“What I’m most excited about right now is what’s new in machine vision. … This combination of new sensors, advanced processing, and sophisticated algorithms will unlock new spaces for vision.” 

While robots and AI may steal some of the spotlight from other technologies in industrial automation, machine vision technologies are being used more than ever before, providing eyes for the machines and actionable data for businesses across all industries. In this special episode of the Manufacturing Matters podcast, TECH B2B Marketing’s Jimmy Carroll is joined by Steve Kinney, director of training, compliance, and technical solutions, Smart Vision Lights; Jason Covar, general manager, The Imaging Source; and Jeffrey Adolf, vision/AI Specialist, 3M, to discuss the latest developments and what to expect in the near future when it comes to all things machine vision.

The panel covers the intersection of machine vision and AI, some of the latest opportunities that machine vision has enabled, the most interesting vision developments in the last few years, and a shift in embedded vision system development. Additional topics include vision standards updates, advances in processing technology including the cloud, and a look toward the future.

.

.  

Jimmy Carroll: [00:00:03] Hi, everybody. My name is Jimmy Carroll. I’m the vice president of operations at TECH B2B Marketing, and welcome to this special episode of the Manufacturing Matters podcast, where we discuss the trends and technologies shaping the manufacturing industry today. I’ve got a great panel of folks today to discuss machine vision trends. I’ll go down the line here. Steve Kinney from Smart Vision Lights, Jason Covar from The Imaging Source, and Jeff Adolf from 3M. Guys, thank you so much for taking the time. It’s a busy day here at the Business Forum, but I appreciate it. 

Jimmy Carroll: [00:01:15] We’ll jump into questions in a minute, but I’ll start by asking you guys each to just kind of introduce yourselves and what you do at your company. 

Steve Kinney: [00:01:25] Okay. I’m Steve Kinney, director of training and technical solutions for Smart Vision Lights. We’re a leading supplier of LED lighting for machine vision. And as my title implies, I do inside training, outside customer training, and other videos and promotional material for us, and advanced technical solutions. And we have custom designs or other advanced interfaces or applications. 

Jason Covar: [00:01:51] Jason Covar, general manager for The Imaging Source, camera manufacturer [of] machine vision cameras. I oversee the US operations, which also includes more systems and software as the market has evolved, which I think we’ll talk more about today. 

Jeff Adolf: [00:02:09] Jeff Adolf. I work at 3M. I’m a division technology lead for our internal manufacturing technology team. So we service our own internal customers only. We work with manufacturing teams around the globe to help solve their vision problems, improve processes, improve quality. 

Jimmy Carroll: [00:02:26] I’m looking forward to diving in here a little bit. But, just a kind of fun opening question, what are you most excited about in machine vision today? Jeff, if you want to start on that end. 

Jeff Adolf: [00:02:35] Yeah, I think it’s going to be probably the expected answer is kind of … The intersection between AI and vision seems to me to not just enable new opportunities, things that we obviously couldn’t have done, you know, five, 10 years ago. But I also think there’s a huge opportunity to drive vision into places that it maybe wasn’t before. I know we’ll talk probably about some embedded solutions too, that are doing some of the similar things on another end of the spectrum. But I really like the idea of AI being able to enable more rapidly deployable solutions, maybe solutions that are maintained at a lower level. And then of course, your very high-end solutions that can only be done with that. So that’s my answer. 

Jason Covar: [00:03:24] Yeah, I think it’s an exciting time in the space. I think there’s a few changes in the market. There’s a lot of talk about AI, but hardware-wise I think there’s a transformation. To what Jeff said, there’re other opportunities with markets that weren’t traditionally our customer base a few years ago. So the customer base is opening up, but I think it’s the responsibility of the companies to match that, right? Meet the customer where they’re at. So that’s an exciting time for us to be able to put all that together. 

Steve Kinney: [00:03:52] And I would say things along similar lines. We could talk about linear axes of various technologies, but as it all comes together, I think two things are happening. One, the capabilities and the ability to solve problems for a wider variety of customers, combined with the market awareness. And now is the time for automation and the market is coming to us. I think all of us would agree, we’ve had technical capabilities beyond what the market’s demanded for quite a while, and the awareness and maybe the buzz around AI actually makes customers. We get more inquiries, we’re doing more and more things. And those areas include wider areas than ever before —— food, agricultural applications, all kinds of places where we can apply vision and make a difference in their applications and end results. 

