Latest / UnDocked: The Maritime Transformation Show / Path Robotics' Heather Carroll on Robot Dogs That Weld, Welder Shortages and Shipbuilding's Future

Transcript

Path Robotics' Heather Carroll on Robot Dogs That Weld, Welder Shortages and Shipbuilding's Future · UnDocked: The Maritime Transformation Show

• Heather Carroll: what happens is the AI will try in these in simulated environments with the real data in the playground. they're just welding millions and millions of times, It's crazy to watch. It's just you can see it just spin up all of these practices. And what they do is they will encounter ideas and try things that we would never think to try. And then it's better. if you limit the intelligence and the experience just to what we as humans have, are we getting the best possible outcome? Nick Chubb: Hello everybody and welcome to another episode of Undocked. Hello to you, Raoul Harris. Raal: Hello to Unick Chub. Nick Chubb: and a very special hello to you, Heather Carroll. Welcome on the show. Heather Carroll: Thanks, Nick. Great to be here. Nice to meet you both. Nick Chubb: Likewise, I'm super into niche tech, but I think that robotic welding is about as niche as I've ever gone. Raal: Yeah. Heather Carroll: Yeah. Nick Chubb: So maybe let's start there. You don't look like a welder, Heather. So Heather Carroll: Yeah. Nick Chubb: how did you end up in the wonderful world of robotic welding? Heather Carroll: That is such a great question. I get that often, which is totally fair. And you would not want to see me weld either. So this is what the robots Nick Chubb: Hahaha Heather Carroll: are good for. No, I spent about, I'd say 25 years in AI and robotics and automation in my career, but that's always been in supply chain and logistics. So not in manufacturing at all, certainly not in welding. and when I got the call to consider coming over to the manufacturing side and working with Path, I really had to think long and hard about it because in the world of supply chain logistics and in the world of what we do, relationships are important. I had a huge network of people. I didn't know anybody in manufacturing, but my husband's a welder. And so I we've been married 32 years and I've seen him go from being a welder himself out on the floor. To owning his own business and starting his own manufacturing facility about 10 or 15 years ago here in Nashville, where we live. And I watched him go through this time of booming construction in our area. and he hit a wall scaling his business. They were exploding in growth, but he could not find skilled labor and he could not find skilled welders. Eventually, he sold the business because it he just couldn't scale it any further. So when I got the opportunity to think about taking a technology and an automation that I understand really well and marrying that up with a problem that I saw my husband come home every night struggling with, I really got excited about being able to help take technology to solve a real world physical problem that every manufacturer in the United States that has any part of their process involving welding is definitely facing right now. And it's not just here, it's globally. This is an absolute global issue. and it just it was something that got me really excited about joining it. So that's how I got here. Nick Chubb: Nice. When you said your husband was a welder, I thought at first you were going to say he was upset that you've decided to back robots instead of humans, but I can see how that might solve a real problem for him. Raal: I think it's interesting. Nick Chubb: The Raal: because just to open with that as a as a gambit, because so much of the story of automation and fear of robots and and and Heather Carroll: Yeah. Raal: AI will get into into the AI dimension of this as well. but so much of that is about this idea and fear of displacing humans. Whereas you're coming at this from this being a the problem being the lack of humans to start with, not that not you know, that's Heather Carroll: Yeah. Raal: quite an interesting opening position, I think. Heather Carroll: Yeah. Yeah. And it's a very, very important one. It's one that we definitely should spend some time talking about. And I'm pretty passionate about that piece of it as well. Raal: Adding another digital system to the fleet is easy. Getting it to work seamlessly with everything else is where the real work starts. Disconnected platforms create duplicate data, manual workarounds, and unnecessary complexity across a fleet. Insight and SmartC bring it together through a common integration layer, connecting onboard operations, applications and data into one end to end environment. A fleet that's easier to manage, easier to scale, and ready for whatever comes next. Because the best technology doesn't work in isolation, it works together. Connect your fleet with Insight. Follow the link in the episode description. Heather Carroll: What we often lose in the conversation is the actual data around the numbers. And it's really important. so I'll give you a couple of stats that really blew me away as I got into this area and tried to start getting my mind around, okay, how big is this problem? So in the United States, I can speak to those stats specifically. In the United States today, according to the American Welding Society, the average age of a welder is 55 years old. The average retirement age is 63. The number of welders in the US that are 55 and older is 21% of our workforce. So that means that literally less than 10 years, we lose 21% of the workforce. That next band, which is 45 to 55, is 22% of the workforce. So if you retire at 63 and you've got 43% of your workforce within a 10 year window of hitting this retirement cliff, you don't have the people. We do not have young people coming into this trade. And this isn't a job like other skilled labor. This isn't one that you can just pick up. Welding is exclusive. Extremely hard. It's an art and a science, and it takes time to get good at it and to do really critical infrastructure multi-pass type welding, which is what we need to do things like building data centers, building energy and infrastructure, building ships. That is that requires a very skilled workforce that can take eight to ten years to get really qualified. well enough to do this. So we've got a problem. I mean, we this these numbers are not on our side. We have got to have help. And the technology should really be looked at as a tool. Just like you would have any other power tool to help you do a job faster, to help you do a job better, and to do more, that's what what that's how we should be having the conversation. How can this automation unlock capacity and help the skilled labor that we have Raal: No Heather Carroll: do the critical work and do more of it. and that that's what I love about this space and the opportunity to really have a real conversation about the data. Nick Chubb: I want to come back to the shipbuilding point because obviously our audience is primarily maritime professionals so that's going to be a real area of interest and I can see from some of the announcements you've made this year you're looking at shipbuilding in quite a serious way but before we do you mentioned it takes kind of eight to ten years to train a human to become a good welder. One, how do you train a robot to become a good welder? Heather Carroll: Yeah. Nick Chubb: to how long does it take. Heather Carroll: Yeah. So you train a robot. you should train a robot to become a good welder, the same way you turn train a human to learn how to weld, which is through practice. and so that is the piece of physical AI that's so critical. And one of the reasons this is so hard is we don't have the data. with LLMs and things like ChatGPT and Claude. You can go out and scrape the internet for all of this data that's out there and train a model to be able to generate ideas and and and come up with things, right? You cannot do that with with physical AI and