellow team members still believe that today, but even more so as they see their products come to life in their customers’ hands. Skydio’s drones are on the rise, continuing to turn heads in sectors from construction to bridge maintenance to the military and public safety. In January, Skydio received the CES 2021 Best of Innovation Award for Drones and Unmanned Systems.
“I think it’s a really exciting time for the industry. My co-founders and I believe that the whole market and industry is still very early, and there’s a lot of work to be done to build products that can really scale and serve customers the way that they should,” Bry told D2P. “We’ve made a lot of major technology investments to get to where we are now, but we continue to double down on that, and we have a pretty exciting [array of projects] that we’re working on that I think opens up more and more of these things over the next few years.”
As an innovative company with clients in the defense industry, Skydio takes intellectual property and supply chain security seriously. To help safeguard both, the Redwood City, California company relies on a U.S. manufacturing base that reflects its commitment to supporting domestic component manufacturers. Bry said that Skydio’s domestic manufacturing base helps the company develop its products faster while maintaining close control of quality standards.
“We’re always looking to find suppliers that are cutting edge in their particular material or area,” Bry said. “This applies across the board for all the mechanical and electrical components. And whether it be molding or machining, or forging, we have really strong partners that have powerful capabilities that are key pieces that make our product possible.”
Skydio COO Mark Cranney echoed that sentiment in a blog post in which he wrote, “Domestic manufacturing is a part of our DNA. We believe there is no better place to build the world’s drones, with the world’s most advanced AI, than the nation that leads the world in AI — the United States.”
One of Skydio’s suppliers, Arris Composites, partnered with the company on a redesigned airframe that replaced a 17-part assembly with a single, multi-functional structure, using Arris’s Additive Molding™ technology. The technology combines continuous, aligned fibers and electronic components within topology-optimized structures. It is said to provide the strength and stiffness of titanium at a fraction of the weight, enabling the Skydio X2 to increase its range and speed.
According to a news release detailing the accomplishment, the unique properties of Arris’s Additive Molding carbon fiber allow Skydio to optimize not just the strength and weight of the X2 airframe, but its radio signal transparency as well. The technology also enables Skydio to change material composition within regions of a single part. By leveraging 3D-aligned, continuous fiber composite materials for complex shapes, Skydio can optimize its carbon-and-glass-fiber layout to meet functional requirements of individual regions of the airframe.
Bry said that to achieve its goals, Skydio will need to sustain its R&D and go-to-market execution. Today, Skydio employs about 200 people, approximately half of whom work in R&D. He also said that regulatory frameworks, especially for flying drones beyond the visual line of sight (BVLOS), need to be modernized to expand market applicability and use cases.
Another challenge, given estimates of future market demand, will be building enough drones to keep up with demand. A recent report from Levitate Capital forecasted the global drone economy to grow from $15 billion in 2020 to $90 billion in 2030.
“Autonomy will be the driving factor that will unlock this growth by reducing barriers of adoption, opening up new use cases, such as remotely piloted docked drones, and allowing regulatory modernization for Beyond Visual Line of Sight (BVLOS),” Bry said.
Adam Bry took time out to talk with D2P recently about Skydio’s technology, its commitment to innovation and supply chain security, and how technology innovations are impacting hardware designs. Following is a transcript of our conversation, edited for length and clarity.
D2P: What were some of the technology developments along the way that culminated in getting to where you are now with the Skydio X2 Drone?
Adam Bry: Our core focus is autonomy, and it’s primarily a software problem. We’ve invested a huge amount in building a software system, and our goal with that is essentially to replicate the skills of an expert pilot. The ultimate goal for the product is that, for anybody who’s using one of our drones, it’s as if you had the world’s greatest drone pilot there, flying it, for whatever task you’re trying to perform, whether it be capturing amazing video for consumers, or inspecting infrastructure for businesses, or getting situational awareness for first responders or for soldiers.
There’s a huge amount of software technology, but we also made a big bet, which, I think has paid off for us, that you can’t really make that software work unless you develop it very closely alongside the hardware. So, throughout the life of the company, we’ve tightly interwoven hardware and software development to make the products that we have today possible.
There are a number of really interesting innovations throughout the stack—a few that I point to that I think are pretty significant and lead to the level of performance that we have today.