Jimmy Carroll: [00:04:41] You know, in the larger space, in the industrial automation space, there have been some notable developments as of late, whether they’re acquisitions, mergers, or large investments in production and manufacturing facilities in the US. And obviously, Jeff, AI is among them. And feel free to continue to bring it up because, as as you know, there’s reason for it. Right? It’s very exciting. But back to the question: What are some of the most notable developments to you in automation today? 

Jeff Adolf: [00:05:14] Yeah. So I’ll focus on vision, obviously, because I think as a broader category, there are some really cool things that are happening in automation, you know, from the robotics side. But I think from vision, I’m excited about —— a little bit to your point, Jason —— I’m really excited about the opportunity to use vision as a sensor in so many more applications to deliver data that’s relevant and then usable through AI. So it’s not just, hey, you put AI together with the image-processing backbone of your vision system and you get something that’s new, right? That’s cool, and that’s what we’ve been doing. But I also think there’s a lot of opportunity now. Hey, I can generate unique data from vision sensors that just didn’t exist before. And we can take that data and incorporate it with our process data, incorporate with our SAP data, incorporate it with our safety data, right? And those kinds of things to me are opening up avenues that, as technology people and as automation people, are incredibly exciting. 

Jason Covar: [00:06:23] Yeah, I think sometimes we overlook, you know … We look at what’s in front of us and AI is talked about a lot, but sometimes if we zoom out a little bit and we look at the fundamentals that are right in front of us, there’s a lot there. And I think vision falls into that group. You know, we work with groups that have supply chain optimization through AI where it really is an asset. But now they’re looking to build digital and vision groups to do the verification. So supply chain is optimized all the way to a certain point. But that visual representation is key. And so I think we’re seeing that as a big part. And for the adoption, when you talk about adoption, I think, like, Steve’s role is critical for our companies. The pace for adoption of technology comes to the training and educating. And I would say that Steve is in the perfect position to do that. And in a lot of companies that does inhibit the, you know, the growth of that. So, yeah, I think that’s a just point. I think vision is an area of growth as well. 

Steve Kinney: [00:07:24] Yeah. And I think it’s an exciting time to be in vision. My history is in cameras and I’m working in lighting now. But I’ve been in this industry doing vision since 1998. And I think, again, if we start talking about technologies, the sensors, the resolution of the sensors, the multispectral aspects, the spectral ranges of some of these sensors, other advancements in multispectral, non-visible imaging in general and the widespread use of that now, or the increasing use of that, whereas it used to be a little esoteric … I think all these things are really interesting. I think there’s technology behind each one of these that are making them each enabled more and more on their independent axis. And the things we can do with vision —— it just helps widen the breadth of solutions we have. And thanks, Jason, for that tag there about education. I think it IS key. I think through the A3 we offer CVP, and other courses through private companies like SVL. We do trainings twice a year on site. You can sign up for free, and it’s important to make sure you understand a little bit about all this too, so that, you know, as you’re coming to us with a problem, you understand in some context around there about what we’re going to solve for you and maybe understand some things we’re going to talk about. 

Jason Covar: [00:08:48] Yeah. And I would say I think … AI in general, right? You could take the fireball approach of putting it everywhere in your company. I do think the hardware companies are the gateway to a lot of this, to deploy it in a way that works and scales. It does rely on the hardware. So I think for the hardware company sides, I think we’ll see benefit from this. 

Jimmy Carroll: [00:09:10] Yeah. And it’s important to kind of give vision it’s flowers, right? Automation and the automation space, at a lot of these trade shows, robotics and of course AI and especially humanoids kind of steal some of the thunder. But machine vision is a critically important enabling technology. And you look at the ways that all these technologies work together to solve new challenges for customers. And without vision, you can still solve certain things, like, robots can do pre-programmed paths. Think about AGVs versus AMRs with the incorporation of sensors into the AMR. It’s a lot more flexible than an AGV is. But there’re a lot of other examples of that too. What are some examples of how vision is kind of opening new doors in automation in general? 