welding. The data just doesn't exist. So you have to go get it. And the way you go get it is by actually doing. and the challenge with that is it's very expensive because to train a model to weld, you actually have to weld metal. And on shipbuilding, that metal is really, really expensive. And so what we've done at Path and what makes our solution unique is we've been at this for about eight years. And over the past eight years, we've had all of our robots that have been deployed in the field across manufacturers all over North America. All of that data has been collected. So every single inch of weld that we've done for eight years, it's millions and millions and millions of inches. That data's been collected, it's been labeled, it's been put into a weld world model, and we've created our obsidian model using reinforcement learning to train that robot to practice millions of times to do every different type of combination of weld. So you have joint types, you have material type, you are training the on what happens when I change the travel speed, when I change the amperage, when the temperature changes, when I'm doing a T joint versus a V groove, all of these different scenarios. That is possible when you've got all of this data and it's in a an environment where it's using real-world data, not synthetic data. So that's that's what we do. We have a data farm. We leverage all of the training information from all of the robots out in the field plus the data farm. And we just go through the the process with reinforcement learning of training robots. The outcome of that is superhuman quality. I and give you some some data on that too. I just this is fascinating. I love all the metrics around it. Yeah. Raal: Well t but maybe maybe just before you do, I think Nick Chubb: Hahaha Raal: it might be worth us just giving you a couple of minutes to talk about path and what it is that you actually actually do. 'Cause there's a there's a million threads I want to chase after already, but I think if we just h help the listeners out by giving them a real grounding in tell us about the company. Heather Carroll: Yeah. Yeah, yeah. So as I said, we started about eight years ago. We're based in Columbus, Ohio. we were started by two brothers who grew up in the manufacturing world. you know, they started welding, our CEO started welding at eight years old in his garage, where his dad had a fab shop. And so they grew up in that environment of manufacturing. They both went off to get their PhDs, one in computational neuroscience to train and figure out how to build an an AI model that will think like a human. and that's our Alex, our CTO and Andy, our CEO, worked on his PhD in AI-driven humanoid robotics. So how do you get a robot humanoid to learn to walk and move like a human would using AI? So once they got out of college, they went out and talked to hundred over a hundred manufacturers. To try and figure out what is the biggest problem to solve, what are the areas that are stopping manufacturers from being able to automate and take on more work in the US? and what they found every time was welding. It was just the bottleneck and it was the constraint. Nick Chubb: you Heather Carroll: the skilled labor shortage, particularly in that area, was just acute. and given their backgrounds and their knowledge of it, they set out to start a company that would build based on AI and physical AI. build a model that would be able to train a robot to do what a human could do when it comes to welding. so that was the foundation of the company with the goal of really bringing back manufacturing jobs. They watched the Midwest in the United States that area got hit really hard when all of manufacturing was offshore. And it was in a time where they saw that. They grew up watching that. So their goal and mission was to provide resources robotic labor and resources to help augment what we had so we could reshore and bring manufacturing back to to our areas in the United States. Raal: And today what's the kind of what kind of verticals are you in in in is it just supplying Heather Carroll: Yeah. Raal: with the robots or you're actually doing the the the construction Heather Carroll: Yeah. Raal: contract? Heather Carroll: So we do what Path does is we have an AI model called Obsidian. Obsidian is the brains behind it. So we provide that solution paired with our proprietary sensor system. So the vision system that can see everything that's happening during the welding process. And we provide a turnkey solution. So you'll get the robot, the AI model, the vision system, the robotic cell, the positioners, everything you need. For it to be turnkey for you to put this into a work cell to begin welding. and that our focus in our core verticals are number one defense. and in the defense sector, we do a lot with shipbuilding. we also just signed our first contract for wire harnessing and dexterous assembly. So that will be to do wire harnessing for missiles. so it's it's not just welding, we're providing physical l AI for robotic labor cross-manufacturing. So we're starting to move upstream and downstream of welding. and then we also focus on AI infrastructure. most people don't think about the components that go into a data center. it always fascinates me when I walk around a massive data center construction site and a shipyard, how much those components look alike. They're a lot of the same things. On a ship, Raal: Really? Heather Carroll: you've got to have backup power generation. You've got to have all diesel generators. You've got to have skids and units and foundations that are built up. You have the same thing in a data center. You have the you even sometimes have the same equipment. It's just wild. And all that stuff has to be welded. so that's another vertical for us. and then energy and infrastructure. Raal: And I guess all the I guess all the monitor all the monitoring systems as well and everything so you can kinda keep see what's running hot and all that sort of stuff. Nick Chubb: Yeah, I was chatting to an OEM recently who's in maritime, in gas turbines specifically, and they said they've seen a huge spike in orders of gas turbines, Heather Carroll: Yeah. Nick Chubb: not for ships but for data centers, because it's a Heather Carroll: Correct. Nick Chubb: kind of highly reliable power source and quite efficient. Heather Carroll: It's it is. It's it's and what this is causing, which everyone's feeling it across the industry, is a terrible lead time problem because all of these components require welding and skilled labor to do this, and both of these verticals are seeing huge growth and demand. It's for the same same labor pool. So now we've got AI data centers. Here in the US, we've got five hyperscalers that are spending over a trillion dollars a year to build these facilities. And then you've got the government spending one and a half trillion dollars on defense, with lots of that going towards shipbuilding. And you've got the government spending $1.2 trillion on infrastructure improvements, which include grid modernization, which require all of this stuff. So now you've got all of this money coming in to for capital improvements. And it all requires the same type of product and it all requires the same labor pool. So in the United States, by 2030, we will need 600,000 welders just for those three verticals I just mentioned. And the and you see it, they'll why would you work at a shipyard to do welding when you're outdoors, you're in confined space? It's if it's a hundred degrees outside, you're out there suited up, welding in that heat. If it's 10 degrees outside, you're out there in the cold. welding when you could go work on a data center site. And so it's it th we're really seeing it hit our shipbuilders pretty hard. the turnover and attrition is pretty bad. Raal: again, loads of different points I want to pick up on there, but I just want to zoom in on one thing from the founder perspective. Because for