We’ve developed a fully deporting base perception system that takes input from multiple cameras and uses deep neural networks to predict the 3D structure of the world around it. It’s a really powerful system that overcomes a lot of the failings of conventional computer vision algorithms that get confused by smudges on the lens, or sun flares on lenses, or big, textureless surfaces—things that are kind of visually ambiguous, but also things that humans have no trouble resolving, because we bring a bunch of contextual information. So, when we look at a scene, we see contextually—I know this is a building, or I know this is a pond, and that’s why it’s reflecting.
The deporting perception system that we’ve developed has that kind of contextual understanding built in, which makes it really, really powerful. And that’s pretty tightly coupled to hardware decisions that we’ve made as well.
X2 has six 4K Fisheye Camera navigation cameras—three on top, three on bottom. So, it sees everything in every direction, at 4K resolution. And I think we’re one of the only—if not the only—shipping devices that are using computer vision on rolling shutter Fisheye cameras, which brings with it a huge host of algorithmic challenges, but is also really, really powerful in the amount and the quality of the visual data that we get.
So that was a big bet that we made in the core technology that’s going into X2 that’s worked out really well for us. And it’s an example of hardware and software being pretty tightly coupled to get to the end product that we have today.
D2P: How important are the innovations that Arris’s Additive Molding technology brings to the manufacturing of the X2’s airframe?
AB: Arris’s technology enables consolidation of a 17-piece assembly into a single part while offering a 25 percent weight reduction. This impressive performance gain was achieved within fairly solidified part and system-level constraints as we became aware of Arris’s capabilities during the development of X2. We are tingling with excitement about what we can do for next generation products with their capability in our toolbox right from the start.
We feel strongly that the capabilities of our products reflect the pace of innovation at Skydio. Our partnership and collaboration with Arris highlights this. We went from being completely unaware of Arris’s capabilities to productizing it and bringing it to market in one year. This is just one example of Skydio’s bold decision making and pace of development.
D2P: What makes the X2 innovative and unique?
AB: X2 is bringing together a lot of really amazing technologies. It’s got our core autonomy system with the NVX2, which is really the most powerful computer that you could possibly fit onto a drone of this size: six 4k navigation cameras. And, with X2, we’ve put that together with a really robust, hardened airframe design for enterprise users. So, I think that combination is really powerful and unique.
One of the things that we’re really excited about is the partnership with Arris Composites. I think that’s an example of a company that has figured out an incredibly cool and innovative way of applying 3D printing style technology, along with conventional molding work, to get incredible results. And the component that we’re using from them is an integral part of the airframe, but we’re also very excited about what we can do with that partnership going forward. We see all sorts of possibilities for using these parts that have impossible geometry and incredible material properties to make our products lighter and stiffer, and higher performance.
D2P: I understand one of the advantages is that properties of a part can be sort of fine-tuned according to the specific region of the part.
AB: Exactly. One of the things that they do which is really powerful is they have the ability to direct the fibers at a pretty granular level—like at an individual strand level, in every piece of the part. Which means that you can use the expected stress load to line up exactly what you think the fiber geometry needs to be to get the properties that you’re looking for. And that is really useful.
D2P: What are some of the opportunities and advantages that Skydio sees in manufacturing in the U.S.?
AB: We started manufacturing in house in 2016, and the reason we did it was because we felt it was our best path to get to the best product in the quickest way. A drone, in a lot of ways, is a cutting edge aerospace device disguised as a piece of consumer electronics. So, that tight coupling between hardware and software that I mentioned carries over to manufacturing, and we’ve invested a lot in building up that competency as a company. All the software and processes and tooling and fixtures, as well as the people aspect of managing a manufacturing line, is really important to being able to manufacture these devices and have them be reliable and high quality.
So, having manufacturing very tightly coupled to R&D has been a huge thing for us, and we’ve benefited a lot by being able to manufacture close to home.
And then, if you look at our supply chain, the most critical component with the most coupling to the software is the processor and the silicon that we’re using. And the suppliers that we’re working with there are Qualcomm and NVIDIA, both of whom make cutting edge chips that have unique capabilities that you can’t get anywhere else. And so, I think having close partners like that, which are U.S. companies, is another real advantage that we get by being a U.S. based company.
D2P: How does the IP protection and cybersecurity required by your customers, particularly in the defense sector, impact what you require of your suppliers?
