In this special machine vision panel discussion episode of the Manufacturing Matters podcast, Jason Covar (The Imaging Source), David Dechow (Motion AI), and Dale Deering (Teledyne) join TECH B2B Marketing’s Winn Hardin and Jimmy Carroll to discuss all things current and trending in the machine vision space. Topics include the intersection of robotics and machine vision, the real value of AI on the factory floor, line scan imaging, and the latest image sensor developments. Additional topics 3D imaging, SWIR and hyperspectral, industrial computing, and much more.

Jimmy Carroll: [00:00:06] Hi everybody. My name is Jimmy Carroll. I’m the vice president of operations at Tech B2B Marketing. And welcome to this special edition of the Manufacturing Matters podcast. I have the pleasure today of being joined by my colleague and friend Winn Hardin, along with a great group here: Dale Deering of Teledyne, Jason Covar of The Imaging Source, and David Dechow of Motion AI. First guys, thanks so much for taking the time. Really appreciate it. Glad to have you here. So I want to jump right in. And you guys had some meetings earlier today at the A3 Business Forum about, strategy meetings and machine vision in general. So let’s just start off right now and hopefully all you guys can chime in a little bit. But what are you excited about most in machine vision right now? And I guess David, if you want to start down on your end.

David Dechow: [00:00:49] Well, I know that we like to talk a lot about emerging technologies and what’s the latest and what are the trends in machine vision. I’m most excited, though, that machine vision as a fundamental enabling technology continues to be widely accepted. I see more and more people using machine vision and, again, not the pure cutting-edge technologies but the fundamental machine vision technologies to enable automation in a wide, wide variety of marketplaces. The marketplace usage is expanding as well. And not to completely cut out the emerging technologies, there’s a lot of excitement around some of those as well. And really technological advances even in things we’ve been using for years, like 3D imaging, hyperspectral imaging, and so on.

Winn Hardin: [00:01:41] How about you Dale?

Dale Deering: [00:01:44] We’re looking forward to a year of growth, which is one thing that’s always exciting to look forward to. And there’s an integration of technologies today, whether it’s the integration of AI into imaging that’s enabling better defect detection and acceleration of imaging use into different applications that perhaps traditionally was difficult. Or also the integration of technologies, whether you have SWIR imaging, different wavelengths that are being used, or hyperspectral imaging. And all these are finding new applications to solve and new ways to give people more data about the applications or the products that they’re trying to inspect. So these are some interesting trends.

Winn Hardin: [00:02:32] We’re going to talk more about SWIR and AI specifically in a little bit here.

Jason Covar: [00:02:37] Yeah, I think the most exciting thing for us in the vision part of the business is that vision is a component in everything that we see. You know, we talk about automation as a whole, but we’re an integral part of that. And I think that’s exciting for all of us, as well as some of the newer technologies in the sensors themselves is something I’m starting to see that will funnel down to the market. But I think, yeah, this will be interesting for us to see as the year goes on.

Winn Hardin: [00:03:07] That’s a perfect segue to the next question we were going to ask, which was, we’ve heard for years that machine vision tends to follow robotics around the world. You’re going to get a lot of robotics adoption, and the machine vision, they’ll start doing more advanced VGR-type systems. Can you talk a little bit about the relationship between robotics and machine vision right now? And is it complementary and beneficial to everybody or how does that go?

David Dechow: [00:03:35] I’ll jump in. I think that first of all, I liked what you said about machine vision following robotics. But I want to clarify that a little bit. Machine vision, we’ve always taken the marketplace and let’s say the success of machine vision and tracked it with robotics, much like we track it with the manufacturing index, other manufacturing indices, and machine vision as a marketplace or in terms of its success in the marketplace, in the numbers does track robotics. But I see machine vision as a little bit of a different beast when we think of it with robotics. Machine vision is without a doubt an enabling technology for robotics. And one of the things I think really emerging in the last perhaps 10 or 15 years is that more and more advanced robotics must have machine vision in order for it to succeed. And in some cases, the marriage between robotics and machine vision has become seamless. It’s become something that we expect to see, expect to work, to just plain work. But it’s not a trivial thing. What we sometimes see as being the marriage between robotics and machine vision, while working, is still, as you mentioned, is still an emerging trend. It’s still an emerging technology in the marketplace. We had a great discussion this morning about that, that robots, I think, many of us think in the market, that robots are becoming a technology that needs machine vision as an enabling force. And that’s great for vision, that’s also great for automation in general and advancing automation in a variety of ways, even into the more esoteric robotics like humanoid robotics and AMRs.