Jeff Adolf: [00:10:00] It’s a good question because I think there are so many areas. One that I mentioned and I’ll bring up again is, I think the ability to access not just data, but information from things on a process line is enabled by cheap embedded vision. So if you have, for example … Just a simple example: you have a new piece of capital equipment for, whatever, making a widget, coming onto your floor. It’s now almost de facto that that’s coming with three or four cameras that are looking at the process, saving to an NVR device, and maybe there’s some intelligent video saving for when there’s an upset or something. This is just such a simple thing. And it’s not something that we couldn’t have done 20 years ago. It’s just something that now is de facto, right? Vision is becoming kind of a, hey, I need information. How do I get it? I’m going to put a camera on there. I also think I’m personally really excited in the area of, more of the broader area of factory management as opposed to maybe specifically automation. I’m super excited for what AI-enabled vision can do. Simple example would be, human safety, right? This is not … We’re not quite in the automation world, but we’re still on a factory floor. But you can imagine, and we have at 3M, implemented solutions that help maybe real-time alarming for no-go zones, right? Or shut down lines if you’re in  an unsafe area or, even simpler, we have a lot of systems that are simply understanding what are the safety risks in certain areas of the plant. What are the ergonomic risks that we are continually seeing with our installed CCTV network that, again, most plants have. So this is using existing camera systems and leveraging the new AI technology to make them more interesting. But also just the proliferation, little cameras are everywhere, and I think we’re going to continue to see that. 

Jimmy Carroll: [00:12:02] And when you say that, you’re talking about, like, Agentic AI? 

Jeff Adolf: [00:12:05] So, it can be Agentic AI for sure. I think that big companies move slowly, right? So 3M is moving slowly. But even if you have, let’s say, dashboarding of true information, not just data, but hey, these are the insights that I have gleaned from these 17 cameras that I have looking at this process. Right? That’s a dashboard where a process engineer can look at that and understand what’s going on in this line, or a safety manager can understand about the risks that are happening in his plant. We’re not quite, at least at 3M, to the Agentic side, where you can take an automated action, but just getting that information —— information, not just data —— in front of an engineer or a technician, makes the time to problem solving so much faster. 

Steve Kinney: [00:12:56] I think that’s important. And as I listen to that, we talk about vision going in more places than ever? And I think I agree with all this. You know, I’ve worked personally on applications where we’ve done people-counting and stores and stuff, right?, to know where they go. Study the data. And these are more accepted now. But I think the point in there is there’s a whole range of these things now where we’re used to having these simpler cameras everywhere. And then you can start gleaning data like this and tying it in all the way to the high end. We have 50 and 100 megapixel cameras and we’re able to do stuff down at micron level, and the inspection world and stuff —— and everything in between. I think it’s a tremendous time for vision in general and just the explosion in these areas. I think AI just enables the possibility and furthers that explosion. 

Jeff Adolf: [00:13:51] Yeah. And if I could just quickly add on to that, I think the onus is on us. Right? As in “us,” I guess, *here*, to not allow us to be stuck in one of those domains. I think it’s too easy to say machine vision means this. It means a 50 megapixel camera looking at a semiconductor wafer. Right? That’s not it. There’s a huge, broad range now, and I’m super happy you brought that up. 

Steve Kinney: [00:14:15] I think that’s a really good point though. When I was product manager for JAI, you know we started hitting that point and even PULNIX a little before JAI bought them. But we started hitting those points in the early 2000s, where you started seeing things. You started seeing the simple drones and smaller cameras and what I called cell phone cameras everywhere. At first, as a camera guy selling, you know, we’re selling Kodak imagers on bigger pixels and dynamic range over Sony and other things. You tend to focus on that first. I kind of dismiss these as, oh, well, if it’s that far down the market, it’s not ours. But the truth is we have to own all that as vision and that if you’re doing that, you’re leaving stuff on the table. And I’ve learned to embrace all this too. But I think it’s important for the market out there to look and make those adjustments as well. 

Jason Covar: [00:15:04] And I think you hear the computer vision, right? You hear this terminology —— and I think we talked about this last year and you know the definition of that. But I think it feeds into what they’re saying. You know, when you think about the AI, the output is dependent on the input. And I think Jeff brings some good points as well as Steve. So, vision can be a critical input to that. Right? So you watch a prosthesis with cameras, then you feed the AI to have a better output. So I think it’s complementary in that way. But others may see it in the opposite. The AI will do all the work. But the input is very, very important for successful output, right? And vision, as you said, if you put something on the floor, you put it into the process and you actually watch it. That’s very, very valuable for these systems to work. 

Jimmy Carroll: [00:15:46] So yeah, I mean, if you think of AI as programming with data, if that data is poor, then you know, what do you have? 