most people, AI is a relatively new thing and we've seen exponential growth in it. robotics have been around for a long time, right? R factories having robot Heather Carroll: Yeah. Raal: elements. And clearly, from what you've talked about in terms of the skill profile and all that sort stuff, this really what you're doing really wouldn't be possible without AI. And yet your founders started from a position of of already making that bet that AI would come and would be able to solve that seems really interesting to me that you you started from a position rather than it having a robotics play and then AI suddenly becoming an enabler. It sounds to me from what you're saying that the origin story was to to be an AI company first. Heather Carroll: Right. And that is critical in this. It is true. most people when they think of robots, they think of an assembly line. They think of automotive, tier one automotive, where you've got the big, you know, yellow fanic arms on an assembly line and the the parts are coming down the line at a rate and those robots are coming in and zzz they're welding all along and it it's the same thing over and over. And that is great for tier one automotive, and that's been around for about eighty years. That is not new and it works really well. But those robots are pre-programmed. There's someone that has written a program, has a teach pendant, and they're out there touching up the program, and it is very specific. And if anything shows up on that line at a different speed, at a different time, the part is not fit up perfectly, it will fail. And you you look at what you do in a shipyard, it is none of that. I mean, everything, the fit up is gappy, nothing's perfect, there is no assembly line for parts to just show Nick Chubb: you Heather Carroll: up. These are massive, massive parts that just the material handling moving of those things is challenging. So the idea of taking that part and moving it to a robotic station. Well, that's just not reality. And so the only way to do that is to have something that is intelligent, that can adapt just like a human would, and that can move to the part. So one of the great things that we did this year is we launched a new product called Rove. Rove is it's our same obsidian and vision system model, intelligence put onto a quadruped. And so if you've seen the quadrupeds like Boston Dynamics spot, We have now outfitted Spot and we've named him Rove. And he is outfitted with our intelligence platform and can now go to the park. So we launched this with Saronic here in the United States at their Franklin shipyard. and it's that there's a video out there and we can get you some of the footage if you want to use it, but it's game changing because now you can have these huge, huge parts in a shipyard and you can have the work. The worker, the robot, actually go to the part and not have to worry about moving the part into an area where you can get the automation. so it's it's those types of changes, the adaptability, the ability to do this without having to pre-program something, to have the robot be able to look at a joint and see that it's a couple of millimeters off and adjust the fill plan based on that. So true real-time adapting. is what makes this possible to be able to move t to to areas shipbuilding. Nick Chubb: I remember 10 years ago those Spot the Dog videos sort of going viral and thinking, what on earth Heather Carroll: Yeah. Yeah. Nick Chubb: is anyone going to use one of those dogs for? And here we are 10 years later, it turns out they can weld. Didn't see that coming. Heather Carroll: They can weld, turns out. No, I don't think a lot of people did. Yeah. Raal: no. Nick Chubb: You mentioned the kind of crisis in welder demographics, 600,000 welder shortage predicted. My view on these things is often net-net they become job creators rather than Heather Carroll: Yes. Nick Chubb: job destroyers, but the nature of the work changes. So when working alongside a kind of a robotic workforce like this, how does the nature of a human welder's work change? Who's working in a shipyard, for example? Heather Carroll: Yeah. I mean, f at first it's not gonna change a ton because we just don't have enough. so Nick Chubb: Right. Heather Carroll: it is more I mean, honestly, if if if your listeners are in the maritime space and they work in these these yards, they know what I'm talking about. It is a huge, huge problem. The attrition rate, the turnover rate in this space, in it what you will find online and through sources here, like the fabricator. is 70%. But what you, when you go to a yard and talk to the yards in the United States, they'll tell you it's 100%. That if I hire, 50 people within a year, I will have lost almost every single one of them. And you can't retain the knowledge. This is such a skilled job to do, especially shipbuilding. there's so many things about the process of what you do in a certain order, how you do it, when you do it. The nav sea qualifications that are required. All of this stuff is learned over time. And that's what makes it so take so long. So it doesn't take eight to ten years for somebody to learn how to weld something. But to learn how to weld a submarine to sub-safe standards, it does. And when we're trying to increase our output of these types of ships all around the world. we have this problem with AUKUS too, which is the program that's between Australia, the United States, and the UK. Every single program is struggling with this this issue with skilled labor, particularly in the welding, across this program. so it's it's a real, real issue that something like what Path is doing is the solution. And day one, it doesn't change what the existing welders are doing. It just gives you more capacity to do that work. Further down the line, the way I would think about it, and what we've seen in other areas that don't have such an acute shortage in other manufacturers is the way that it changes is you now can take that really skilled labor and put them on the skilled stuff. They're not doing the repetitive stuff. You can you can put the robot into the areas in shipbuilding, there's confined spaces. It's incredibly dangerous. I I Raal: Dangerous. Hmm. Heather Carroll: was at a yard last month where it was a submarine yard. And they had this one area of the submarine where the box opening was about this big for the welder to climb in. And he had to climb in and go all the way down this long tube. And he's welding this kind of area and backing himself out, but it's completely confined. So there's two people that have to go in. The welder has to climb in, army crawl down to weld with his little headlamp, and a second human has to sit there just to make sure he doesn't pass out. From the fumes and the gases. this is crazy. And so we have automation and technology that can help with that kind of stuff. Put that human doing something else that's not that dangerous, not that confined, and is harder to automate and harder to teach. There's there are things that the robot's not going to be able to do. There are reach issues sometimes. There are certain types of welds that the technology's not there yet. and so you're always going to have this really highly skilled human element that's needed in these other areas. And that's what you'll see them starting to do, the higher value work. Raal: Hm. I think yeah, I I think that answers it. I th just following up from Nick's question there, when you were saying about you could see these jobs moving overseas and then you brought them back, I think a lot of people will ha find a bit of cognitive dissonance in saying, Yeah, but you're bringing them back to be done by by robots. But what you're saying is that this actually there's there's still a role for the human. What this is doing is is is making that more doable, a better, more interesting career, maybe, Heather Carroll: So w we've got a real yes. Raal: And and that's how you see it, is it? Working you you see a future for this kind of work where a human is is buddied by robots, is having robot minions doing parts of the job whilst still being in control, let's say. Heather Carroll: So let's take a real world example on what I'm talking about. right now in the United States, and I'm I'm sorry I'll just have to use these of the stats that I know, but I think it's important to for people to talk about the real data. So right now in the United States, we have the ability, we have shipbuilding capacity for 0.1% of all shipbuilding done in the world. 