“As much as we feel like the change is never ending right now, that’s going to be the constant for the next couple of decades. We’re going to have more options for how we increase productivity, how we collaborate with robots and each other, and how we build value. “

Right now, we are at such an incredible moment in history in terms of AI and robotics, and we’re just at the beginning of this journey, suggests Anthony Jules, Founder and CEO of Robust AI. The hard part, he says, is going to be selecting from the incredible abundance that’s available, not, “Is there a way to get X, Y, or Z done. In this episode of the Manufacturing Matters podcast, Jules joined TECH B2B Marketing’s Jimmy Carroll to discuss the current state of AI, and more specifically, physical AI and how it is applied to robots to enable dynamic navigation and perception, which helps businesses today improve processes and become more efficient. The discussion also touches on machine vision advances, humanoid vs. wheeled robots, human-robot collaboration, and much more.

Jimmy Carroll: [00:00:06] Hello, everybody. My name is Jimmy Carroll. I’m the vice president of operations at Tech B2B Marketing. And welcome to this episode of the “Manufacturing Matters” podcast, where we discuss the trends and technologies shaping the manufacturing industry. Today is a Friday, and I have the pleasure of being joined by Anthony Jules, who’s the CEO of Robust AI. Anthony, thank you so much for taking the time. I really appreciate it. It’s a pleasure to have you here.

 

Anthony Jules: [00:00:30] Jimmy. Thank you so much. It’s my pleasure to be on the show.

 

Jimmy Carroll: [00:00:33] Well, I love it. Thanks so much. Let’s get right into it then. So for those who don’t know, tell us a little bit about your background in the company and what you do there.

 

Anthony Jules: [00:00:42] Yeah. So my background is I studied robotics and AI at MIT many, many years ago, then was part of the founding team of a company called Sapient that built very large information systems for fortune 500 companies and had a career there, also doing video games, and then ended up back in robotics, which landed me at Google building robots there for a few years. So my career has taken me through a few large companies and many different startups. And my focus has always been on building systems where people and technology have a very intimate connection. And that’s everything from collaborative robots that I do now to video games to systems that people spend eight hours a day in front of where you really need the user experience to be polished.

 

Jimmy Carroll: [00:01:38] Yeah, a lot of interesting companies you’ve worked for there. So tell us a little bit about the current one and what you’re doing and what you guys are most excited about right now.

 

Anthony Jules: [00:01:47] Yeah, thanks, Jimmy. So at Robust AI we’re building collaborative mobile robots. Our first robot, Carter, is a little bit tongue in cheek. It’s named that way because it actually looks like a cart. It’s a shelf that drives itself around. It’s the most advanced autonomous mobile robot that that we know. But it’s very intentionally made to have a form factor that’s familiar to people, so that adoption is very easy. And it’s designed for manufacturing and for warehouse automation.

 

Jimmy Carroll: [00:02:19] So that’s interesting, right? One of the things that’s come up over the years in terms of robotic adoption and, sort the general public’s acceptance of robotics, is the way they look. I don’t want to venture into humanoids, at least not yet, but is that also sort of an intentional idea of, like, “here are these people that might be working in warehouses or factories or wherever?” We want to make something that feels comfortable and not something that feels threatening to them. Is that something that’s…?

 

Anthony Jules: [00:02:53] Absolutely. That’s a big part of the design decision. I have a few principles —— it’s probably too long for for right now —— but, you know, one of them is every robot is a promise. And the way it looks tells you a lot about what that promise is. This is true of every product, but it’s even more so with robots and things that articulate in the world. So we very intentionally built the robot to have a familiar form factor so that it’s non-threatening, because the promise is we increase productivity, but we don’t replace all the people. That’s not what we’re built around. So that literally comes out in the shape of the robot and all of the experience that you have in terms of how you interact with the robot.

