Welcome to Advanced Automation, a podcast by Calvary Robotics, where you'll find industry leaders and experts sharing their thoughts on the world of automation. Hello, everyone, and thanks for listening. I'm Josh Prevelli, your host for this installment of Advanced Automation. Today, we have with us Eric Danziger. Eric Danziger is a cofounder and CEO of Invisible AI. After serving in the US army for nearly a decade, Eric studied computer vision at the Robotics Institute of Carnegie Mellon University. Prior to founding Invisible AI, he worked as the director of software engineering at Luminar Technologies and led a team of over thirty engineers working on some of the hardest problems in autonomous driving. In twenty eighteen, he found Invisible AI to bring state of the art computer vision technology from self driving to industrial applications, blending his military experience and technical knowledge to help companies streamline operations. Thanks so much for being here. Absolutely. It's great to be here. Yeah. So I gave a quick overview of some of your background. It's a little maybe deeper dive coming straight from you would be wonderful. Absolutely. So, kind of as you mentioned, I, I I served in the army. And one of the things that that, I was kind of really thinking about when I when I transitioned, was just, you know, what is it ultimately that that I wanted to do? Kinda what was the the kind of area of the of the world that I wanted to work in. Right? And, I think robotics really just is incredibly cool. I think that's something that that everyone who works in in robotics and and manufacturing can can kind of understand. You get this this amazing kind of confluence of really hard technical problems and things that happen in the real world. You're not just stuck behind a computer. You get to actually see the impact of what's happening. And so that's that's kind of what led me to, to Carnegie Mellon. And I would say that that, you know, one of the things that you see again and again is that, there's so many problems to be solved. Right? So robotics and computer vision in a lot of ways is still in its infancy. We're still we're still kind of at the beginning here. And so that makes it honestly even more exciting for me. But it means that you need to be very, very clear headed, about problems that you can solve and about areas to work on. So, I think there was a period of time, not long after I got out of graduate school where self driving was a year away, or something something something like that. And I think it's, you know, again, it's it's really easy kind of when there's a big change in technology to kind of overestimate how much that's gonna impact all the problems that it can that it can touch. And so I think that there's we're still, you know, many, many years away from what anyone would call kind of real self driving. And I think that that at the same time, those technology changes were very real. Right? So being able to now do all the stuff that we can do with deep learning, has changed a lot. It hasn't changed everything. It hasn't changed some of the things maybe that people assume that it has. It's definitely not a cure all. And I think we we have a habit of always overestimating these kind of technologies, as human beings, but that that isn't to minimize what's possible now. Right? And that's the thing that I'm really excited about is that I think that we're definitely on the cusp of some really, really exciting stuff in terms of the the capabilities that we can deliver to industrial customers. That's very exciting, to be on, like you said, the cusp of what's next. You're solving it and creating it, you know, all in all at once. So do you can you give us then a nice maybe a boilerplate or your rendition of what Invisible AI is? The way that I that I like to kind of think about Invisible AI is is really trying to build a new layer of information in these manufacturing facilities. Right? So we have, very specific products that we're that we're starting with, and those, I think, have have huge impact. Our customers are are very excited to get those. But I think that the the purpose of Invisible AI, the goal is to really be a new stream of information, a new place to get operational information about what's happening inside of your facility. So our first products are around manual assembly. This has traditionally been an area with very little digital information. Anytime you wanna understand what's happening in manual assembly, you literally have to go walk and look. You have to see. You have to take a stopwatch and actually measure yourself. And this is actually part of the the kind of Toyota production system. Right? Again, so you can boot the actual kind of go and see. Right? That's a that's a core tenant. And that's just because the actual environment is is challenging. Right? You you have to understand why things are hard and things are hard not because, you know, this one station wasn't set up properly. That one station's inside of a facility. Right? And and anytime, you know, they run out of parts, it's not that they go two feet over. They might have to go to the other side of the factory to go get something. Right? And so understanding why things are happening, you really need to be in that context and you need to to understand it. And I think that, for us, giving that visibility is kind of the the goal. Right? Being able to see kind of what's happening. But we're seeing that already with, safety. Right? So there's there's things around robot safety, fork trucks. Right? The the vehicle is actually moving throughout these facilities. Right? There's a ton of places where we wanna have a little bit more visibility, understand what's happening, and make sure that we're kind of always working towards better safety, quality, and cost in manufacturing. Absolutely. What what is on the horizon then? What new innovations for Invisible AI? Yeah. So we've been, you know, I think we we basically made a decision pretty early on the company that we were gonna tackle, some of the harder