0.1%. That's the only capacity we have. And During at right after World War II, we were building thousands of ships a year. We can no longer do that. All of that labor moved out. All that shipbuilding has gone to other places. Over the last 30 years, China has invested heavily. China now has the capacity and they get the orders for 60 to 70 percent of all shipbuilding because they've made the investments. We have not. So now we're trying to increase our shipbuilding here, and we don't have the people to do it. We don't have any. it the yards to do it. We don't have the equipment, the automation, none of that. So now what you're seeing is investments coming back into this space. And just this last month in September, I was at the groundbreaking of Port Alpha, which is going to be the largest shipyard in the world once it's completely finished. And that yard alone is going to be able to provide just the first phase of phase one of this yard. Is going to provide 150,000 tons of capacity. All of the yards in the United States today is 100,000 combined. So this is going to be transformational. And they are hiring 10,000 workers, 10,000 humans to work in this yard. But it's not possible if they don't bring that alongside the automation because they can't compete with the other yards around the world that would get that work. If we couldn't do it here in the time frame that it's required. So what we're actually seeing is the ability to put automation in allows us to bring these investments back and create jobs in these areas. And I've seen it too in everyday manufacturing across the United States. In everyday things making, whether it's parts for data centers, whether it's trailer chassis, you know, we have customers that make all kinds of things. We have see it all the time. They have order volumes that they can't take because they can't hire enough welders to run a second shift. But when they can use the automation to then split the welders they have and put some on first shift and some on second shift with the automation. Now I've got to hire people for that second shift. I've got to have quality inspectors. I've got to have supervisors. I've got to have material movers. I've got to have people there working doing all kinds of things. So yes, it is a net new job creator. It is not a job killer. That's just not what we see in the data when you look at bringing in this type of automation. Nick Chubb: is it worth just talking briefly about Saronic? For anyone who doesn't know them, their primary focus, as I understand it, is building autonomous vessels or autonomous platforms for defense purposes. But I think a lot of the technology could have wider civil uses as well. And as I understand it, this yard that's being built is not just going to be churning out ships for defensive purposes, although that's the kind of primary purpose. But the kind of wider US merchant marine could be built there as well, kind of Jones Act type ships. Is that right? Heather Carroll: That's right. That's a hundred percent right. Yeah. Saronic is going to be doing more than just building for the defense in this area. It's a commercial yard as well. Raal: just just a a qu a quick question about that. I mean you you talked about that capacity. I mean Nick you what you just mentioned there. I think it's worth saying that when people think about defence, obviously you think about battleships and military hardware and stuff, but but actually being in control of your Merchant marine is also a really big factor in in in warfare situations to be able to bring in the energy you need, the food supplies, and certain particular rare things that you might need and might not be able to get domestically. So so I think there's that angle to it as well. But given that focus and how you know how much you've talked about the the the sheer scale of these things, what does this mean for Path in terms of your growth trajectory and that sort of stuff? Can you give us a sense of how big How big your operation is today and and how do you see it growing in the in the near future? Heather Carroll: Yeah. we have been on a rocket ship. I got here about two years ago. and that was I came in as the chief revenue officer of Path two years ago. And when I got here, we were primarily working coming out of COVID, we were working a lot with trailer manufacturers. So it it what we had experienced here was when people couldn't travel. All of sudden they bought boats and ATVs and Raal: Yes. Heather Carroll: all kinds of stuff that needed Nick Chubb: You Heather Carroll: to be put on a trailer and taken to campgrounds or to the lake or wherever because you you just couldn't go anywhere. but then that was leveling off. You don't buy those every year. And so as I looked at what were the markets that really had the most demand, to me it was pretty obvious. It was AI and data centers. Two years ago, it's obvious now, two years ago, we actually did not have as much spending in this space as we do now. so I I kind of wanted us to start focusing on grid modernization and data center component structures. that paid off. But I also thought about if that is not that investment's gonna go down, what would cause it? And it would be a regional conflict. And whenever we have that in the United States, if if the market tanks and there's a regional conflict, our defense spending goes up. So I wanted us to be able to take the technology that I knew would work very well for that same type of welding and start to build those customers and build into that space. So we started just a about a year ago on the defense side. And very quickly, within the first year, yeah, Nick Chubb: was well timed. Heather Carroll: within the Raal: Yeah. Heather Carroll: there, yeah, there's always some luck in it, right? but when the within the first year, we'd signed three contracts. and these three contracts are with, you know, two of the neo primes, one of them being Ceronic. and Huntington Ingalls, which is the largest shipbuilder in the United States. They do all the aircraft carriers, they build submarines. they're one of two major prime shipbuilders in the US. and so later, that was the end of last year, we've continued in this space and we've now expanded. and just in the last couple of months, we signed a $600 million contract with Huntington Ingalls. To be their physical AI partner for all phases of their shipbuilding program they call Hyper. So Hyper is high-yield production robotics facility where Path will be sitting in that facility and they will drop plate at the door and we will do all of the things upstream and downstream of welding. We'll do fit up autonomously, we will do the welding, there will be a quality inspection piece. We have a partner that will be doing the sanding and grinding and coating and painting, but raw plate will drop at the door and out will come complete units. And so that is the type of growth that we're seeing. It's just exponential. And to your point, it's not just in the defense space. I do a lot of work with the Department of Commerce in the United States. And they are they also have a shipbuilding office and it is on the commercial side. So they are working on programs to help bring in commercial shipbuilding into the United States. So we do a lot of work. There's a trade development program where we're working with officials from Japan and Korea and other allied countries to bring that shipbuilding for commercial