The Future of Autonomy

Summary:

In this engaging panel discussion, Tyler Sweatt is joined by Donald “Chee” Gansberger, Salsa Lopez, and John Goodson to explore the evolving landscape of autonomy in defense technology. The conversation covers the impact of commercial tech on the battlefield, the importance of adopting commercial standards for interoperability, and the future of autonomous systems in contested environments. Listen as these experts share their insights on the shift towards more adaptable, resilient, and cost-effective autonomous systems that promise to redefine the future of military operations.

Transcript:

Tyler Sweatt (0:19)

Tyler Sweatt, honored to be joined by a esteemed panel here. Talk a little bit about the future of autonomy. I'll let everybody introduce themselves real quick. Just a quick 30 seconder.

Yep.

And then I'll sort of dig into the conversation.

Chee Gansberger (0:34)

Chee Hansberger, I'm a Program Manager for a program called Rapid at AFWERX the robotics and autonomy platform for interoperable devices. I'm also the Software Manager for AFWERX Prime.

Salsa Lopez (0:47)

Hey all, Salsa Lopez, [inaudible] assigned to AFWERX, working as the Robotics and Machine Learning Lead.

John Goodson (0:55)

Hey Everybody, John Goodson, CEO at Darkhive, former Navy combat tech sporting West Coast Seal teams. Yeah. Doing drone stuff now.

[Chee Gansberger] Hi - I'm a JTAC [Joint Terminal Attack Controller].

Yeah.

[Chee Gansberger] I was like oh, we're going there.

I'm surprised you didn't volunteer that immediately.

Tyler Sweatt (1:11)

All right, so as we start, I think recent events, you know, especially the last couple years have I think shown the impact that, you know, drones, autonomous systems, commercial software, can have on the battlefield. I think it's come a long way. I remember I think the first time I was getting Ravens and some of those, we couldn't figure out how to use 'em so we just crashed 'em into people. We're now able to use them to do a whole bunch more. What do you think the implications, and Chee I'll give it to you first, as you look at sort of Ukraine, some of the stuff that's happening there, also look at some of sort of the proving ground aspect of all the commercial tech coming in. What do you think sort of the implications are broader for US National security and defense, how we're looking at some of our larger programs and then where do you think we have issues that maybe we're not seeing yet?

Chee Gansberger (2:06)

Well, I'd say the biggest impact, how you worded that is not even so much just, and I know this panel's about autonomy, but I'd reach onto the commercial side, because the moment you said the Raven, I just think of as a JTAC all the different systems I've been trained on and it's been a ton. The ones that were commercial were really easy to get my hands on and figure out. And the ones like the Raven that came from a prime built by a bunch of requirements documents that went through a program office, we would just crash them into people because that was the only effective use.

[Tyler Sweatt] Worked.

You couldn't, I mean, no one was gonna ever actually use it. What we're seeing in Ukraine is taking that commercial off the shelf stuff and because it iterates so rapidly, the dudes on the ground are able to actually develop TTP [tactics, techniques, and procedures] and then of course the enemy develops TTP. And you've ended up with this really, really concentric OODA loop of not just acquisitions but tactics that are building on top of each other only because commercial innovation is happening so fast. We're now seeing that innovation being so fast leaning into autonomy for software can actually now overpower the speed of the human mind that's involved in these, you know, jamming or remote controls where there is no remote control. I just, I let the robot go do it. But we also have to step back and say, hey, we're gonna take the gloves off and let commercial innovation especially, the US commercial innovation and artificial intelligence, kind of take the lead on that. Otherwise we're just gonna get buried by other peers.

Tyler Sweatt (3:43)

Salsa as you look at concepts of autonomy and sort of, I think there are a lot of words thrown around. I think DOD is great at acronyms and program names and buzzwords and they usually sort of, maybe they skimp a little bit on the technical understanding. Where are you seeing sort of pockets of actual adopters that are looking and trying to build in at that technical level and really understand how we're looking at interoperability and autonomy? And conversely, if you wanna name and shame, where are you seeing sort of resistance or maybe fluff without some of the meat?