 

Jimmy Carroll: [00:03:42] Okay. So that’s a really good response because it’s kind of a segue into a question that I wanted to ask. Maybe 7 or 8 years ago, somewhere in that range, reports started coming out saying, hey, robots are not actually taking jobs. In fact, they’re oftentimes creating jobs. And it feels like, in the last couple of years anyway —— and this is just my own experience,  you can let me know if you think I’m wrong or if you agree —— it feels like we’re sort of almost getting over the hump of, hey, robots aren’t taking your jobs. They’re here to do specific tasks —— it’s a cliche, but dull, dirty, dangerous jobs, repetitive jobs. I mean, if you look at labor shortage —— there are jobs available now that people just don’t even want to do. So, robotics are kind of —— it feels anyway —— starting to be accepted at least. But then —— and this is a perfect question for you because your company is called Robust AI and you have AI-enabled collaborative mobile robots —— now comes the fear of AI, right? And the fear of AI, it’s mostly in the general public, I would say. In the industrial space, it might lean a bit more towards skepticism than fear. But, it’s kind of a two-part question: one is, when you’re talking to somebody who is outside of the industrial space and they fear it. When you tell them that you work at an AI company, if they don’t already know, how do you let them know that, hey, AI is not here to enslave us or do this or that. It’s here to better society and make jobs more efficient and so on.

 

Anthony Jules: [00:05:22] Yeah. So there’s a few big, big ideas in there. The first thing that I start with is —— you know, I’m honest about it, and I give my real perspective, which is these technologies won’t replace people everywhere. But I’m really honest in saying there IS going to be labor disruption. Jobs are going to change. There’s no way around that. That has been true and accelerated really from the Industrial Revolution through now. And that’s not a surprise to most people. The real question they have is, okay, “so then what kinds of jobs should I be thinking about? Is my particular kind of job going to change?” Then I usually just share more detail about what I think is going to change and what’s been surprising. So I think when people think about robots and think about AI, they assume that it takes away all manual labor. Well, one of the things that we’ve figured out is it’s not actually ideal either financially or in terms of process to get all people out of workflows. People and robots add different value, and having people in the process gives you an adaptability and a different set of quality checks than automated systems give you. And by combining people and robots, you actually end up with something that is higher productivity than over-indexing on either, higher quality than over-indexing on either, and higher adaptability than over-indexing on either one. So there is this win win, which is the future that we want to design, where the combination of people and machines actually get a lot more done and do it at a higher quality level. And that’s really the goal. And the final thing I’ll share on that is —— and we’re at the absolute beginning of this —— we’re designing what these patterns are going to look like. We’re all figuring this out together right now. And because of that, we’re going to see a lot of change before we get to patterns that are stable.

 

Jimmy Carroll: [00:07:42] It’s such a good answer. And that’s one of the things that, working in this space —— obviously not at the same level as you, but we’re in the industrial space in terms of our clients —— when I talk to people about AI is that AI is only as good as the data that you train it with. And people are required, like humans-in-the-loop input is required for constantly improving AI models. You need people for that.

 

Anthony Jules: [00:08:10] Yeah. I think you need people to improve the AI models, but I also think the best systems have people in the loop in a supervisory or monitoring capacity. Because again, what that gives you is all the power of whatever that AI system does as well as the contextual reference and checking that a person can bring to it. So, you know, these systems aren’t perfect and are never going to be perfect. And what’s really clear —— and I think many people have had these examples now —— which is, when they’re wrong they’re wrong in ways that seem so obvious to people that it’s often comical. Because they’re NOT intelligent like us. You know, it really is a very different mode of computing, where these things are trained on lots and lots of examples, and then you build a system based on all those examples. And when you meander into questions where there isn’t a specific answer that it’s been trained on, it does some type of guess in between the things that are closest. And sometimes that’s good and sometimes that goes off the rails. You really do need people, a human in the loop, at multiple levels for these systems to be the most productive.

 

Jimmy Carroll: [00:09:37] Yeah, absolutely. Staying within the context of AI, “physical AI” is a term that I’ve been seeing used quite a bit more lately. And I’m curious: What does that mean to you? And then within the context of Robust AI, specifically, what impact does Physical AI have when it comes to the design patterns for robots that work alongside people?