Steve Kinney: [00:15:52] I think the other side to that, and to build on that a little … I’m also vice chair of the Vision & Imaging TSB (Technology Strategy Board) for the A3, and within the imaging TSB —— Jeff and Jason are on here too —— as we talk about these things on the Technology Strategy Board and help the A3 develop their vision for this we’ve asked the question, “what does vision look like 10 years from now?” Because we’re used to, oh, it used to be a CCD camera. Now it’s kind of CMOS and now it’s kind of visual, but, oh, we’ll go outside into the non-visible. But as a board we’re looking at that, and vision is going to go far beyond that. We had someone exhibiting CT scanning for example, at the last Automate, which I found really interesting and an alternative to x-ray. And you have the whole terahertz scanners in the airports, which haven’t really permeated themselves down to the automation world, but I think it’s coming —— or that there’s potential in these areas. And vision, at the end of the day, is the input to all this. But vision may not be traditional human-based vision like we think of just a camera doing it. 

Jeff Adolf: [00:17:09] That’s a great point. 

Jason Covar: [00:17:10] Yeah. 

Jimmy Carroll: [00:17:11] We get the chance to talk about this a lot. But you think about hyperspectral imaging kind of migrating beyond the lab, or 3D imaging kind of coming of age or, even more recently, developments in sensors have kind of made sphere sensing much more approachable for factory floor beyond just R&D. What are some other technologies, whether it’s from the vendor community or whatever, that are kind of opening new doors for different applications, different, ways to solve applications. And then specifically, one thing that comes to mind is processing, right? There’s been a lot of really interesting developments in FPGAs and SoCs, and GPUs are no longer this big, you know, Jetsons and all these other things. So, Jason, I know we’re going to talk embedded vision too, but … Let me stop there! 

Steve Kinney: [00:18:01] Well, I think the magic in that is it’s easier than ever. The processors are more powerful, they’re smaller. And in some cases you don’t need much of a processor. I’ve seen people deploy Raspberry PiS and Arduinos, for example. Raspberry Pi in particular can grab an image if it’s not high bandwidth. As you exceed that, then you can get into the embedded processors. And then as you go to AI, we, of course, then start talking the new Jetsons and all the things around that. I think the important thing for the audience to understand is there’s more processing power in incremental steps available at price steps that are proportional on that, and that you don’t overdo these applications. It’s important to look at this, get a definition about what I’m trying to do, deploy a lean amount of resources towards it. But I often ask people … There’s nothing wrong with the smart cameras and building that in, but in a smart camera, you get a certain selection of images and stuff there. There’s a certain framework that’s baked into that. And I often ask, “what’s the difference between a smart camera and having a cigarette-size processor sitting under the table?” And then I can use any camera, including much smaller cameras where there’s a weight or a size issue, as a tether to that. And it’s basically a remote head smart camera at that point. So I think a lot of people get into this and say, “oh, I need this” or, you know, “Cognex told me this smart camera…”, which is absolutely a fine answer, but make sure you’re weighing that out against the benefits of that architecture versus maybe making a discrete system. And where we make it all work together today like it’s a unit. 

Jason Covar: [00:19:44] Yeah, absolutely. I think the smart device, single piece of hardware, I think there is some decoupling from that in the market for sure. When you think about scalability, cost always comes into it. So there’s one attribute of embedded that helps.  I think when you look at the software side, data scientists, others working with all of the data, love the GPU or love the capability; I think the market, there’s a disconnect between what I want and what I believe we need and the deployment. And again, I want to go back to Steve and training and educating, I think as a group and a community, we really have to educate the customer side for this to scale. When we talk about embedded vision. From a manufacturer perspective, it’s also a system-level sell. It’s not a component. Whereas most of the machine vision world is used to selling at a component level, you know, putting things together. So this is a shift as well. And then the other thing I would like to say is, synthetic data? When you talk about new systems and how it’s working, I think this is a big improvement and does help a lot. So I think this is an area you’ll see more and more. You hear it more. But I think it is a big step forward that helps a lot of things that may have been successful in the past with having this data and being able to work with it in real time. So that’s what I see. Embedded, with synthetic data is more and more commonly required and asked for today, I should say. 