Salsa Lopez (4:19)

For sure, so, I'll start with where we see things going down range, right? It's this push into commercial, the push for this OODA Loop, right? That needs to be so fast. In order to do that I need to adopt the commercial, not the government standard. And so I think in the DOD previously when we built these large weapon systems, we generally would create standards and we have standards everywhere. We have MISB standards, we have lots of really great things that actually do allow us to do good information sharing, good data sharing. I think as we lean into what I will call "robotics SPICE autonomy" and there's a reason for that. When we lean into the robotics sector and we lean into something like UAS, those things have already been developed. And if I'm gonna pull something from the commercial side, it's likely also going to come from academia. So I need to lean into both of them and what they're using as their standards and what they are using is ROS MAVlink. And these are protocols we already have. So why do we as a government believe that we will create a better standard? One that will be more widely adopted, likely not, and that we will not incur any tech debt by adopting that new standard that we now require someone else to need their product too. And so folks like DIU have kind of pioneered what's called the robotic autonomous systems interoperability protocols. And through the kind of joint small UAS working group, we've started to build protocols that are based off of ROS based off of MAVlink on what we expect your robot to tell us. Not to say that this is what we need to have as a standard for a new protocol, but how are we going to use the existing protocols? And I think that that's part is really important, because now I've reduced my technical debt, when I go and I'm starting to apply a new technology that maybe just came out of academia or out of industry towards the TTP, to make those OODA loops much more concentric.

Tyler Sweatt (6:10)

So from a commercial side, you know, some what sort of building platform coming in looking at, right, how do we not have all of these, you know, tightly integrated sort of vertical silos of capabilities but we're actually building scalable capability? John, I'd be curious one, where are you sort of seeing, you know, the government maybe leading in the right way and then conversely sort of what's the implication for you as a commercial provider of maybe some of the disconnected nature of how the government's been approaching autonomy at scale robotics, so to speak initially?

John Goodson (6:50)

Yeah, great question. And I think Salsa already hit on this actually really well, which is from commercial industry's perspective, when they're being directed, if they wanna work with the government, they're being directed to utilize a particular architecture, a particular protocol, whatever it is. I mean, assuming there's a lot of engineers or people who lead engineering teams in the room and what's the first question is, okay cool, where's the documentation? Where are the resources that I can look to, to better understand how to leverage this platform? And as someone who has been outta uniform and working in defense tech for a number of years now, primarily as an integrator, I can tell you that having gotten that direction from a government sponsor and being told, you know, go find this resource over here. And I say, okay, that's fantastic. I found it, but it says that I need, you know, I need permission from a government sponsor to do it. Yeah, call this person. Okay, I call that person. No, don't talk to me, talk to this person. Okay, talk to that person. Okay, cool, you need a SIPR token to be able to get on a classified network in order to access this. That's frustrating for me and forget me telling my engineering team that anybody who runs an engineering team, particularly on the commercial side, what's the answer eventually to a question like that is, screw it, we'll build it ourselves. Like I'm not gonna waste our time doing this. And that is really beneficial to whoever the likely traditional defense prime is, who's building that system and was actually advocating quite heavily for it to be on the other side of a classification barrier so that they could control it and own it and define standards that only their systems are interoperable with. These are the things that, all the way back to the question are really important, important considerations for me that Salsa hit really well, which is you're not gonna do better than ROS, it's already bought, it's already broadly adopted, it's industry, it's academia. You're not gonna do better than Mavlink. You can come up with all sorts of reasons why, you know, you need to add custom messages and things like that. But you're not gonna do better than a- Particularly if you make it a government standard that exists within a silo and is completely inaccessible to anybody else. The success of whether it's, you know, the stuff that you know, Salsa and Chee are leading and they understand this quite well or success for really high profile programs like Replicator. All of this is going to really hinge on if we want to create an ecosystem of partner software and hardware that we expect to rally around these types of architectures, they gotta be discoverable. Especially if you wanna pull commercial industry in, that means that when your engineering team is told, you know, hey you gotta integrate with this thing over here and they say, fine you yet another thing that I have to go and learn about to integrate with. But when they actually go and Google it, they can actually find an API defined, commented out very, very well. They can find examples, they can run online fully simmed, you know, examples of their software running on a representative platform. Like these are things that if you do that, that gets the engineers really excited, that gets guys like me who aren't engineers really excited, because you made things really good for my team and it made it happy. You know that that's the type of thing that I think about a lot from the commercial side, building solutions that notionally will plug into some of these types of programs.