vessels into the US. So there's a lot of that work going on as well. Raal: You you mentioned I've got to jump on it, 'cause I love a new term. I don't think we've we've talked about it, robotics, we've talked about AI. You mentioned physical AI. That's a a kind of emergent genre, let's say. What's different about I mean obviously, you know, there's robotics involved, but what's really what's different about physical Heather Carroll: Yeah. Raal: AI, do you think, from the the way AI is being more commonly used? Heather Carroll: Yeah. So the physical AI is basically bringing those two things together. It's taking the intelligence that's in the AI model and marrying marrying that up with something physical like the robot. and you need both because for the robot to be able to move and to understand and to take actions based off of that understanding, you've got to have that brain or the AI part. But If you just have the AI and you have no way to physically do anything, well, then you've got Chat GPT or Claude. Raal: Yeah, yeah. Heather Carroll: You can interface with it, you can talk to it, you can write good emails, you can ask it all kinds of questions, but you're not gonna solve a real physical problem with that. You gotta have some way to take that intelligence and take action. So it's the marriage of the two. Nick Chubb: yeah, I think we've probably all seen the Chinese Robot Olympics and all that sort of stuff that's been doing around on the internet the last few months. It does feel like China has become incredibly advanced in a very short space of time on robotics. And obviously China is already in a leading position on shipyards and shipbuilding. It is in the dominant position. It's done it very quickly, sort of in 20, 30 years, as you talked about earlier. When it comes to shipyards near enough from scratch and thinking about where this technology could go, you mentioned you already looking kind of upstream and downstream welding itself. What does a shipyard in the US look like in five years time or 10 years time that could be built from scratch that might put you in a position or might put the US in a position to compete with China on shipbuilding? Heather Carroll: Yeah, they're Saronic, I'll I'll say it again just because it's really fresh. I was just there within the the last week or so. You should go look at the video of of their model because it will give you a great idea of what that's gonna look like. and they're not the only ones. As I said, there are other major shipbuilders in the US that are also working on these types of investments and improvements. An example is Huntington Engels Hyper Program. That is again going to be designed, and the automation within that facility is going to be driven by Path and our partner Gray Matter, where it's all autonomously driven to take raw metal at the door and work that all the way through the process. And so that is the future. That is what is being conceived and built when we're talking about greenfield type spaces. Brown fields are going to be more challenging. You've got this existing infrastructure, you've got existing constraints about what can be done there. So the automation that you're going to see put into those brown fields is likely to look different because you're going to be constrained by all of those other parameters. But there are still ways to automate it. A great example, if you think about, we talked about Rove, the quadruped, if you go into a brown field today and you see the automation that's in the yard. Shipyards are automated only about 10%. So 90% of a shipyard is still manual. The average of manufacturers in the US is 80% manual, 20% automated. But a shipyard's even worse. It's just more difficult. The stuff's so big. and so what happens is you go in, and the one area that you usually will see some automation in these yards will be in the panel line. So people will have the big gantry automation welding the panels together, right? But almost every yard I've been in, at the end of that panel line, there will be 10 to 20 humans laying on the ground doing welds that the panel line can't do. And they're all over the Nick Chubb: Right. Heather Carroll: place. And they're coming off that panel line. So you got great automation, automation, automation, and boom, now you got a bottleneck. Now you got all these humans coming in trying to free this bottleneck up. So imagine something like Rove. that you could use as almost a swarm of capability where you've got 20 roves that can be doing other welding and other things just like humans would. And then when that panels come off the panel line, you can send rove over there to do those welds that are normally you would be laying on the ground to do. So I just think it's gonna look a little different in a brown field versus a green Nick Chubb: you Heather Carroll: field where you can think ahead. and build the flow with what the right automation would be at each step, it's gonna look different than the automation that you might put in a brown field where you're dealing with the footprint you got. Nick Chubb: And following on from that, what advice would you give the shipping industry generally in terms of how to think about shipbuilding? to, I guess, explain why I asked that question, I was speaking to the CEO of a ship owning business a few months back, and they basically said, if you want to build a ship today, earliest yard capacity is sort of 20, 30. The market's pretty well. locked up. There just is very, very few slots to build ships right now. And really, the only option, or one of very few options, but really the only option is China to do that because the capacity exists. And if you don't build your ships in China, Chinese shipping companies will build. will take the same slots and build ships in China and you'll end up sort of getting competed out of the ship owning market as well as the ship building market. So the ship owners across Europe in particular that I speak to often feel that that is the only option. And so if it is possible to kind of build out a highly autonomous shipyard. near enough from scratch on a greenfield site. And it's possible, obviously with a lot of capital, to do that fairly quickly. Where does that leave the rest of the commercial shipping industry in terms of what they should be thinking about when it comes to Heather Carroll: Yeah. Nick Chubb: shipbuilding from 2030 and beyond? Heather Carroll: Yeah, I I think you should look for the areas where people are deploying this type of automation. The advanced automation is what's gonna help make it possible and to do it faster. again, I'll go back. Apologize for going back to seronic, but these guys, I mean, and and if just look at what they not what they're saying, but what they've done. so they bought a yard, a commercial yard in Franklin, Louisiana. I was at that yard October. October of last year, and it was sugarcane fields. Literally, it was this little bit of yard was doing some, you know, commercial ship production and a bunch of sugarcane fields. If you go to that yard this October, they're already producing in one year, they have produced an entire 160-foot autonomous ship. Not only have they produced one, they're on multiple holes of that ship and In three months they will have the capacity to build twenty of those ships a year out of that yard. Twenty Raal: Wow. Heather Carroll: a year. That is and they have the orders. They just got orders for ten of those ships. It's incredible. Nick Chubb: which is, I think that's four times the shipbuilding output of the US in the last 12 months, something like that. I think Heather Carroll: Yes. Nick Chubb: the US produced about five last year. Heather Carroll: That that's why folks like the Vice President of the United States were at Port Alpha in September Nick Chubb: Yeah. Heather Carroll: for that groundbreaking. It's it is groundbreaking. It is it is really transformational. It they will start producing 150,000 tons