At the U.S. Army, where he retired as the service’s chief scientist and the head of research and technology, Dr. Philip Perconti helped leaders get focused on the most critical and emerging development priorities, including using artificial intelligence for mobility and maneuver and advanced sensing capabilities.
Now in his fifth year as senior vice president and chief technology officer for Leonardo DRS, Perconti continues to work to solve battlefield challenges related to bringing AI-enabled sensing and computing to the tactical edge, from undersea operations to low-Earth orbit.
Defense News spoke with Perconti this month about how physical AI is becoming integral to military operations – and Leonardo DRS is supporting the transformation. Some responses have been edited for space and clarity.
Defense News: At the Army and in your current role, you helped determine which technological investments should take priority. What does that process look like in this moment when the rate of technology development is happening so quickly, particularly in spaces like counter-UAS and artificial intelligence?
Dr. Philip Perconti: First of all, you have to understand your customer, and you have to be connected to their mission. You have to spend a lot of time with customers at every level, all the way from the soldier or sailor on the ground up through the senior leadership level to understand where their concerns are, where their gaps are. DRS has a great reputation with the Department of War, so you know that’s one of the things that we pride ourselves on, is understanding and listening to the customer and taking the voice of the customer at face value and then planning technology development around that.
Q: Leonardo DRS is at the bleeding edge of counter-UAS, contributing to the Golden Dome missile shield, and building the nation’s Columbia Class submarines, to name a few things. What are the challenges to integrating physical AI into this diverse range?
A: What we’re trying to do is put artificial intelligence at the edge, on the platform itself, so that even if it’s disconnected, it remains mission capable. That level of integration is where DRS spends much of its time and energy.
When you start thinking about AI on tactical platforms, those systems have to operate across wide temperature ranges and in adverse conditions such as weather, smoke, dust, and vibration. All of that matters when you’re integrating AI into a tactical platform in a way that keeps it mission capable and still meets the customer’s price point.
Many AI models are developed in the lab, and then people try to fit them into the platform, only to find they do not work there. The model may be too large, or the GPU may not perform within the required temperature parameters. All of that has to be considered with physical AI. You have to start with the constraints, here is the operating envelope, and here is what the mission requires.
Q: How do you do that?
A: You start by assuming the system is disconnected. It has to make local decisions first, and then, if information is available and reach-back is possible, of course you take advantage of that. But on tactical platforms, that often is not the case.
That could mean a submarine, where you might have a large AI system on board acting as a server afloat; a space-borne AI application for communications, tracking, or sensing; or a tactical platform on the ground or an unmanned surface vessel at sea or undersea. Across all of those environments, the same principle applies.
If we’re really going to deploy AI, that includes both the foundation models people often think of, such as large language models, and what I would describe as computer vision or perception models for target detection, tracking, or classification. Those are not necessarily large language models, but they have their own discrete architectures built around specific tasks.
Everything we do is modular and based on open systems. We also have our own real-time, secure, edge-based physical AI operating system, SAGEcore™, which runs these models in real time. Our company is ready to make that possible.
Q: Where is some of that testing and development taking place right now?
A: Some of that testing and development is happening in space right now. Our GPU-based processor chassis is flying with our communications module today. The idea is to receive signals there, process them on orbit—whether that means analyzing a threat, detecting something, or sending information down so action can be taken against a threat. That is physical edge processing in space.
We also have programs underway with the U.S. Army to bring automation for target detection and recognition into its legacy fleet of thermal sensors across a wide range of platforms. The Army wants to field that capability, but the challenge, again, comes down to price point and space claim. These legacy platforms are already packed with electronics. Finding room to add another box—especially one that requires meaningful size, weight, and power for this kind of processing—is a real challenge for our customers.
So we continue to work with them to demonstrate that we can put these kinds of GPUs into ruggedized chassis and integrate them into their platforms.
Q: You talked about the challenges of developing these capabilities for various challenging physical environments. What has Leonardo DRS learned about how and when to integrate the warfighter into the testing process and what kind of surprises do you encounter when technology leaves the lab?
A: One of the clearest messages from the Department of War is that speed matters most. The priority is to come forward with a good solution and then work through the capabilities and limitations that may come with what is, at first, perhaps an 80% solution for the requirement.
That means bringing users into the process early. We understand the environment, and we can make informed decisions about what is ready to show as a prototype or demonstration. When we do that, we are transparent with the user. We tell them where we know it will work and where we know we have not yet met every specification, especially when some of those specifications may not be critical to what they are trying to accomplish.
One of the surprises you can encounter when technology leaves the lab is that a commercial part may not perform well in the field simply because it was never designed for that environment. That is part of DRS’s differentiation—our ability to understand the platforms, the processes, and the emerging technologies, and to bring them into an operational environment in a way that will hold up. Even if the first version is an 80% solution, we know how to make sure it survives, and if the user wants to scale it, we know how to build it at scale.
Q: As we’ve seen in Ukraine, contemporary battlefields can be extremely instructive about technological needs and capabilities. What have you learned and are you learning from the recent conflict in Iran?
A: You go to war with the kit you have, so you’re always fighting the last war first. The challenge is to get out in front of that. We spend a lot of time asking how to look ahead and how to make decisions based on where warfare is going, not just where it has been.
It is a logical progression to say that if warfare today involves a single drone, warfare tomorrow will involve multiple drones or coordinated swarms that can create real challenges for less mobile forces. You also have to think about how those systems might be used in other applications, from surveillance and intelligence to other mission sets. Drones are not going anywhere. They are here to stay.
The level of autonomy they bring to bear, I believe, will be driven by physical AI, specifically, how much capability you want to put on those platforms and at what price point. We think about autonomy not only from the perspective of drones, but also from the perspective of unmanned vehicles more broadly. We are still learning from all of that, and we are helping shape how our customers think about it.
I define disruption as a technology or capability that causes you to think, act, work, or fight differently. If it does not do that, then it is not truly disruptive; it is incremental. So the next question is: what is the next disruption? In many cases, it comes back to price point. Do I have to use a missile that costs hundreds of thousands of dollars against a $55,000 drone? I think that is a solvable problem.
That’s the problem today, but it’ll be superseded by something else, and so that question is what we’re trying to get in front of. I think AI at the edge, or physical AI, is going to have a lot to say in the future about these problems.