 

Anthony Jules: [00:10:02] Yeah. So, first I’ll start with the definition. So Physical AI is intelligence applied to systems that move and act in the real world. It’s a lot of what we would just have called robotics previously. But I do think it allows a bit more breadth in definition and it lets people pull in ideas around generative AI and things that are now more familiar to the general audience. And it’s not just about perception, but robots need to understand context and behave accordingly in the world. And it’s really trying to bring a rich understanding of the world into the decision loop that some robot, or even some thermostat, is using to make decisions. So when robots share space with people especially, design patterns shift. So you need… We believe people have to have agency so that they stay comfortable, and then you’re not fighting backlash constantly. And, so that when people know the right answer, they can implement that change immediately and effectively and therefore get more productivity, handle exceptions in seconds instead of, you know, minutes to hours. So you need agency, clarity, and responsiveness in these systems. So our approach blends computer vision, real time reasoning, a lot of user experience and user-centered design so that the way that people interact with these systems feels incredibly intuitive and really straightforward.

 

Jimmy Carroll: [00:11:49] In a past life I was an editor at a magazine that covered machine vision, and industrial automation, too, but primarily machine vision. So a lot of times my brain goes to visual inspection for AI. But in the last couple of years it’s really gone much, much further than that. Given all the use cases we’ve seen —— real world actual AI deployments, whether it’s mobile robots or AI for predictive maintenance or whatever it may be —— what are  some other ways that you’ve seen AI benefiting manufacturing and warehouse environments today?

 

Anthony Jules: [00:12:28] Yeah. So, I think the first one that comes to mind is smarter routing and workflow optimization. That’s been a problem that, you know… I have this joke: whenever AI gets really good, it stops being AI. So there’s a lot of stuff around routing and workflow optimization that has its basis in AI, and it continues to have some parts that are more modern AI, but a lot of it now is just thought of as operations research. But this was all built in AI labs over the last 40 years. So the stuff that’s new is doing that smarter routing and workflow optimization with robots and people adapting in real time, that makes the computational complexity of the problem explode. So that’s a place where AI is useful. You also mentioned predictive analytics and maintenance. I can’t NOT mention that one, because I think it’s so important. And as much as we already do, I think the penetration on that’s maybe 5%. There’s so many more places where we are not yet using the data that we have access to to understand how to keep, uh, have systems have higher uptime and higher productivity.

 

Anthony Jules: [00:13:51] The other piece is really using AI and sensing to have better interaction and collaboration with floor associates —— workers in manufacturing, or floor associates —— in warehouses. This is one of the places where I think WE really push the envelope because a person’s experience with the robot governs how that system then gets adopted over time. And by using AI and understanding where the robot is, where the person is, how the robot’s addressing the person, having the robot present as few choices as possible so that the cognitive load on the person is as low as possible… These are, again, just some of the things that we’ve started to do. And I think we are just scratching the surface on the power there. And I think just fundamentally AI turns data into decisions. It has the ability to improve everything, but there’s a lot of nuance and a lot of iteration that’s necessary for us to get to the right answers.

 

Jimmy Carroll: [00:15:05] It really is more of an observation than anything, but it really feels like, AI is not new, but it feels like it’s the dawn of a new era almost. Like Dr. Andrew Ng says, AI is the new electricity. And then at the A3 Business Forum this past January, there’re industry leaders saying, “hey, industrial automation has existed for a long time, but in a lot of ways, we’re only just beginning.” So it’s going to be very, very exciting to see how AI gets used. You mentioned like 5% deployment, but, how can AI transform everything really? And one last thought there is… I saw somebody on LinkedIn post a meme that said something to the effect of, “AI is not going to take your jobs. People who are good at AI are going to take your jobs.”

 

Anthony Jules: [00:15:59] So I’ve seen that one and I 100% agree with it. It’s funny, I was going to say that in response to what you were asking, because it is pervasive and it’s stunning how pervasive it is when you step back and you look at the AI that we’re talking about. So generative AI that’s available to the broad audience works on text and works on generating images and generating short amounts of video. That is all these systems do so far. As much as, you know, I’m a roboticist and we’re building an AI-enabled collaborative robot, we are JUST starting that journey. If ChatGPT represents what something like chatbots can provide in terms of value —— and let’s say they’re halfway into that journey, or 25% into that journey I suspect they would say —— you know, I think Physical AI and robotics is 2% into its journey. We don’t even have the data to really make these systems as effective and capable as they’ll be in five years and ten years. So, I think the adoption of technology or the adoption of AI is just beginning and starting to accelerate. I think some of the things we’re going to see is more and more rise of collaborative robots, for example, using more vision and more information from the real world to drive that change.