of sh tonnage of ships out of Port Alpha in 2028. They just broke ground Raal: Mm-hmm. Heather Carroll: in September. it's it's really incredible. So I think what I would tell people is. The same thing we did back in World War II. In World War II, we were told it wasn't possible. We had a massive crisis. We needed to turn out ships quickly. We had no capability to do that. And we had industry come in, people that had never built ships before. Never built ships before. Henry Kaiser came in and said, you know what? We can do it. We're going to build these yards. We're gonna create a way to do this where we can do distributed shipbuilding, where it all doesn't have to be done at one yard. We can build components that can come together. So you can make the components all around the country, bring them together to build the ship. And we went from taking 240 days at that time to build a commercial ship to turning a ship a day during World War II. And we did it because we used new technology and we did new ways of building. And that's what you're seeing right now. You've got commercial companies, venture-backed companies coming in and saying, we can do this a different way, but it's going to take new technology and new thinking and new ways of doing it. And what I would encourage the industry to do is really look at this and say, why? Why would we just say it can't be done? Why not go and try something new? Apply this new technology that can help us do this faster with better quality. We didn't talk about the quality before. But there is a whole quality element to this. When you deal with quality on I'll speak again to the military only because it's the most stringent requirements. ABS and other shipbuilding requirements are out there as well, but for the military, for NavC, it's the it's the most difficult to pass. And when you talk about manual welding, the average first pass yield of welding, which For folks not familiar with that, that's your quality metric Nick Chubb: you Heather Carroll: of I weld something one time it goes through inspection. Does it pass? If it doesn't pass, I have to grind out the weld and now I have to manually repair it. Now, Nick Chubb: Wow. Heather Carroll: if you've got a hull or you've got a part of a ship that requires a hundred passes, and you inspect, and somewhere down in the tenth pass, you've got porosity, you've got to grind that whole thing out. It can take days and Shifts. I was in a yard one time where that thing had been sitting there for months being ground out to be repaired. So quality is huge and it is one of the big things that causes delays in shipbuilding schedule. First pass yield for manual welding is 70%. That means that you have got it's 60 to 70 percent. But let's take the more, you know, happy path of 70 percent. And you've got seven out of 10 welds pass. Three of those are going to be stuck going through that repair process. So what happens in robotic welding, and what we're seeing with our obsidian model, is our first pass yield is at 98%. And what we're able to do, and I'm gonna have my team make sure we send you some of these images because it is just when you see it, it just everybody that sees it, it just blows their mind. What we're able to do is there's this you have a WPS. So your WPS is your weld procedure, and it's going to tell you all the parameters of what is qualified. So you have to keep it at this temperature, keep it at this travel speed, you have to have this torch angle. You have, it's all these rules, right? You have to have this size fillet weld, etc. That WPS is your rule. That's what good looks like. How do you measure that with a human? You measure it after the fact and you destroy the piece or you do a UT test, an ultrasonic test to find out whether it stay the human stayed within those parameters. That's why first pass shields is so hard with a human. With a robot though, we monitor millimeter by millimeter, millisecond by millisecond, and it does not go out of those bounds. It can't. And so what that does is it gives the quality inspector the data on every single piece and component of when it was done How it was done and did it stay within that WPS? And that is what is creating this much higher quality. So we're not even talking about that, but that is another piece that allows you to use new advanced technology to do this faster at higher quality. And I think that that's what people should learn a little bit more about. don't, don't take my word for it. Go, go see that, see people that are doing this. and see the results they're having. It's really just incredible. Raal: You you mentioned Nick Chubb: I'm Raal: trying new things, right? And you also talked about the brownfield side of things. You also mentioned earlier on the benefits of being able to introduce a robot into confined spaces and that kind of thing. And that's where that's where I started thinking about this the Rove example and the where this may end up. When we think to the future with technology, you talked about these giant machines. I mean it these are giant constructions, so you Obviously you want scale there, but it's getting more and more possible to do smaller and smaller kind of machines if we if we need them. So my big question to you is do you see a future where this technology is not only building the ship but is also on the ship? And the the reason that I mention it is because you you talked about the the lack of skilled welders available to shipyards and industry. One of the other roles that's really a a struggle. For people crewing is fitters on board who are actually Heather Carroll: Mm. Raal: doing this kind of work with machinery and and and so on there. And there's all sorts of reasons why that that is that we could we could potentially get into, but but ultimately this could be something that you could see a a use for. Is it something that you're thinking about at par? Heather Carroll: Yeah. so fitting for us means something different. It's it's the fit up to weld when I say fit up earlier. So I want don't want people to think I'm I'm talking about something different because what you're calling out is is very important. My my father, my entire life worked on offshore oil rigs and shipping. So I've been when you say that I'm like, yeah, I I need to remember there's other things around fitting that happens in y on ships as work. we aren't doing that. but is that something that there are people out there and companies out there thinking about? It would make a lot of sense for there to be. It's just those are some of the hardest challenges to solve. getting one of the reasons you don't see a lot of humanoids today. and look, there are companies out there working on this problem right now, and it and we're we are going to get there and we're going to see this. capability, it we just seem to be hitting this inflection point where every day there's some new crazy thing that's happening. so I don't I can't pretend to even know what time frame we're working with. So but I don't think it's I don't think it's 20 years out, let's put it that way. we will we will have this capability. But right now it's with the technology we have with humanoids it's really difficult. there there are problems with stability There are things with just, we've we're trying to build something that looks and moves like a human. And that's very difficult to do. I mean, we've struggled with dexterity with the hands. We've struggled with n the knees. If you look at some of these different humanoids, it's like the way they walk is different because we're trying to figure out like how can we do it in half stability. And that's gonna be required for some of those other jobs and tasks Raal: But Heather Carroll: that you're talking about. Raal: but yeah, I I guess I wasn't necessarily jumping ahead to the kind of things that Nick was talking about with the kind of robot Olympics and stuff. I was kind imagining more something that's just on wheels and you