Dale Deering: [00:05:40] And I would just echo some of that. I mean, if you look at a technology generation 10 years ago or 15 years ago, you had robotics or automation that perhaps didn’t require imaging, but those are solved. Those applications are done a long time ago. The applications today that they’re working to solve are things like fruit picking or agriculture, things like that. They’re not in the same location every time. You must have some sensing in order to tell where they are. So image sensing is today required for all of the applications I think that robotics is exploring or advancing. And even not just visible imaging but infrared imaging and different kinds of information that you want to gather, like in fruit, you need the ripeness and things like that. So you need the size and the shape so you know how to pick it.

Winn Hardin: [00:06:33] Agtech particularly excites me an awful lot and just food production in general, right. That industry has typically been a laggard. But I hear you. And so it sounds like it’s very complementary and it’s hoping to grow the market potential for both technologies. Is how we’re manufacturing products also impacting that? We talked for years about flexible manufacturing, short runs, more customization, things of that nature. Are we having a give and take where as machine vision and robotics get better, they’re enabling manufacturing industries, OEMs, product producers to actually do more customization or maybe come faster to market? But is it a give and take both between our customers and . . .

Dale Deering: [00:07:24] Yeah, I think there has to be. I think there has to be an adaptation if you’re using a robotic process, there are certain constraints that make things easier. And just to use the fruit example, in our area, they grow a lot of fruit trees, like apple trees. Even the way they’re growing the apple trees today are different so that they can more easily automatically pick them. So if you have a huge apple tree and you have a very complex branch structure, it’s very difficult mechanically to get in there, whereas if you grow them sort of two dimensionally, then it’s very easy to access them.

Winn Hardin: [00:07:58] And so it’s actually impacting the pruning of the tree.

Dale Deering: [00:08:00] Absolutely.

Winn Hardin: [00:08:01] That’s wicked cool.

David Dechow: [00:08:02] Dale stepped down this path just a little bit. But one of the things that is, again, another very exciting thing about machine vision in today’s marketplace is its close marriage with the ideas of smart manufacturing, smart factories. The idea, and you mentioned, of having more flexible automation, but not only flexible automation but also automation that is a data consumer of information that we can provide as machine vision technologists. We can provide that data to a big data-consuming factory and in turn allow that factory to make better decisions, to process better, to improve their productivity, because they can collect the data now that we have available.

Jason Covar: [00:08:56] Agreed. And I think it creates efficiencies. I think where you’re going with that. And that efficiency is then pushed to either more productivity or to other parts of the overall process. And it does come from from imaging. So I think we’ve all gone into customer environments where you first put vision in and they start to think, “Oh with this data,” to David’s point, “what more can we do?” And so you compound that efficiency created to then further produce.

Winn Hardin: [00:09:24] And it’s always blown my mind that machine vision had this ability to monitor, identify problems before, but so few customers back in the day were really leveraging it. It seems like the world now is just completely embracing the data lake as we all know. So it’s nice that the world is finally catching up and leveraging what machine vision can deliver.

Jimmy Carroll: [00:09:44] I think food production is a good jumping-off point to a question I wanted to ask you guys, and I agree, it does seem like food production is an industry or market that’s ripe for growth. Pun very much intended. AI, we have to ask. We have to talk about AI. But it does seem like food production is an area where AI could really lend itself in terms of being able to handle inspection of organic matter and parts and products or whatever that are in different orientations. And every piece of chicken breast or orange or whatever, they’re all different. Talk a little bit about that. You know, where do you see AI fitting in with when it comes to food inspection? But where else? Where else is AI adding actual value on the factory floor or beyond the factory floor today?