Jeff Adolf: [00:21:13] I’m glad you brought up synthetic data. That was the first thing that came to mind when you said that, and I think there’s some maybe horizon two, three technologies that are not quite there. But, yeah, synthetic data is something that you see extremely commonly. And I think it can certainly be overdone as well. I think there’s a lot of applications where if you were simply to deploy a camera and get real data, it’s just simply better than synthetic data. But I think having that alternative, maybe for huge skew mixes and things like that, we see that happening at 3M. The other, maybe more of a question for you guys than a comment, right? Because I’ve been pondering this same question about where the processing happens in a chain? So, typically, you can have your smart cameras, right? Very, very simple, it processes on the camera. You have a huge amount of power to do that. And then you get a result out of your camera. You can do, I don’t know, ten years ago everybody was saying, “well, everything forever and ever will be in the cloud. It will always be processed in the cloud forever.” And that hasn’t panned out right. But, there’s a place for cloud processing. I’ve been seeing as, again, this trend that I’ve mentioned, where you start getting more of sensor-style cameras. Things to just monitor. You know, where you may have had a sensor before, you have a camera now, right? I’ve been pondering … Do you guys see a shift now where individual cameras will be making, you know —— you have 30 cameras in a small area, all making individual small decisions and then feeding that up. Or do you see it percolating all of that data to a larger processor, maybe to the cloud, maybe on edge? I’d be interested to hear your take there. 

Jason Covar: [00:22:59] Yeah, absolutely. And I think you’re right. We’ve seen an evolution. Everything was cloud, then was on prem. There was, you know, it was kind of driven by AWS. 

Jeff Adolf: [00:23:09] You missed the fog. That was the between. Between is the fog.  

Jason Covar: [00:23:13] Yeah! What I hear … And in part, I mean, you know it as well, probably, as any of us. You know, with a company like 3M, if you want to scale something, you know, the IT layer does come into it. I’m sure this is a big piece of slowing down, stopping completely, you know, taking things off course. But I think on prem for the data, but also for quick decision making. 

Jeff Adolf: [00:23:35] Yeah. 

Jason Covar: [00:23:36] You would inference on this device and make a quick decision in a binary way. We don’t need to send it up to the cloud. And we’re not doing all this calculation, where we really can train and we can inference on this small device —— literally closest to the problem —— and make that binary decision.  

Jeff Adolf: [00:23:52] Yeah 

Jason Covar: [00:23:52] And this seems to be the trend. I think the cloud piece, from what I hear in the market has moved on. And I think also companies’ security, where your data is sitting, is also another one that you hear they say, well, we’re not so comfortable, you know, in a POC or something where you’re pushing a lot of data. But enterprise level, you don’t want all of this data going up to the cloud. And so I think the on prem is certainly the way. But to your point, if you are to inference and make a binary decision where there’s plenty of this, you know, opportunity within a factory, I think that’s part of the attribute as well as embedded. You get closer and you’re able to do that with the compute. 

Jeff Adolf: [00:24:26] So, I agree. Completely. And that was a leading question. But also, let me play devil’s advocate real quick. And I want to hear your idea too. With AI now being able to glean more information than it ever has before from data, there seems something almost heretical about getting rid of all of this beautiful, rich data before it hits a real processing unit. You know? Like, is there … Maybe it’s just application dependent, maybe it’s whatever. I couldn’t hear you any more about it. And cybersecurity at a company like 3M, it’s a problem, right? But  I always kind of have this back and forth, like, I’m getting rid of so much data and look how many more things there are to produce to use this data with. But at the same time I don’t need that data. I just want an answer and I want it to be done here. It’s self-contained. 

Steve Kinney: [00:25:17] I think there’s an interesting parallel to that. I think some of that, it’s very much a futuristic question and a little bit to be seen, some of it by application. Some of it will be by which architecture wins. And what I mean by that, if you think about drones, there’s a whole thing about, well … I have a drone and I can do it. And you have farming and hyperspectral drones and stuff now, but there’s also a whole contingent of drones where I have a hive mind. Right? And we all kind of share that, and they can move and act as one thinking. But I believe that’s trickling down to our level, too. So when you ask that kind of question that way, you made me think about that a little. And depending on the application, other things, you may very well have smart cameras that have a little bit of smarts, but kind of a hive-mind communication. And they all see a little something, flip a little switch, but make an obvious… 

Jeff Adolf: [00:26:12] We’re back to IoT. We’re back to IoT. It’s like this is the big circle of life, isn’t it? 