AB: Skydio extends FAR and DFAR clauses to our suppliers, requiring our subcontractors to safeguard DOD information. DOD currently regulates contractors’ information systems security through DFARS 252.204-7012 (Safeguarding Covered Defense Information and Cyber Incident Reporting), which we require all our subcontractors to undertake. This DFAR clause requires compliance with various cybersecurity standards (e.g., NIST 800-171) based on the data being handled by the contractors (including subcontractors). Skydio takes precautions to only load software and IP to our products on our internally owned and managed systems and networks.
D2P: How do you ensure that your software, which you develop in-house, has the most robust cybersecurity?
AB: Security is a critical part of our software design and development process. By developing all aspects of our software in-house, we can consider security holistically, and as a fundamental part of the system design.
Skydio is a software company at our core. We combine those resources with world-class expertise in cybersecurity across all levels of our tech stack. We conduct internal and external security evaluations of our product on a regular basis, enabling us to iterate and improve our cybersecurity posture on a continuous basis. We leverage industry standards and best practices for encryption, communication, and system integrity.
The Defense Department has recognized the fact that our defense and enterprise products offer especially high levels of cybersecurity and supply chain security. Last year, DOD selected Skydio as one of five Blue sUAS companies, solidifying our status as a trusted vendor to the military and other federal agencies. That designation followed exhaustive testing—including reverse engineering—to ensure our products met the highest levels of cybersecurity.
D2P: Skydio has stated that it sources its processors from U.S. companies, to provide a high level of supply chain security and serve as a trusted partner to government customers. Do you source any other electronics within the U.S.?
AB: We source all critical components from U.S. and allied vendors and go to great lengths to source components in the U.S. where we can.
D2P: How would you summarize the role that AI plays in Skydio’s drone?
AB: I think the simple answer is that it’s at the center. Everything about the hardware is designed to support a really powerful AI system. And in contrast to conventional, manually flown drones, the software system was architected with modern software flexibility in mind.
This doesn’t matter so much to the end user—they don’t see this—but the drone is running a modern flavor of Linux Ubuntu. We have a lot of modern software frameworks that we’re using to run all the different algorithms on board, which gives us software flexibility. It means that we can continually ship software updates that make the thing smarter and better and add new functionality. And it gives us the horsepower that we need to run the really advanced algorithms that give the drone its capabilities. So, I think, a very common reaction we see from our customers—and customers that have the most experience with drones are kind of the most pronounced in this—is that there’s really nothing that flies like a Skydio drone when you get it in the air, because of the AI capabilities that it has.
D2P: Have advancements in artificial intelligence impacted the hardware designs? How have your engineers worked to optimize components for AI?
AB: The most direct things that I would point to are integrating the really powerful computer in the [NVIDIA Jetson) TX2, which comes with a lot of hardware work and hardware challenges. That computer consumes a lot of power, it generates a lot of heat, and we’ve come up with some pretty creative ways of dissipating that heat and managing the thermal properties without adding weight to the airframe, by leveraging the structures and airflow that are already happening inherently with the drone. And then there are the camera systems that give us an omnidirectional, high resolution, multicamera field view of the world.
So, when we sit down to design the drone—any new product—the first step is, the hardware architect sits down with the folks on our computer vision team. We play with a lot of different camera placements, and we optimize everything to make sure that we’re going to get the visual field of view that we need for our algorithms to be successful.
D2P: Can you point to any major design tradeoffs, or compromises, that were made?
AB: I can point to one. The product that is in market now is Skydio 2, which is a smaller, shorter flight time, more consumer focused product. It doesn’t have folding arms because the cameras are out on those arms, and there’s a lot of complexity associated with having a camera, where you care a lot about the calibration, on an arm that has a folding mechanism. But with X2, we were able to overcome that challenge, and X2 does have folding arms. So, that was an area where we made a compromise on the first generation version of the product that, thanks to a lot of work across hardware and software, we were able to overcome on the second generation.
We try really hard to get to a product that is state-of-the-art and competitive with all the typical drone things that you would care about, in terms of flight time, and top speed, and camera resolution—all the sort of typical hardware specs, but then has the thoroughly advanced, AI autonomy capability on top. So, there are slight tradeoffs: The drone is a little bit heavier because it’s carrying an extra computer and camera, but the wins that you get from ease of use and from efficiency more than compensate for that.