 

Jimmy Carroll: [00:17:46] I could probably sit here and ask you questions about AI for the rest of the day, given your background and my overall curiosity with it. But I do want to cover a little bit of the robotics stuff, too. And you just said something that kind of tripped my brain. There’s there’s been some standard news about the definition of “collaborative robot” and what that actually is. But in general, like on the topic of robots working alongside people, what are some of the —— obviously AI —— but what are some of the trends and technologies that are reshaping the way that robots and humans can work together on the factory floor or the warehouse today?

 

Anthony Jules: [00:18:25] I think probably the biggest overall trend is the need for systems that are more adaptive and flexible. I think this is… This has been affecting manufacturing really over the last couple of decades, where you have more and more attempts at doing things that are high mix, medium- or low-volume. And what that means is you need to set up a process, run it for some amount of time and pull it down, as opposed to set up a process and have it run largely the same way for years at a time. And that need for flexibility is one of the drivers for collaborative robots that work with people and are more adaptive. So some of the things that drive that are using modern AI to have much better and much more flexible vision systems, to allow dynamic navigation in environments, using techniques like visual SLAM —— visual simultaneous localization and mapping —— so that robots can know where they are just by what they’re seeing in cameras, and then using those cameras to also understand the environment semantically; understanding that that’s a person, that’s a machine, that’s a forklift. And that context allows the robot to more appropriately move through the world or address people or other tools in the world appropriately. And it’s really the weaving together of those things that allow us to, WILL allow us to have systems that are very adaptable and dynamic in the environments that they’re in.

 

Jimmy Carroll: [00:20:31] Speaking of the environments they’re in and, going back to that thought of in a lot of ways, industrial automation is only just beginning… A lot of small- and medium-sized enterprises or businesses today aren’t really using industrial automation technologies like robots and AI —— or maybe they have some level of it —— but I think that area, that space, is ripe for growth. You guys refer to this idea of designing robots for real people. Do you see  your technology being an enabling technology for these companies that maybe don’t have on-site integrators or engineers that can support these systems to be able to maybe lower the barrier to adoption?

 

Anthony Jules: [00:21:20] Absolutely. Thanks, Jimmy. I’ll say, that’s where we’re going. And that’s a big part of our vision. We’re a startup and we have giant contracts with a couple of giant companies right now. So that’s really pushing us forward. But we’ve designed from the beginning, we’ve built for associates on the warehouse floor, on the shop floor, and not just for engineers. The vision here is this technology needs to become simple enough to integrate and use that it can serve that middle market, you know, especially small and medium enterprise. What we’ve seen, or from the research we’ve done, we believe that a slow adoption there is primarily because the value proposition is unclear and the ROI is unclear. There’s a lot of friction, and people are forced to be in a position where they think, “okay, I need to write a big check just to figure out if this thing might work for us.” And as anyone, including myself, who’s run a small or medium business, that’s not a decision you can make. So part of our vision is really reducing that friction, as robots should feel more like tools that you already know how to use so that they can look at it and say, “oh, I know how this can impact my business,” and they can make that first step. And that’s really what we’ve been designing to. So I mentioned, we have a big partnership with DHL Supply Chain. And we just recently made public that one of our deployments with them showed a 60% productivity increase in the picking operation that we worked on with them. That’s what the big guys are going to see. And as we make this simpler and as we make it more approachable, I actually think people, companies in mid-market, can see much, much bigger impacts than that because they’re actually starting at a less optimized place.

 

Jimmy Carroll: [00:23:39] Yeah. And, you know, respect to companies like DHL and others that allow you to go public with it, because oftentimes that ROI number is out there somewhere, but it’s under heavy NDA and you can’t find out what it is and no one knows. And it’s difficult for these smaller companies to learn from the bigger companies because they say, “yeah, we’ve got these contracts with X, Y, Z, but we can’t talk to you about them.” And that’s the challenge.

 

Anthony Jules: [00:24:03] It is. And I think we’re very appreciative of DHL letting us share that. And that’s the result of an almost two year relationship.

