WASHINGTON — While Shield AI might have started out with a quadcopter capable of flying indoors in GPS-denied environments, it plans to take its ever-evolving artificial intelligence and autonomy technology to the next level by buying a drone company that has caught the attention of the military.
Shield AI’s purchase of Martin UAV, maker of the V-Bat vertical take-off-and-landing unmanned aircraft system, became official on July 30.
Shield AI is aiming to break into the Pentagon market and has already worked closely with the Defense Department’s technology hub, the Defense Innovation Unit, and it has raised more than $50 million in venture funding since 2015.
Martin UAV’s V-Bat has been evaluated for use across the military services, but particularly by the Army for several years. The aircraft was spotted in Germany at the first European-based Joint Warfighting Assessment in 2018 where it was directly compared to Shadow capability.
And Army soldiers spent the better part of a year evaluating V-Bat’s capability as part of the FTUAS competition. A request for proposals is expected to hit the streets soon. The Army will then select a winner to build a first tranche of UAS.
Martin UAV was also chosen to build a prototype in a U.S. Navy competition for a future UAS and Marine Corps units are flying V-Bat as well.
Defense News spoke with Brandon Tseng, a former Navy SEAL who is Shield AI’s cofounder and chief operating officer, in a July 29 interview on what the acquisition of Martin UAV would mean for his company’s growth and for V-Bat’s future.
This interview was edited for length and clarity.
How has your business evolved since its inception and why is buying Martin UAV a good move for Shield AI?
From the beginning, our mission is to protect servicemembers and civilians with artificially intelligent systems. This mission stemmed from my experience as a Navy SEAL, operating in Afghanistan. But I think what a lot of people miss out on is because there was just a quadcopter and everybody saw a quadcopter. A lot of times people said, “Oh, Shield AI built an indoor quadcopter and they’re a quadcopter company.” One of the things when I was starting Shield AI, there were a lot of customer conversations and so I was talking to Navy SEALs, Army Rangers, Army Special Forces about a problem that I was pretty familiar with — clearing buildings of threats. And that was what led to Nova [quadcopter], but a lot more broadly, I was interested in bringing artificial intelligence and autonomy to the defense sector because my hypothesis was that it could be game changing on the battlefield, in terms of improving mission effectiveness and reducing risk.
So while I spoke to a lot of ground warfighters, infantry, special operations forces, I spoke to a lot of pilots, I spoke to a lot of general officers about their strategic problems. And what became a very common thread, and boiling down hundreds of conversations into a core problem, was we have issues operating in high-threat, GPS-denied and communications-denied environments. From drone operators to F-18 pilots to Apache pilots, I really started to learn about this problem of integrated air defense systems and how that was becoming an increasingly more prevalent threat on the battlefield that was denying our ability to maneuver. And so if you talk to a fighter pilot, and you say, “Hey, would you ever want to go up against integrated air defense systems,” they would say, “No, never, odds of survival are minimal, like you’re just not going to win against 14 surface to air missiles… You’ll get shot down immediately.”
That is the core problem that we look to solve, operating high threat, denied environments, GPS, and communications-denied environments and what that led us to was the solution being self-driving technology applied to aircraft.
We built a quadcopter that can go inside buildings autonomously, without GPS communications, not to be a quadcopter company, but to show the department the power of AI and autonomy and the value it provides in high threat, GPS-denied, communications-denied environments. What it has always been about for us is getting our AI and autonomy stack on platforms of increasing strategic consequence. And so we talked about climbing the unmanned systems food chain. We, a lot of people don’t realize, for the past two years, we’ve been working on fixed-wing autonomy, but looking for the right asset and platform to apply that autonomy stack to, and I guess the criteria for that was it had to be a fantastic standalone capability, it had to be beloved by warfighters. And it had to be ready for integration of Hive Mind, our autonomy stack, on board. And it had to represent a stepping path as we get to these platforms of increasing consequence.
So what about V-Bat convinced you this a platform of increasing strategic consequence and to pursue the purchase of Martin UAV?
I think we first heard about Martin UAV, and the V-Bat, either late 2018 or early 2019. We heard it from a variety of different sources. We heard about it from investors that were familiar with our thesis and our plan of climbing the unmanned systems food chain. We had heard about it from customers, saying, “Hey, it would be fantastic. What do you think about Hive Mind on board something like the V-Bat?”
One of the challenges is we were still a small company in 2019, and you have to prove out, to marshal the resources. To do something like an acquisition of Martin UAV, you still have to prove out that you’re a viable company to investors, a fast growing company with what you originally started with. And so while it was on our radar, I would argue, from a Martin UHV perspective, and a Shield AI perspective, joining forces back then, wouldn’t have been a really viable path.