D2P: What are some of the new use cases that you envision for the X2 drone?
AB: I think that there’s kind of an endless stream of interesting things that people can do with drones, but I would also say that, for the point the industry is at today, I think a lot of the most interesting work to be done is making existing use cases, or use cases that people are kind of experimenting with today, work really well and work really reliably with autonomy.
I’ll give you one example of that. People have been talking about and wanting to do bridge inspection with drones for years now, because conventional methods of bridge inspection are very expensive and slow and dangerous. Oftentimes, you’ll send people climbing around manually on the underside of a bridge, or you’ll use something called a snooper truck, which is a giant piece of heavy machinery that rolls along the top of the bridge, and then has a big cantilevered arm that hangs underneath that people can sit in, and visually inspect a bridge. In the best case, it’s very slow and expensive, and, in the worst case, quite dangerous. Sometimes, these things tip over and can cause all sorts of risks and damage.
So, being able to use a drone to fly underneath a bridge to inspect it seems to make a lot of sense, but it’s also incredibly challenging. And conventional, manually flown drones, basically, just fail because they’re reliant on GPS. And as soon as you find the bridge, you lose the GPS signal and, even if you had some other way of knowing where you were, there’s still a bunch of obstacles to run into.
There were a couple of examples of really world class drone pilots doing bridge inspections with conventional drones, but we felt like there’s an enormous opportunity there to make that work better and more efficiently with autonomy. It becomes much easier for an operator to just say what they want to look at, and the drone does the rest and captures the data.
So, that’s one example. Another one is house inspection. So, this is especially relevant for insurance claims processing after a wind or a hailstorm. The conventional methods of doing house inspection involved a person getting on a ladder and climbing around on the roof, which is also fairly dangerous and expensive—sometimes, people fall off. They also risk damaging the roof while they’re up there. So, it makes a lot of sense to send a drone to do that job, but it’s very difficult for a human pilot to fly the thing, avoid all the obstacles around the roof, and capture the kind of precise data that you need. And that’s something that we’ve been able to fully automate with the software product that runs on our drones. We call it House Scan.
There are use cases like that that which are really within reach right now, that people have kind of dabbled in, that we think autonomy really unlocks. But there are also more forward looking use cases, which we’re also very excited about. With autonomous drones deployed at scale, eventually, I think every construction site will have a drone that lives on the site, and flies itself whenever it needs to, to track progress and spot issues. I also think every piece of critical infrastructure—power plants, energy substations, railyards—all these things really should have drones. They can be useful tools for the people that are monitoring and running those sites.
But there’s a nice incremental path to get there, where it doesn’t have to happen all at once, in a big bang. As the products get better and more mature, they’ll be deployed in more and more places.
D2P: Do your customers sometimes have new ideas for use cases?
AB: I think that a lot of our inspiration for the products we build comes from spending time with customers. One example of this is crime scene and accident scene mapping and reconstruction. This is something that really wasn’t on our radar until we engaged more deeply with customers in the public safety space.
We knew that they were using drones to get situational awareness in dangerous situations—being able to essentially replicate what you would do with a manned helicopter, but in a much more cost effective and safe way that can be deployed in way more situations.
There’s obvious situational awareness value there, but it turns out that being able to document crime scenes and accident scenes is incredibly valuable for them. And, right now, they either do it in a very manual process where they kind of walk around and take pictures, sometimes make notes in a notebook. But there’s the potential with autonomous capture, with our drones, to be able to very easily digitize a crime scene or an accident scene. So, you have a complete digital record of it that can be used later if there’s a court case, and by investigators.
So, that’s an example that really wasn’t on our radar until we got out there and saw what customers were talking about and wanting to do with drones. And I think that’s likely going to become a major use case for us in the public safety space.
D2P: With all these diverse use cases, what would you say binds them all together?
AB: That’s an interesting question because it’s really 95—or even more, 99 percent—same stuff. There’s a huge, huge swath of use cases that revolve around capturing some kind of visual data about the world and wanting to do that in as easy and streamlined a way as possible.
No customer, no first responder, no soldier, no enterprise inspection worker wants to crash their drone. None of them want to have to do precise, finicky manual piloting to get in and get the data that they want. So, there’s actually quite a bit of generality between all these different use cases, in being able to have a drone that can fly itself, avoid obstacles, and autonomously capture an area or scene. And that’s really the stuff that we’re most focused on.