Salsa Lopez (9:58)

I think and if I may, I think it's also really important-

[Tyler Sweatt] You may.

On the government side to provide access to those tools. And so one of the things that we've started adopting on the military side is Stitches and we got your team on there. So www.stitches.mil, not an AFWERX program, but it's another government program that has-

[Chee Gansberger] It's about to be.

Yeah, sorry, it is about to be, we're adopting it.  It is an IL4, ITAR repository, where I can have Jupyter Notebooks, I can have a GitLab and I can get, it's IL4, it's ITAR, I can get contractors and I can have there and I have now a repository of Git that is government owned, that lo and behold has lots of other things, like messaging standards. So whenever I ask the team over at Darkhive like, hey, I would like you to integrate with these messages. I can point them to a repository that is documented that we can put GitHub issues. That goes back to the developers, right? And that's so important. And many times even within the government, when I'm told of something new that someone would like to integrate, either with our product or that we wanna integrate with someone else's product, the first thing I ask them is "Where's the code?" And even me as a government person, has a hard time-

[Chee Gansberger] Oh god.

Getting hold of the code.

[Chee Gansberger] Yeah.

(11:06)

Or they don't understand the code, 'cause they maybe never ask for it. Maybe it's GPR and some things are, and there's an argument to be made for non GPR code, but for the stuff that is GPR, government purpose rights, where is it? How do I go and touch it and did it actually deliver? And a lot of times the government teams, unfortunately, will ask for government purpose rights but never actually get the code. And when they do get it, they realize that it wasn't quite what they wanted. They got the right product, but not the quite the right underlying code that they needed in order to continue expanding. Or we have so much technical debt, because we haven't been maintaining it and now we are stuck having to rebuild it. And that's another massive problem.

[John Goodson] Yeah. If no one else, I mean, yeah-

[Tyler Sweatt] I felt like he was gonna talk, so I paused.

[John Goodson]No.

[Tyler Sweatt] Stutter or an idea?

[Salsa Lopez] Yeah. I think I just, I just made him happy.

John Goodson (11:52)

You made me real happy, but I mean the, the other side of this, I mean, is always, it's a little bit, you know, scratching one layer deeper that and when it comes to whether it's, government owned code, government owned application, industry or otherwise, and this notion of open, you know, when a solution is put out there to be open, to be government owned, whatever it might be. Open to me has always meant, again, coming from the perspective of a systems integrator, open to me means that there's good documentation, there is an interface that is well defined, that I can easily point someone to and they can very quickly get to work and actually begin interacting with whatever that application may be. And that if I have trouble, there are resources that are available to me to help me leverage it effectively. And those are kind of the three tick boxes for me. You know, where's the code? Certainly, but, how easy is it for someone who has no familiarity with that code to get it up and running, to do build from sourced and or to troubleshoot things independently without having to get whoever the originator of that, of that application of getting them involved effectively. Like that's open to me. That's kind the standard.

Chee Gansberger (13:11)

I'd say where we're gonna see a lot of bang for our buck, especially AI, ML and autonomy. When you start looking into, what we're trying to do with robotics is, and I'm a keyed on this from you saying GPR, I have a suite of problems, like just a whole lot of things that we're trying to get after. And some of them are very niche to the government. I have a just a set search on Amazon. I keep waiting for them to build me drones optimized to go kill an S400. It's probably never gonna happen. So we're gonna have to go write that code ourselves. We're gonna have to go write a lot of things that go do military specific use cases. Obviously we want the GPR for all of that 'cause those are our problems. We have to solve our own problems. But there's so many other things that have a dual use case, not just from a hardware perspective, but a software perspective. Leaning on both of those as much as possible. It's easy to make that case for hardware, because you're gonna buy a lot of hardware en masse, it's harder to tell that story about like, I would rather just buy licenses from a commercial entity for this dual use software, than go build my own in my own software factory. A lot of people think, well then what, we would own it? Well then I would own something that was only good the day I built it, and for as long as I'm willing to maintain it at the level I'm willing to maintain, if it's got a large market, there's lots of competition and those vendors are going to innovate and I would rather just buy those innovations, as as they get better and better years down the road. So, for how we look at this is, now I need to rapidly onboard commercial partners, which is why Tyler's on this panel. But that's difficult to do.