Steve Kinney: [00:26:17] I think the difference is … I do agree. And a little bit of “Stranger Things” as I talk about the hive mind there too, for those fans. But, I think the point in that is, where’s the data going though? Because in the hive mind, yeah, there’s a little bit of IT and IoT going on there, but it’s all just right here before there’s a decision made and then it goes upstream. And in a lot of cases that’s important because you can overflow everything going upstream if you’re not doing some of that. 

Jason Covar: [00:26:47] I think you’re right. So I mean, the asset is the data and the image. And as a vision person, we would always see that you would implement a vision process and you say, oh, I could do this, that, that, four more steps out of that data. I think what you see is, you see the decision making, but maybe the data doesn’t lose. It’s just not in real time. 

Jeff Adolf: [00:27:03] Yeah, sure. 

Jason Covar: [00:27:04] So it’s kind of more of a, you know, canary update kind of style where it’s not there, but overnight. 

Jeff Adolf: [00:27:10] Yeah. 

Jason Covar: [00:27:10] When you’re downshift, you still keep the metadata. But, you’re right. Yeah. I think they’re doing the targeted decision making. But that is an asset and it should be still. 

Jeff Adolf: [00:27:19] I’m trademarking my fog. This is between the cloud and the edge is the fog. This is where fog fits. 

Jimmy Carroll: [00:27:26] When it comes to those applications, though, that … We talked a little bit about embedded vision. I guess this is kind of a question for everybody. When you need —— like, I want to say real-time but near real-millisecond latency, for applications, what are some of those applications that you’re seeing out there these days that require those? And what are some of the solutions, whether it’s FPGAs or, you know, SoCs, GPUs, like what do some of those systems look like? 

Jason Covar: [00:27:56] Yes, I think if you just step back and look at the architecture of embedded, like, this was one of the attributes that made it become more popular. The latency. Right? So when you started in vehicles and you had these ADAS systems, you had these systems where you need really, really low latency. So the idea was: grab an image, put it back, and process it on the board. It’s kind of spun out of that, I think. You mentioned drones, for example, right? Something that is dependent on the flight of something has to have a low latency. When you think about that, and that’s one that comes at the top of mind, because if you’re controlling some kind of vehicle or aircraft and you have to make a decision off vision, this has to be almost a real time. 

Jimmy Carroll: [00:28:37] Especially in defense, right? 

Jason Covar: [00:28:38] In defense, right. So a drone —— if you’re making a decision on where that’s going. So that would be one that’s also popular. You know there’s a lot of conversation around defense today, here and in Europe. That would be one that’s comes to my mind. 

Steve Kinney: [00:28:53] I’d echo that too. I have a significant amount of experience with cameras and defense, and some of this is in the architecture again of these solutions. This is why you pick certain interfaces or architectures as well. I can remember in the high-end, I mean, 15 years ago, I had a group putting cameras on the axles of Humvees and some of these other things and using them like accelerometers, because these guys are often piloted by a guy in the back, or even if there’s a driver, there’s a guy in the back that’s watching all around the vehicle and making strategic choices about what he’s seeing 360 around the vehicle. The number one thing is when the vehicle moves, that guy gets motion sickness. So they actually put cameras on the axles and watch the vibration of that axle to stabilize the image and stop the motion. But the point in that is camera link is the only real-time spec that’s out there, and it’s still out there, even though we … It’s a long time … I see Jeff making a face. 

Jeff Adolf: [00:29:55] I have all of my installed infrastructures camera linked. 

Steve Kinney: [00:29:58] Yeah. 

Jeff Adolf: [00:29:58] I’m not hating that. 

Steve Kinney: [00:30:00] It’s funny, people don’t talk about camera link, and you got to know I love USB vision because I’m a guy that used to have to drag around a frame grabber, a PC, and just show the camera. They have their place and there’s nothing wrong with the USB and Ethernet. They have high bandwidth and we get the job done. But by the nature of the packets and the nature of their communication, it takes time to put those together. When they were doing accelerometer acceleration on the axles, they’re reading … Camera link is real time. Here’s my clock cycle, here’s my AD bits, depending on the configuration. And, absolutely, it’s just a modified parallel LVDS. But my point in that is, in the time you’re putting a packet together for Ethernet, they’d already read the first 10 lines of the image and were computing the motion vector long before they even got the whole image in there. So I think a lot of this depends on the application. And again, we have high-end capabilities. But that kind of capability is not for everything either. And there’s vision everywhere with lower-end capabilities. And you need to think about the architecture and not overdo the solutions or underdo the solutions. 