There’s also, obviously, big differences. After you capture that data, what a police officer who’s documenting a crime scene wants to do is very different from what somebody who’s inspecting a bridge wants to do. And there are a lot of downstream tools that these folks are already using that our products integrate with to seamlessly fit into their workflows.
But I think drones have this really interesting and cool property of being a very general purpose tool, and you can see that in the market today. Skydio 2 is being used by consumers to film snowboarders, but it’s also being used by Departments of Transportation to inspect bridges, by first responders to get useful information in emergency situations, and almost all the technology that supports that is common.
D2P: Besides the ability to meet really tight specifications, what else do you specifically look for and value in a supplier?
AB: Information security and cybersecurity are key things for our product that our customers care a lot about, and that we put a lot of work into in everything we control, and that we also look for in our suppliers. We do a lot of work to secure our manufacturing line and make sure that access to the code base, and access to the software that’s being flashed onto the device is maximally secure.
And then contractually, with all of our suppliers, we make sure that they’re adhering to the relevant cybersecurity standards. That’s a big part of what we enforce contractually, and also look for when we visit and do tours of partner sites.
D2P: How would you describe Skydio’s company culture?
AB: I think there’s a few things here. Our technology spans a lot of different disciplines. So, we have AI, autonomy, robotics, as well as large scale software systems that span embedded devices, mobile devices, and the cloud; and hardware engineering of many different flavors. Our hardware team brings together cutting edge thinking from consumer electronics, along with a lot of aerospace design principles.
And then, on the customer side, we’re in public safety and defense, and industrial applications, and creative content capture applications. So, the breadth of what we’re doing is enormous, and that’s reflected in our employee base. We have people with all kinds of different skills and interests, people that come from a lot of the industries that we’re selling into, and I think one of the things that we work really hard at, and is one of the keys to getting to where we are now, and is going to be really important for us going forward, is a super collaborative culture, where everybody is willing to bring what they’re expert in, into the conversation, and listen to what other folks have to say, and collaboratively figure out how to solve the problems that we’re working on.
D2P: What else do you look for in people that you’d like to bring onto your team?
AB: We’ve had a tremendous amount of success with folks that, essentially, have that combination of hunger and drive, willingness to learn and try new things, and inherent curiosity. I think that those qualities span a bunch of different ranges and experiences. We’ve had folks that we’ve hired right out of undergrad who’ve been phenomenal and have learned a ton with the company and are now making core key contributions. And we’ve also had success with people who are very experienced, and have come with years, decades of experience, oftentimes in other industries, but come in with a curiosity and figure out how to apply their previous experience in other industries to what we’re doing at Skydio.
D2P: How do you see further developments in technology—whether they’re in AI, or quantum computing, or advanced wireless systems—impacting technology hardware design in the future?
AB: I think that we’re at the beginning of a really exciting period in embedded AI devices. And probably the thing that is most impactful, which really hasn’t come to fruition yet is silicon, designed specifically for AI and machine learning, reaching maturity in embedded devices.
We’re seeing early examples of this in the cloud. Google has a processor that they developed in house that they call the TPU, the Tensor Processing Unit, which gives them enormous efficiency wins for training and doing inference on machine learning models in the cloud. That’s kind of an early proof point—it’s not deployed at scale yet.
There’s a lot of innovation happening, trying to bring that kind of technology to embedded devices. And I think that we’re going to see that wave start to break over the next couple of years. It’s coming from the big chipmakers, as well as a bunch of startups, and that’s going to be profoundly impactful. That’s like an order of magnitude increase in the compute power, oftentimes accompanied by an order of magnitude decrease in the amount of power you’re using.
So, it’s a really significant thing that I think will be probably beyond even the progress that we saw in the heyday of the PC, where CPUs were just reliably doubling year over year. We’re at the beginning of a pretty exciting period for deep learning specific hardware, which I think will be very impactful.
So that’s a big one. All the work that’s going into electric vehicles is also paying off for drones, in battery technology getting better and more efficient. I think that’s one where we’re not likely to see a discrete step change, but just continued progress that makes them lighter and fly longer, which is useful and can open up more possibilities for customers.
Probably the last one, which I think is also on the near horizon, is persistent connectivity through, first, LTE, and then 5G and beyond, which is another really sort of key piece for getting the most utility out of these things.