could put into spaces. But even what you were saying there about the accuracy, even if you had something that you was that the human was assisted, we see some uses for surgery and things like that, where they're actually, it because I loved what you were saying about the idea that, y You don't have to then test the output if you know exactly what happened at the point of point of work occurring. That's interesting, isn't that? So is there a hybrid Heather Carroll: Yeah. It it it is. Raal: point? Is there a hybrid point before, necessarily robots are doing that kind of work on board ships where where the robotic augmentation of tasks is is something that might Heather Carroll: It there are companies doing that type of thing, and often it's called human in the loop or teleoperation. but to get the type of quality that I was mentioning, you really don't want someone touching it and making the changes, right? I mean, it's Raal: Right. I see. Heather Carroll: so it depends on the task. I can say that for the Nick Chubb: That's interesting. Heather Carroll: For the quality and the welding piece. And we see it. I mean, it's so interesting. So we'll have operators that are welders that have been out in the field and they if you give people parameters to play with and dial in, they'll play with them and they'll dial them in. And not always, but sometimes you will find that those those manual parameters that people have used for years, and that's just how they've always done it. will not perform as well as when you let the trained model try. It what what happens is the AI will try in these in simulated environments with the real data in the playground. Just think about it as a playground or a sandbox where they're just welding millions and millions of times, an hour. It's crazy to watch. It's just you can see it just spin up all of these practices. And that's happening in the training background. And what they do is they will encounter ideas and try things that we would never think to try. And then it's better. And so if you if you limit the intelligence and the experience just to what we as humans have, are we getting the best possible outcome? Shouldn't we train and allow it to try things that we may not have thought about trying before to find a different result. so It it's an interesting conversation to have. And I think that it's different with different tasks. I will be really clear about that. I think there are tasks that we don't have enough data on yet. Teleop is great and you need it. think about how Tesla trained cars to drive, right? It went out and collected all of this real-world driving data to create simulations and do this training that people weren't even thinking about was happening. You don't even think about it. Your car is just collecting all all of this data everywhere you drive and it's feeding into this training system that is making it better. And It's the same kind of thing with physical AI. Nick Chubb: That's really interesting because I think we see the discussion generally across the maritime industry today when you look at how any kind of application of AI in operations or automation systems is the human in the loop is the kind of gold standard for what's required. But that's almost a kind of frame breaking idea that actually there are tasks where human out of the loop might be significantly superior. And I hadn't really thought Heather Carroll: I Nick Chubb: about it that way. Heather Carroll: and it it's because you have to have the data. If you don't have the data, then Nick Chubb: Yeah, yeah, yeah, yeah. Heather Carroll: you are not able to do this reinforcement learning. You are not able to do the training to get the fidelity that we you would need. But that is what we are seeing. and that's what we are seeing in the results when when we run these scenarios and these tests and compare the the outputs. Raal: think that's an interesting point that maybe worth just getting our heads around a little bit because I think as Nick was saying, I think the the logical conclusion of a lot certainly what we're talking about in maritime is that we we get the machine to watch the human over and over and over again to learn what the human does well and then kind of do it. But what you seem to be suggesting was that you've also got within the model and I'm guessing it's within within your model, obsidian, that they Heather Carroll: Yes. Raal: that they are actually say encouraged but but they are permitted to to go beyond what they've learned from humans and actually challenge perhaps the way that things have been that they've sort of seen things done in if they can kind of find a way of doing it more efficaciously. That's a very long word but Heather Carroll: And they and they have to do it within the bounds of the WPS. It's not allowed, it's not allowed to go Raal: So how does that work though? Give give us an idea. Heather Carroll: outside of the bounds of the WPS. So Nick Chubb: What's WPS, Heather Carroll: that's the weld procedure. So if you go Nick Chubb: Okay, okay. Heather Carroll: and look up ABS will have their WPS, Nick Chubb: Right, okay, yeah, yeah. Heather Carroll: I think Lloyd's Register may have one, NAVC for Navy welding in the United States. They have standards that have to be followed. Those standards, that's the guardrails that we put around our model and our obsidian AI training system, right? So when it's doing the reinforcement learning, it's got to stay within that. And within that, it will create a fill plan. And the fill plan is gonna be how the beads of the weld are laid down, right? And what happens in systems that are, and you'll see, and now it drives me crazy. And I'm sure drives everybody else crazy too. It's like everywhere you turn, this is AI enabled, this is AI, this, AI that. It's just so overused. I almost don't even want to say it because it means nothing to no one anymore. but what what really needs to be happening is you need a vision system that is looking at what the what the robot is gonna see to do to join those pieces together. And it needs to be able to say, okay. I see that this joint is coming together like this, but there's a gap between it. And I I've got to fill that gap now. Because it my CAD file said it was like this, but it's actually like that. And so if I just use a standard fill plan that was pre-programmed with a teach pendant, it's not gonna fill that gap because it thought the metal was fully together, right? That's the problem with pre-programmed. That's the problem with systems that require a robotic engineer to create these programs and then do these teach pendants. And so if the robot starts welding on that and there's distortion because of the heat, it's gonna fail. It's gonna blow through, it's gonna get porosity, it's not gonna, it's not gonna fill that joint the way it should. You cannot have failed welds when you're talking about a ship and you can't have it when you're talking about a utility pole that's taking power over your kids' playground. you this stuff cannot fail. So it's really critical that the quality is there. So what we're able to do is have the robot look at that and say, okay, I know that I've got to stay within these bounds. I can't travel faster than this speed. I can't have a torch angle other than out of these angles. I have to have this many inches of fillet weld because of this type of base plate that I'm welding, right? It stays within that. But what it's able to do is change th other things within that fill plan. And it's able to adjust within those parameters. So what we find in the data is better quality. We also find less consumables. It uses less wire. It uses less gas. It's like those kinds of optimizations that make a big difference. Raal: Well, I've got a question as well. You you mentioned ABS earlier. this is a world where, as a classification society, so they're coming in and they're verifying that everything's seaworthy and all that stuff. Very traditional