[Tyler Sweatt] Thought it was my moderation capabilities.

(14:58)

You bring bring a little something to the fight. But yeah, it's being able to get those commercial partnerships, especially as if we're really gonna take advantage of the US having a pivot ahead of our near peer competitors. We have to look at what the US market does well, we don't manufacture well compared to our near peer anymore, but we do innovate in, especially in IT, as an information economy way better than anybody else. So we need to be leaning on that commercial advantage. And that's gonna take a pivot in how the military acquires things.

Tyler Sweatt (15:28)

So using that as sort of the launch of the next section here and we think about, you know, this last conversation was around sort of building community documentation, right? The right stakeholders around the table and starting to scale that. And you sort of turn the corner into changing that relationship between the communities, right? Which then begs the question, how are we thinking about it from procurement or you know, congressional sort of regulation and then what does that look like? You know, flash forward two or three years and what does the program look like? How are you, what are you guys trying to put in place? You know, I'll leave it broad, so you can go wherever you want with it.

Chee Gansberger (16:11)

As soon as you say congressional. I think of, you know, have some friends that worked in congress eight, nine years ago, that honestly the congressional tools, the legal tools are already there to be doing innovative things. It is, this term gets thrown around a lot, but the "frozen middle", it is reinvigorating the acquisition arms to like look at these alternative, AFWERX exists as one of these, DIU exists as one of these, but we are fringe players. We're the lessons that we have for the DOD, especially in like commercial partnerships, are not the standard in acquisitions. So I think that needs to change.

Salsa Lopez (16:50)

Yeah, absolutely. I think ideally as a government specifically in the robotics space and we're talking about delivering these capabilities, we are trying to figure out, I think right now with our government partners on what does that tactically look like? What Airman is going to be flying these things, what airmen is going to be doing, the logistics supply chain, charging them, updating the software, pulling the data off, what does the data pipeline look like? These are all massive problems that we don't have answers to yet. And we have some folks across the world in different little bespoke units, they're helping us try and understand how to connect all those things. But in three to five years, I think what I would love to see, is a community of practice and web text surrounded around this tactical problem of robotics and how are we best using them, and how are we connecting both of our ML operations as well as our robotics operations. So for ML, how are you moving data from point A to point B and how are we effectively connecting data to the right places in the cloud at the speed of relevance? Are we updating and training our computer vision models and then redeploying them? And how are we verifying that those models and the code that's coming in is going to be something that has an acceptable level of risk for that robotics system to go and and move forward? So ideally on something like Second Front’s Platform or other platforms as well, that means lots of commercial companies coming onto the platform and being available for licenses that these Airmen can go and pick up, so that this company maybe does really good ML ops. That's likely something the commercial sector is gonna do better. More than than the government, right? Because everyone does this, this isn't a government problem, right? But training the specific computer vision model, might be an Airman function, right? For this specific thing, for that specific mission. But the infrastructure can be theirs. Same thing on the platforms, right? Darkhive has some great platforms that they're building for us and they're gonna have some software that needs to probably go through some CVE testing before we deploy it back. And then someone else may come out with some new software that's open source. Maybe ROS 3, ROS 5 comes out in three years, right? As ROS gets updated, how do I push that through the platform and get it there and now start having new heterogeneous and homogenous capabilities? How do I enable the Airman to grab whatever robot is on the shelf, 3D print whatever parts they need, optimize their sensor, their shooter, their radio packages and their all of that onto a platform with the right software, with the right ML ops, put it together and send it to range? With an ATO and send it down range knowing that that is all been okayed? That would be the ideal three to five years from now. And that happens at the tactical level, with support from headquarters helping build all of the scaffolding, allowing them to move through that process.

Tyler Sweatt (19:33)

Yeah, flatten that kill chain and allowing time to value from a kinetic standpoint, or a sense standpoint to get pretty quick. 

For John, on that,

[John Goodson] Mhm.

So taking that concept sort of where Chee and Salsa started talking through it, and then juxtapose that with-

[Chee Gansberger] Nice word.