Jeff Adolf: [00:31:09] Yeah, I think my answer is maybe a bit anticlimactic. Honestly, the vast majority of the applications that I see are not latency constrained. They’re throughput constrained. That’s so industry dependent, right? I understand 3M is not maybe an automaker or defense, obviously, but for us, it makes a lot more sense to have that balance of, hey, we know what throughput we need, right? We have film that’s being made, high-speed film that’s being made. And the question then is, what’s the architecture that makes the most sense to match the throughput? And so some of that is, hey, now we can do parallel GPUs in a single server. And now our costs are way down. We don’t have to have a bunch of servers. Maybe we simply just have an embedded GPU, little NPU, whatever it’s called, right on board and hey, it processes it quicker. So, for me, I don’t see a ton of very, very high-, or very low-latency applications. But throughput is obviously something we talk about every day. 

Steve Kinney: [00:32:06] And it’s interesting you say that, because throughput and latency are two different things. The application may need one or the other or both. And there’s certainly a relationship with latency and throughput. But they go … And I think it depends on where your experience is though, too, because we talk about food and other things. And I often tell the story about sorting rice or beans. Nowadays that is done on what we call a waterfall. They spread them out on a vibratory conveyor, get them one layer deep, run them off a waterfall, and if there’s a rock or a piece of foreign matter as it’s falling, it goes by the camera. They detect that —— and often in the non-visible range —— hey, this is a rock, or at least some kind of matter that’s not the product. And they actually do what they call “puffing out of there” and puff it. And when it goes this way, it hits a divider and it falls in the reject bin. Those things are incredibly real time. And you have to make these decisions with very low latency. And there’s many others, you know, in inspection and other things, where we have to take action on things that are going by 20 a second sometimes. So there are applications that need it. 

Jeff Adolf: [00:33:15] 100%. 

Steve Kinney: [00:33:15] And there’s other ones that don’t. And even though they may be … The films and stuff are a good example, it’s a process. You know what’s running out, the throughput counts, but not necessarily latency, because you’ve got 20 feet to make a decision, right? 

Jimmy Carroll: [00:33:28] I’m glad you brought up vision standards because we’re kind of on the cusp of GigE 3.0. It’s coming. Well, it was coming a couple months ago, right? And then it’s coming in a couple of months. But it’s coming soon, though. 

Jeff Adolf: [00:33:39] April. 

Jimmy Carroll: [00:33:39] April. Okay. But what do we have to be excited about there? There’s Rocky V2, and then there’s GigE vision approaches with GPU direct, bypassing the CPU for more real-time latency. Like, what are you guys excited about there? 

Jason Covar: [00:33:57] Yeah. So Jeff and I are on the Vision Standards Committee. As you mentioned, GigE 3.0 is there. You know, I think the responsibility of our association in one area we’re in discussion to lead with A3 is something around the MIPI and the the embedded space, right? So there are some companies out there that are looking to do this on a hardware level. But I think as an association, the responsibilities, part of that is standardization and A3s benefit a lot from these standards. So I think we’re trying to put something there. It’s more of a fragmented process now with all the different components and embedded. But that’s an area I think we’re going to look to the future and try to lead as an association. 

Jeff Adolf: [00:34:40] Yeah. I think a couple things to be excited about from an end user, just bandwidth is going to be the thing that is going to be the most impactful by far. I think something that maybe is a little bit under the radar but should be recognized, just with all the European cybersecurity work going on, that impacts these standards. Right? That’s not just a hardware thing that ripples through these standards. So I think there’s a lot of forward-looking work being done to try to get ahead of that, too, so that we aren’t scrambling when that comes about. I also think it’s interesting … Small optical interfaces, I think, also is another thing that —— and we aren’t there yet —— but I’m personally like … We talked about camera link. Camera link is about as old as I am. You know, it’s an ancient. It’s an ancient system. 

Steve Kinney: [00:35:30] 1999 when we did the first draft, I mean … So 20 years. 

Jeff Adolf: [00:35:35] And, especially with a lot of the data center push, I think that we can learn a ton from all of the data center work that’s being done with these optical interfaces, and bring that to the machine vision world and introduce a standard that’s kind of the Goldilocks, right? It has the high bandwidth. It has the long length. It has no data loss. Right? There’s a lot of opportunities looking in the future. But I also think maybe getting harder these days. There’re a lot of different organizations around the world. They’re trying to work together, but everyone has their own interests. I think there’s a lot of challenges there, too. 