problems that that no one else was really looking at. And and, that some of that is around, moving assembly lines. So these traditionally have been very hard from a computer vision perspective because you can imagine things happen at different periods of time while things are moving. Right? People are getting in and out of vehicles. There's a lot of complexity. It's not easy to to manage. So that's something that kind of required us to develop a whole bunch of innovations that we've been slowly kind of, rolling out to our customers. Everything that we do now is in in three d. Okay. We have multiple cameras that have overlapping fields of view. Everything's time synchronized. And so what we're really building is is already this living three-dimensional representation of this moving line, and then using computer vision to pull out the relevant information from that. So that's something that can extend again very easily to a robot cell, that we're extending to crosswalks and kind of intersections inside of facilities. And that, you know, has forced us basically to kind of innovate on how we do things, in-depth, how we synchronize across all these systems, and just really kind of deliver the a next generation suite of of, computer vision technologies inside of these customers. So I imagine pulling all that kind of data is difficult, but what other challenges then are you facing that with the Visible AI? So we we actually try and have everything happen on device. So, all the camera systems, all the the, AI devices that we have, I have all the processing power they need to compute everything locally. And so what we do is we kind of collate all that after we've, run the the various kind of data extraction algorithms to, a database on prem. So everything stays in their facility. It's a lot more secure. There's very little network bandwidth required. And that allows us to install thousands of cameras instead of just dozens, on on networks that may not have been upgraded for that kind of bandwidth. And so it's, you know, it's hard. I think the, my favorite saying is, from I think, like, an eighteenth century general, I said, amateurs talk tactics, experts talk logistics. Right? The hardest thing in actually deploying these things at scale is not necessarily the the individual technology pieces. It's kind of everyday things, like how much bandwidth do you have? Can you actually stream that much video across your facility? Right? That's the thing that tends to break some of these deployments. So, that's something that we've been thinking about kind of from the beginning. And I think, yeah. So that's, I think that's a big part of it. I think the computer vision part of it. Right? So we've been focused a lot on, methods of computer vision that don't require extensive retraining. So this is a big topic for us. One of the biggest problems that we see computer vision, especially deep learning companies have, is that they need six weeks or months to get up and running. They need to take your personal kind of, IP data and send that overseas sometimes to get labeled. Right? It's it's not necessarily very secure. It's very, very it takes a long time. And if any changes have happened in that period, right, if you've changed any of your process or changed part, changed whatever's happening, it starts all over again. Right? And so that that that chain can be very, very fragile. So we try and do everything, again, kind of assuming that these devices are configured by the customer, that everything's gonna run locally, and that this should work out of the box. That's the the kind of, deployments that we're that we like to do. So, as we develop these new use cases, we're trying to always keep that in mind and not make it something that's winds up being more consulting, right, where you have to actually then go and configure for the specific customer and label the data, deploy a new model. We wanna have models that are working across entire, our entire customer base. As far as your customer base, do you work with end users, more with system integrators like Kelvin Robotics, or both? What is your sort of, look at customer So I think we're we're we're still pretty early from a technology perspective as a company. Right? So we tend to work with our customers directly, but manufacturing is only possible. To actually penetrate the manufacturing market, it's really only possible if you can work through systems integrators long term. Right? So, I think that's something that we're absolutely going to do. But we need to make sure that we have things kind of in a place in a place that we can do that hand off. And systems integrators are, in a lot of ways, as demanding as as manufacturing customers for the same reason. Right? They want it to work. They don't wanna have you know, they they wanna have make sure that everything is is kind of easy to install, that it does what it says on the box. And so kind of as you as you basically work through the the kinks of just being a young company and deploying some of this stuff for the first time and some new facilities and new types and new types of manufacturing, it's best to kind of keep our hands on it as much as possible. But, again, the the kind of thesis for us, the the belief is that if we can make this something that anyone can set up, that you can set up in minutes, and that doesn't require our intervention to set it up, that's something that a systems integrator would would definitely, be be kind of able to do. And I think it fits well with the systems integrator kind of mentality. Right? Which is, you know, you have a a a set of specific customers. You know, their needs spiked really well. You're you can do a lot of the customization and a lot of the the actual work to kind of make things specific for them. Whereas the a lot of the kind of core technology companies are building a more kind of platform product that then gets customized, at that level. So, yeah, I think that's I think that'll that'll absolutely be the path going forward, but we're not there yet. What what is