 

Jimmy Carroll: [00:24:14] This month, it’s almost June, at the Automate show —— and in the last couple of years at Automate —— I’ve seen AMR [autonomous mobile robots] just continue to become more popular, more widespread, larger, with larger demo areas at the shows. And,  you can see the booth size growing for a lot of these companies —— even companies that are not startups that have been around for a long time —— growing exponentially over the years. And I’m just curious, from your perspective, as someone who’s been in the industry a long time and, you know, works at a company that makes them, what do you see as some of the primary growth drivers for AMR? I know we’ve talked about AI and vision systems, but what about things like, you know, easier-to-use fleet management software or whatever else.

 

Anthony Jules: [00:25:02] Yeah. So I think you hit a couple of them on the head. We are betting on vision-based robots. So vision-based navigation. We use visual simultaneous localization and mapping. We’ve actually stepped away from the two dimensional LiDAR. You know, a 2D LiDAR just gives you information about one slice of the world. Robots that work indoors and at scale can’t afford —— and it’s not the right sensor —— to use a giant 3D LiDAR. So we’ve gone to only cameras, and we’re writing the exponential of those technologies becoming cheaper and more accessible, and the cost of an incredible amount of AI compute, you know, taking the input from those cameras has come down significantly. So that’s one driver. You mentioned another one. Fleet management software is getting better and better. And, many of the big players have pushed for interoperability, which is something we’ve built for from the beginning in a way that I think will actually serve, of course, the big players well, but will help shape the industry. So that’s certainly another big driver. But if we step back, kind of even one step back. I mentioned a little bit about this before. I think the biggest driver is flexibility. Companies can start small and scale. They can change the layout of a facility and the robots… You know, remapping a space can be done very, very quickly. So, for example, we’ve mapped a space that was a quarter of a million square feet in three hours. You just kind of flip the script on how. It just doesn’t take nearly as much time and effort to set up, install, and deploy these more modern systems.

 

Jimmy Carroll: [00:27:00] And like you mentioned before, like LiDAR, I mean the price came down a bit, but it’s still expensive. And, on the other end of things, by using  just cameras, technologies like 3D have really almost come of age in terms of their size and their cost and their capabilities. And I imagine that that has been a driving factor in terms of companies like yours being able to say, “we don’t necessarily need LiDAR, we can use these now.” Is that the case?

 

Anthony Jules: [00:27:33] Oh, absolutely. You know, my position is, the right set up of cameras and depth sensing gives you a much richer and a much more dependable understanding of what’s around the robot. And that’s, I believe, the biggest input into both safety and capability. You know, if you give a human the equivalent of a 2D LiDAR, you know, you put on a helmet with one really thin, thin line in it. Most people would feel pretty impaired in terms of their ability to move around and not bump into stuff. As a roboticist, I’m proud of what we’ve accomplished as an industry in getting the LiDAR to work as well as it does, but I think it’s pretty intuitively clear that if you have a system that literally understands what’s around it, just by being able to see all around itself and have better depth perception than a person —— these robots actually see better than us because it’s not, vaguely, that a thing’s further than something else. No. There’s an object that’s 1.2 meters away from me. It’s calibrated. I know where that thing is. And I think that type of rich understanding of the world is a direct input to make robots both more capable and safer.

 

Jimmy Carroll: [00:29:05] I always appreciate a good analogy, and that’s a great one. So if I use it in the future, I’ll give you credit. I promised to ask about it because I really, genuinely want to know your opinion: Humanoids, right? It’s a hot topic. There were some humanoids at Automate. A couple of them were moving, so that’s a good thing, right? Where do you come down on the debate of “why do we need humanoids versus wheeled robots?” And if you don’t necessarily want to answer that, just if you have thoughts on humanoids in general, I’d love to know.