Steve Kinney: [00:36:10] I think you brought up an interesting point, though, in that too. And that’s the whole GPU thing, the Nvidias and the various forms of that. And I recognize a long time ago, at first I’m like, wow, the GPU is great. And it is. But in a lot of the cases, the actual pipeline into the GPU is the limit because it has to cross a CPU, go up to the GPU. So now some of these GPU direct standards, I think, will change things a little, because if we can get the image directly in there, almost like the old DMA direct transfers, to get it into the CPU, you’re going to start seeing that. And I wouldn’t be the one to forecast what’s going to come out of that. But I think, almost certainly, it’s important and that there needs to be some standardization there, but it has to be the next generation as well. 

Jimmy Carroll: [00:37:02] We’re going to get hit with a flood of people here shortly. But before we close out, is there anything that you guys are … I know it’s hard to use a crystal ball, but do you have any strong predictions for the next couple of years, things you expect to see in machine vision or hope for? 

Steve Kinney: [00:37:18] Massive adoption! Both expect and hope! I think we did say I didn’t pitch lights in this one, but I think it’s important. I realized when I saw cameras, I’m like, the cameras are always the last thing these guys think about, they do the whole system. No, I’m going to have vision and go, what camera do I put down? But I’ve learned, once I got on lighting, it’s the lighting and the light determines what even comes off the camera. I think all this though, just more, more, more adoption, all these things are coming out. I think it’s an exciting time to be in the vision space. And, yeah, it feels more and more important. 

Jason Covar: [00:37:59] Yeah. I think I’m optimistic for collaboration. I think when we’re here at, like, the Business Forum, you talk to many leaders and everyone has the same stories and same experiences, and you all shake your head and nod and say, yeah, I’ve done that, done that. And I think one of the ways to break that is collaboration, right? I think all of our goal is to hit scale, not only from an output perspective, but to make impact, putting it into factories and putting it into companies. So I’m optimistic there’s going to be more collaboration, working towards the same mission. And I think the end result of that will be some successful technology, exciting technology. So that’s what I’m looking forward to the most. 

Jeff Adolf: [00:38:37] Yeah, I think for me, maybe selfishly, I’m most excited about the creativity that’s about to be unlocked. I think the application spaces that we are able now to reach are broader than they’ve ever been by a large, large margin. So what’s limiting us in a lot of it is the creativity, to think about where vision can be used in applications that it’s not being used in. I think a lot of people are excited about incremental steps, things that have gotten faster: processing times, right? And those are really important things. They allow us to do new things. But what I’m most excited about is just, what’s new? What is totally undone right now that we can do with this combination of new sensors, with new processing, with new algorithms. 

Steve Kinney: [00:39:24] I think that’s a great point, though. We’ve talked about the next generation of engineers behind us and stuff in the forum this week, and I’ve told people before, I grew up in a generation before Apple IIc’s. I remember the first Apple IIc’s coming out, not being rich enough to do that. And going through that whole thing and learning and still growing up in this technology. And now these kids start with Arduinos, Raspberry PiS, all this at their hands as they come on as full-on engineers. All these technologies, everything we’ve talked about, is available to today. And I think the real excitement is just exactly what you said. But fueled by these kids that just take it for granted that this is now the new playing field. And what the heck can they imagine to do with this? 

Jimmy Carroll: [00:40:11] Well, yeah. It’s like that anecdote, there’s more processing power in your phone than there was in the computer that sent the first shuttle to the moon, right? 

Steve Kinney: [00:40:18] Exactly! 

Jimmy Carroll: [00:40:19] You know, some younger people don’t really realize what you have, right? I think it was Alex Cecchini earlier —— and if it wasn’t, I apologize —— but somebody said, and it’s sort of a common sentiment, it’s never been more exciting to be in vision and automation than it is now. And I think that’s going to be true next year and the year after and the year after. And that’s what makes this all so interesting and so much fun. And, so great to be a part of the community. So with that, I would like to close it out. But before I do —— SmartVisionLights.com, TheImagingSource.com, and 3M.com. And if you guys have any questions or comments for Jeff or Jason or Steven, I’d be happy to pass them along at Manufacturing-Matters.com. And with that, thank you guys so much for taking the time. Really appreciate it.