the recommendation you have for your customers looking to implement Invisible AI? That's a a great question. I think, you know, the big thing for us that we see is is, again, you know, starting with some of these first products, there's always, a big, big fire that's going on. Right? And I think that ultimately, inside of a lot of these assembly lines, you'll have just a huge host of problems. It's, you know, supplier's part's bad. You know, you can't you can't, find qualified people. You know, the people don't have enough training. You know, there's a lot of it that that honestly there's just not enough process engineering. Right? I think that a lot of what we're trying to do is basically multiply the effort of these kind of process and quality engineers and make sure that we can help them augment the things that they're doing. Right? So all the data collection, all the kind of manually going there and standing online for hours on end, Those are the things that we can help with. And so I think anyone who is experiencing that problem, which to be honest, is almost everyone in mainstream, can, can, can kind of benefit from, working with us. So, I will say the, you know, we're we're, very much, looking at at working with kind of more mid market companies. We've been working with kind of OEMs. The the one big public one obviously being Toyota. But I think working with kind of the the the tiers and the supplier base, and manufacturers outside of kind of automotive, aerospace, and appliances is something that we're actively interested in. So, that's you know, anyone is interested can always reach out to us, reach out to me, and, and we'll get them set up. Absolutely. Absolutely. So as far as connecting with you, what's the best best way? The best way right now is just, you can you can, go on our website. There's a there's a form submission there or just email me directly, eric at invisible dot a I. And, I'll make sure that that you get connected with our kind of sales and deployment team, and, we can get things up and running immediately. So do you visit you kinda start off and then visit the site, or kind of first steps? Is that Yeah. So the typically, kind of figuring out with the customer what what they kind of what what the reason is that they came to us in the first place. Right? Doing site visit, trying to understand their lines, trying to understand their manufacturing facilities. And again, I think what we see is that typically there's, you know, a specific problem that's stripping them to to look for us. So something that is very just top of mind right now, this line is underperforming or this thing is not being made properly. We can't figure out why. But but ultimately, this is something that that we think has visibility or that that has value across the entire plan. Right? This visibility. And this is something that, you know, one of the key tenets, another thing about being kind of on a device and not in the cloud, is that we are able to offer this for for a much better price, than anyone who's kind of cloud based. And that's the the goal for us, right, is to sit to offer something that's very cost effective that that is so cost effective that you can put it in the entire plant and get value from from this visibility. Right? Because the biggest thing that that I've seen is that whatever the problem was that they came up to us with, if, you know, the whole process of of negotiating and and getting a PO takes a month, they could be having a completely different fire and a completely different part of the plant by the time we actually do get installed. Right? So you don't know where these things are gonna happen. You really wanna see the whole place. And if it's, again, if it's cost effective and it delivers kind of recurring benefits around visibility, root cause analysis, being able to actually see what went wrong, when it went wrong, employee training, all that stuff. It's it's it's absolutely worth it. Wonderful. Well, anything else that you feel like you would like to add before we close things out? I think, you know, one of the the key things for us is is always kind of understanding, you know, what what other players in the market are are interested in as well. Right? So I I think this is you touched on the systems integrators. This is something that we just haven't, honestly had time to to, really dive into. But understanding more kind of from the systems integrators, from robotic suppliers, from other companies kind of deploying automation inside of manufacturing, where they see gaps, what their needs are, and kind of what they could do if they could see everything throughout the facility. Right? So already we've heard from some companies that, you know, remotely, monitoring some of these robotic installations could be super valuable. Just because I'll give you give you an example. Right? If, you set up a, a robotic cell and there's no one qualified in the plant to actually maintain that system. Right? And something changes or someone, you know, puts it out of alignment, then you someone needs to come on-site and actually fix that. Right? Right. And someone, you know, especially if that winds up being kind of a sensitive area or medical area or any any kind of reason that you wouldn't want someone to kind of be in there. It's a clean room of some kind. That can be, you know, orders of magnitude more expensive. So I think there's there's a lot of these use cases that we just, you know, we hear about after explaining kind of what we do. And it's always great to connect with our technology companies. So always interested to to kinda learn more and understand kind of what what's out there. Wonderful. Well, I appreciate your time. Thank you for for talking with us today. Absolutely. Yeah. It was great to talk to you and, thanks a lot. Make sure to subscribe to Advanced Automation wherever you get your podcast. And head over to calvary robotics dot com to listen to other episodes, watch our series, Calvary TV, and lots of other great content about Calvary Robotics. Have a great day.