Both companies had to grow. And in basically, late 2020, early 2021, it was, “Okay, we’re ready, we’ve got the investment and the investors behind us to execute something like this. And now, which aircraft makes the most sense. And we’ve looked at all the aircraft, and not only in defense, but in the commercial space, and where everybody’s playing and have spoken to customers, to consultants, about the pros and cons of each aircraft. It just became increasingly evident that the unique hardware architecture of the V-Bat lends itself to extremely compelling capabilities. it’s just not a limited system. And there’s limitations about some of these other systems out there, that would have been more difficult.
The ability to hover, to stare, to come up from a tree line, or go down into a tree line. That’s just one of them. In my opinion, I think when you look at the logistics footprint, driving down the total cost of capability, it is a tiny, logistics footprint. They introduced the new engine, which enables a 25 pound payload capacity, 11 hours of flight and all that fits into the back of a pickup truck. The payload capacity and the endurance had to be there. And when we started to really dive deep on this system. It just is a Goldilocks.
What is your vision for V-Bat’s capability in the future when bringing your technology on board?
Our goal is to build the next generation defense technology company and something that we talk to investors about, something that we talk to customers about, is what does that look like and it doesn’t look like existing defense companies today. It is a company that has an AI and software backbone or core that is leveraged across these different platforms. It is a software first mentality, an AI first mentality, but at the same time, it matters which hardware you put it on. I do not want to discount the importance of hardware in the military and on the battlefield. But I think it is software and AI first, in terms of the capabilities that are really going to matter at the end of the day. We don’t think about does it matter if a system flies 12 hours or 11 hours, it’s how intelligent is that system. What are the missions that it can operate disconnected on the edge as a team?
The software capabilities are going to be very, very, very important. And so what do those look like, first and foremost, the ability to operate without GPS, the ability to operate without communications. I think a lot of people see operations without GPS, but being able to operate without communications is equally important. And so what you don’t want is a system to say, “Oh, I don’t have GPS, I’m just gonna fly back.” It’s like, no, you still want the mission done. And so that is something that we have demonstrated on our Nova UAS.
Then it comes to swarming, having these things operate in teams. The way that we think about it is it is very similar to the self-driving car industry. You probably heard Elon Musk talk about putting a million robo-taxis on the road to operate in a highly distributed manner. But first, you need a single, self-driving car that works. It’s very similar with Shield AI’s approach, it’s build the highly intelligent system, then scale it, don’t scale out a bunch of unintelligent systems, because you’re just going to get formation flight, you’re not going to get any real value or capability, you won’t if you don’t have intelligent Systems. You need intelligent systems to unlock the the concept of overmatch.
What does that look like? It is training these systems to be able to execute a variety of missions. And there’s a long list of missions from countering integrated air defense systems to reconnaissance to escort operations to sensor emplacement and so we’ve been working on the AI aspect of these mission sets for the past several years.
Last week you acquired Heron Systems, what does that bring to the table?
We’ve leaned in heavily on using simulations that are coupled with reinforcement learning, and basically you design a mission and then you let the system train itself over and over and over and over again, which is what Heron Systems did with the [Defense Advanced Research Projects Agency] AlphaDogfight. It’s how open AI train systems to play hide-and-seek. We are applying those same techniques to these unmanned systems. And that, to me, one of the most exciting things about AI and autonomy is that application of reinforcement learning to train systems to do things.
A lot of people look at Heron as “Hey, they did that AlphaDogfight program,” which, yes, that is something we care very much about, the next generation air dominance program. The system beat actual pilots, like a bunch of F-16 pilots, it beat a bunch of other AI pilots from other companies as well, like the large primes lost to it. But what I think a lot of people sometimes lose sight of is actually a lot of what they’re doing will be synergistic to how we think about AI pilots and autonomy onboard Group 3 aircraft like V-Bat.
What happens if the Army doesn’t choose V-Bat for its first tranche of FTUAS to replace Shadow UAS?
One of the reasons that we acquired Martin UAV is because we believe they have the most compelling Group 3 aircraft on the market and so, ahead of almost everything we think about is that, more than we think about the programs that it’s competing in. It is what is going to delight the heck out of the customer.
Where else do you see Shield AI technology cropping up? I know you have a relationship with Textron for example. Could you be out at Project Convergence?
The core for us is going to be with the V-Bat and Martin UAV and that’s the focus for us. There’s a lot of work to be done there. There are certainly applications. But as you know, you have to have focus, you have to channel your resources in a focused manner.