Tyler Sweatt (19:50)

I had to come up heterogeneous and homogenous. We're gonna juxtapose now guys. With disconnected contested spectrum, with real sort of A2-AD, I think from my cheap seats, I think we often talk about disconnected and edge capability, we talk about all of this capability and then I go back the sort of last 20 years of combat and we fought uncontested spectrum and we had a really hard freaking time with it. So you know, someone building those commercial platforms, thinking about this operational picture and the sort of modular ability for that Airman or that Soldier Sailor Marine to have a degree of ownership and to be able to deploy. How are you thinking about ways you design, ways you build the use cases, sort of all that?

John Goodson (20:45)

Yeah and it's good 'cause it actually, kind of, I think ties back to how Chee opened us up, and in Ukraine today and when we look at what's working, what's not. What's working today is large volumes of commercially available small drones, that are extremely low cost relative to anything the DOD is producing today. You know, the dollar figure that I put on, you know you hear a treatable a lot, you hear disposable, the dollar figure I put on that is $3,000. That's the cost of a Mavic 3 Pro from DJI and that is the platform that they're largely losing 10,000 a month two, right? And that's a really important number and it's a really important concept, because when they're losing 10,000 drones a month, they're not shaking their fist in anger. They're very happy, because the cost per kill, relative to the amount of hardware they're taking out, with those cheap drones is well worth it. And so when I think about, subject of the panel, future of autonomy, what Chee started us off with, again, and you know what Salsa's been hitting here is, in an uncontested environment we get the luxury of all of these capabilities. Connectivity, the ability to constantly monitor health and status of drones, being able to pull F and B feeds all the way back from the edge to do ML based analytics. Well far away from the edge. But when we say that we need these low cost systems to be able to function autonomously in field and environments where GPS is gone, when comlink is gone and we still need to deliver outcomes, we don't need them to be the absolute best of the best. All we need to do is make 'em a bit more survivable, a bit more autonomous in field so that we start to take 10,000 down to maybe only 3000, down to only 1000, even though we still put 10,000 in field. So we increase their effectiveness, their ability to survive in these environments. The ability to take human out of the loop for some of these critical decisions, again to deliver real outcomes in these types of environments. And it's not this, you know, crazy, AI, super future Skynet world, right? It's just, can I get a little bit better local positioning with visual inertial endometry?

[Tyler Sweatt] Good becomes great.

Can I get a little bit better global positioning with maybe not GPS enabled solutions, but maybe preloaded imagery data sets, elevation data sets and being able to recognize global positioning, local positioning and do it with alternative means that don't require com. And that right there is a relatively simple approach with existing technology. This all exists today. There's nothing like, oh man we gotta build all this stuff. It's like, no, it's just the right level of autonomy to get us to take what works today in Ukraine and make it a bit more, a bit more survivable, but still keep the systems really cheap. That's the future where I look at it.

Tyler Sweatt (23:34)

I love it. All right, we've got one minute left. So each of you 20 seconds or less, as you think about the future of autonomy, what are you most excited about?

[Chee Gansberger] Everything Salsa said.

[Salsa Lopez] Yes.

I got to get on my-

[Tyler Sweatt] Homogenous.

I got to get on my soapbox. I don't love the word autonomy. I would use robotics, because when we talk about robotics, we talk about sensing,

[Tyler Sweatt] Didn't you pick the name of the panel?

Salsa Lopez (23:58)

I-? Maybe. We talk about sensing, we talk about computer vision, we talk about trajectory planning, we talk about visual we talk about specific things that robots need, to be effective in the future that all robots need, ground, air, sea and underwater. Now we start moving forward for the future of autonomy.

[Tyler Sweatt] Yeah.

John Goodson (24:14)

Yeah. I'll just say everything that we're talking about right now, if we're successful in doing all, everything that this discussion has been about, it can actually influence a change in the broader ecosystem of platform designers that makes them more open or easy to deploy third party software to, and we start to break open a lot of the silos that are really impacting our ability to move quicker, get more kit to field that's more interoperable, where we don't wind up with F22s that don't talk to F35s, things like that. That's kind of what I think this opens the door, the pathway to, especially if we get more and more industry performers building their platform specifically to be interoperable with the backend that Chee and Salsa are developing.

Tyler Sweatt (25:04)

That's awesome. I'm trying to get us, keep us on time. So thank everybody for spending a little bit of time with us guys. Thanks for all the brain juice.

[Salsa Lopez] Yeah.

And thanks to Rise8, for always putting on an unbelievable event. So thanks everybody.