businesses, has that been a bit of a journey getting classification societies on board with this stuff. Heather Carroll: Yes. Yes. that wa that's been one of the biggest parts of the journey. I I honestly thought it would be getting the government on board, you know, with the defense stuff, getting these big shipbuilders, you know, that that have a lot of existing contracts on board with doing something new. that is that was not the big the biggest hurdle. The biggest hurdle. was getting and it wasn't as so much abs it was more navse in the united states with the government to be open to the the robotic well procedures and what we ended up doing is working with all of the major shipbuilders we had six major shipbuilders in the united states sign on to a white paper that we were able to take to navse and say hey We need this technology. We need to be able to and we'll make sure we qualify it, we hit standards, we run through all the qualification procedures, but we need to be able to use this advanced technology in order to increase our capacity and our throughput. and we just recently were able to get that letter from NavC accepting the interpretation so that we're able to use the technology. So we've cleared that hurdle, which is a huge, huge thing. It took It took quite Raal: Yeah. Heather Carroll: a while to do, but but we finally just last month were able to do that. And then on the ABS side for commercial, we passed the very first robotic autonomous weld qualification for an aluminum vessel for ABS that they've ever approved. and so that was a path system that was able to achieve that. Yeah. Raal: congratulations. Nick Chubb: I Raal: Yeah, yeah. Nick Chubb: Can I ask, maybe by way of wrapping up, it feels like there's been a major shift in the US when it comes to shipbuilding. And I can see that obviously technology is a big part of it, right? So one of the enablers is clearly technology. But it's not the only enabler. And actually, the attitudes of class, I think, is an interesting one in terms of barriers that need to be overcome. I'm not even sure it's massively. political because I feel like for the last 30 years Red or blue ship shipbuilding was in decline And we're in the second Trump administration now and shipbuilding didn't really seem to feature in the first so something's clearly changed I'll be interested to get your view on what it is that you think has changed in the US that's made this such a priority and then what else has to change to make this kind of vision of know, autonomous, high capacity, high throughput shipyards a reality five, ten years out. Heather Carroll: Yeah. one of the things that's changed is we we came out of COVID with a realization that we don't make anything here anymore. all our store shelves were empty. We couldn't get toilet paper, we couldn't get hand sanitizer, we couldn't get masks, you couldn't get food. I mean, there were there were just things that you you couldn't the grocery store shelves were just completely wiped clean. And it was a real wake up call that we don't make anything here anymore. And how do we solve that problem? You can't do it overnight. So you're right, this isn't a red or blue thing. It's not a certain administration thing. It is a realization that we have a problem that we've got to solve. And that starts with bringing manufacturing back to the United States. And it this is why it goes across shipbuilding. in order to bring manufacturing back, we There's a great study that was done by the ARM Institute here, which is a group that works with manufacturers of all types. And they asked, What are the reasons you don't automate? And what are the barriers to automation for manufacturers? And the barriers were the number one was consistent demand. Am I to have something to build? Do I have consistent demand? Because if I don't, I can't invest in my manufacturing facility. I I'm I'm not gonna be able to put capital in if I don't know that I'm gonna have orders to make, right? And so this goes across the sectors. but what has changed is this realization that we have to be able to build here. And if we're gonna be able to build here, we have to start focusing demand to manufacturers that can build here. And one of the things that's changed is at the government level, is look, we can help. With focusing our demand by taking government orders for things and having those orders come to people that are building here in the United States. And that is what is driving this investment from VCs and others that are seeing that there's real dollars being put behind. bringing demand and aggregating it here for manufacturing. So now there's a window of time. And that's coming through infrastructure investments. It's coming through data center construction. And now it's coming from shipbuilding. And all of those things are coming together to create demand that is durable demand, which is going to allow more manufacturing to be built here. So that's one good thing. The second is this thing I mentioned way earlier in the conversation around regional conflicts. if we know that disruption's coming, then to your point, it's not just defense. It's our supply vessels. It's we saw it in COVID. We couldn't get anything. So we've got to be able to accelerate the timeline. Of when we are able to produce more ships here in the US. And that is why the last couple of years you've seen this huge increase in focus and investment from both the the Department of War and the Department of State and the Department of Commerce, which are looking at it across shipbuilding, commercial and defense to try and aggregate that demand so that we can incentivize there to be more shipbuilding done here in the United States. But those are the the two real macro pressures that we're seeing that really are are driving this. Raal: mean Heather, it's been a a really fascinating discussion and really great to to have you on and and I do think this is a leading area that that's kind of we can we can learn a lot from. So I I think it's it's not just been a sort of a discussion around what's happening in the US or what the future of shipbuilding looks like, but it also maybe gives us something of a model of what reindustrialization looks like in a post AI robotic world. So yeah, really appreciate you coming on and sharing sharing thoughts with it. And very impressed the at the recall of stats. I don't think I've had any a conversation with anyone who just Heather Carroll: Ha ha Raal: didn't have to reach for notes to take us through what some pretty amazing numbers in there. So thanks for Heather Carroll: Hey, hey, if you when you when you talk to the CRO, which is their chief revenue Raal: Ha ha. Nick Chubb: You Heather Carroll: officer, you're gonna get some stats. Yeah. We're gonna know some numbers. Yeah. Awesome. Yeah. Nick Chubb: They're to know their numbers. Heather, Raal: Yeah. Yeah. No, so thank you so much. Nick Chubb: if someone's listening or watching and thinks, I really, really would like a robot welding dog, Heather Carroll: Yeah. Nick Chubb: what should they do? Where should they go? Heather Carroll: definitely go find us on LinkedIn. you can find me personally out there all the time. So go connect with me. I love to have these conversations and I love to learn. I'm new to this space. I don't know everything. path, our team. We know AI, we know robotics, we know welding, but we don't know everything. And we can always learn more. So I love having these conversations, particularly with people in the industry and the space that have a lot of industry knowledge. We need everybody working together to solve some of these really big problems. So please do reach out. Let's connect and continue having the conversation. And you can always find path information at www.pathrobotics.com. Raal: Fantastic. Nick Chubb: Thank you so much for coming on the show that was absolutely fascinating. Heather Carroll: Yeah, thanks for having me. Raal: Thanks, Heather. All the best.