 

Anthony Jules: [00:29:40] Oh, I’m happy to answer it. You know, I’m a roboticist at heart. I’m a kid who started programming at 11, wrote my first video game at 16, and knew I was going to study robotics and AI when I was a teenager. So, humanoids are part of the dream that all roboticists share. So, we’re already tipped in that direction. And, when I look at it, I think humanoids absolutely have a place, but I don’t think they’re the only modality that’s going to provide value to customers. So I sometimes will say this to help people kind of tease things apart, which is: Robots are going to walk, some with two legs, some with four. Robots are going to fly. You know, we see drones all over the place. Robots are going to sail and dive. They’re going to be in water moving around. And robots are going to roll. These are all the modalities. And when you think of it that way, you realize, “oh, these don’t actually do the same job.”

 

Anthony Jules: [00:30:52] They don’t actually do the same thing. A drone isn’t going to do what a quadruped does. And certainly not going to do what a submersible does. So when you start looking at it, wheeled robots are really about portage. They’re about moving stuff around. You know, there’s a reason that we invented the wheel. And then we invented the cart. And for 5,000 years, we’ve been putting stuff on platforms with wheels and dragging it because it’s so much more efficient than carrying a box around. We actually can’t move stuff long distances without rolling. So, I think the thing I want people to tease apart is, you’re going to have different form factors. We’re going to have different morphologies. They’re just going to be good at different things. And they’re all complementary when use appropriately. So, on the one hand, I’m of course a big supporter of humanoids. They’re part of the dream that all roboticists have. AND they’re not the solution for all problems. So we’re complementary to humanoids, not competitive with them. So, legs and wheels don’t necessarily do the same thing

 

Jimmy Carroll: [00:32:04] Yeah. Given your background  —— you just told me that you started programming at 11 and knew you were going to be in robots and AI as a teenager, plus, you know, your career —— do you have any advice for young students wanting to get into robotics or maybe robotics startups in terms of scaling and adapting to the ever evolving customer need?

 

Anthony Jules: [00:32:33] Yeah. So I think this is… As free advice —— it’s worth every penny. But, and I think this may come through in a lot of the things that I say, which is, I really think focus on building value. Focus on understanding some customer. Understanding some need. Understanding a set of pain points. And really figure out how you can solve a problem for someone in the real world. And as roboticists and people working in AI, we have to do that and we have to navigate the hype at the same time, because it’s good and bad that we’re now in a space that the whole world is paying attention to in a way that wasn’t true five years ago. And because of that, there’s a lot of hype, and that’s good and bad. You know, it’s good in that it’s a little bit easier to explain what we do and why it might be valuable, but it’s bad in that it sets unreasonable expectations for some of the things that we’re trying to do. And some of the things that can add a huge amount of value.

 

Jimmy Carroll: [00:33:41] Yeah, sort of like the idea of, “find a problem and then find a fix for that problem,” right? Instead of just trying to build a general purpose product and say, “you might need this.”

 

Anthony Jules: [00:33:51] Exactly. Yeah, make sure you have a need and then build for that need.

 

Jimmy Carroll: [00:33:59] Fair enough. Anthony, anything else we haven’t talked about that you feel would be important to mention regarding today’s industrial automation manufacturing climate?

 

Anthony Jules: [00:34:11] I don’t think anything new, but it just strikes me that we’re at such an incredible moment in history, both in terms of AI and what’s happening in terms of robotics and Physical AI, and we’re just at the beginning of this journey. So maybe I’ll go back to the first thing that we said, which is, as much as we feel like the change is never ending right now, that’s going to be the constant for the next couple of decades. We’re going to have more options for how we increase productivity, how we collaborate with robots and each other, how we build value. And the hard part is going to be selecting from the incredible abundance that’s available, not, “Is there a way to get X, Y, or Z done?” I find it’s such an amazing moment right now.

 

Jimmy Carroll: [00:35:10] Yeah. I mean, it’s indeed a very exciting time to be involved in this space, in this community. And I, for one, look forward to everything involved in it —— including following the trajectory of Robust AI and the Carter robot and everything exciting you guys got down the road. So, if anybody else has questions or if anybody wants to learn more about Robust AI, it’s Robust.AI. And if you’ve got questions for Anthony, I’d be happy to take those and pass them along. You can reach out to us at manufacturing-matters.com. Otherwise, thank you everybody for listening. And Anthony, thank you so much for taking the time. Really enjoyed this. I appreciate it.

 

Anthony Jules: [00:35:52] Thank you so much Jimmy. It was a pleasure to be on the show.