David Dechow: [00:10:40] Well, again, I do like the topic. It’s a broad topic when we talk about AI and imaging. But just starting with the food side. Food, as you are alluding to, is just a natural, again, pun intended, a natural product that is conducive to the use of AI for inspection of the various things that we need to do in food. Being a natural product, the natural product variations are difficult to identify in rule-based processing, or algorithmic-based processing if you want to call it that. And it has been a real boon to that type of an industry or to that industry and others to have AI in terms of segmentation, classification, be able to identify the objects and the features that are necessary in that space. And back to something Dale said. This includes using specialized imaging along with AI to get the image that’s necessary and then extract the data from that image. So definitely food is a huge, huge use case and it will and has benefited directly from AI. That said, AI, as you know, my take on that, is not one thing that solves everything for all of us in machine vision. And we found the better implementations of AI and machine vision make use of a complementary mix of both algorithmic and AI functions.

Dale Deering: [00:12:28] I think another area that AI is quite interesting for food production. In food production, I guess traditionally we’ve seen visible imaging, which is looking at the color of the fruit or the color of the vegetable and maybe sorting based on that or shape and size. But increasingly, there’s also other criterion that food producers find important to inspect. And that could be things like sugar content or protein content or things that aren’t immediately in the visible spectrum. And so that’s where spectroscopy and SWIR imaging and other technologies are helping to isolate different things about the food quality that provides more of the data. And AI in the use of that is able, again, instead of just a rules-based approach, you have a combination of different sensing elements or characteristics that you need to combine together.

David Dechow: [00:13:24] And in something like hyperspectral imaging, particularly in the shortwave infrared wavelengths, that imaging data extraction has been definitely improved by the capability of AI in the classification side of the spectral signatures. Definitely a strong use case in there.

Jason Covar: [00:13:48] Yeah. And then even beyond the processing side of it and the efficiencies that are created there, you can use these technologies once it’s picked to determine how far from production to consumer. You can make predictive. And so you can reduce waste. Because if I grow an orange here, I can predict the ripeness and how far out it would be fresh. So there’s benefits even beyond just the production. And I think a lot of that comes from these technologies as well.

David Dechow: [00:14:20] Isn’t that an extension too of the smart factory, big data idea, is that we provide the data, but with the advent of AI, it gives the manufacturers or those creating that software the ability to make some of those predictive decisions as they use the information.

Jason Covar: [00:14:41] Absolutely.

Winn Hardin: [00:14:42] So we’ve talked about SWIR a number of times already. And we mentioned hyperspectral imaging, which is interesting always because traditionally hyperspectral has been a very expensive solution. So normally you’d use hyper-systems to identify your spectral bands, optimize in a multispectral, and go to market. We haven’t talked about time of flight or that type of approach for 3D sensing and everything. So, Jason, do you have any thoughts in terms beyond of where the sensor is going and what’s cool and hot?

Jason Covar: [00:15:15] Yeah, I believe these technologies are trending in the right direction. But I think we’re going to see even smarter CMOS chips that are doing things on chip that help from a design perspective, help companies that manufacture cameras as well as create different features for specific markets. So I think this is actually a very exciting part of the industry that we’ll start to see hit market this year as well as next year.

Winn Hardin: [00:15:42] Now is that primarily for data processing or is this actually going to enable new features and stuff that maybe the audience isn’t as aware of?

Jason Covar: [00:15:49] Yeah, I believe the latter. Yeah, I think there’s things that these technologies are being used today, but I think we’ll see others that are not in the market today that will very much complement what’s there as well as give us new ideas.

Winn Hardin: [00:16:04] Fantastic.

Dale Deering: [00:16:06] Yeah. I was going to say, I think another application for time of flight or one of the applications that seems to be a target is in human safety. And it can be in a work cell environment, where you need to understand how far away an operator is from a certain piece of equipment. And I think the integration of these technologies, let’s say, with long wave infrared, where you can now actually assess a thermal signature and say, that’s not just an object, that’s a person. How far away they are. Or for autonomous vehicles, where you can say, okay, when the bus stops, we have to make sure everybody is clear before we restart again. And so being able to measure distance as well as measuring the thermal signature or something to say this is a person, not just a lamppost, is important.

Winn Hardin: [00:16:54] Are the component costs going to come down to when we’re actually going to be able to deploy that widely? I know it’s always been a bit of a bugaboo.

Dale Deering: [00:17:02] For sure, it’s all volume related, right? Just like lidar systems and cars, SWIR imaging for computing.

Winn Hardin: [00:17:10] I know we had a couple of new sensors come out a few years ago with Sony’s chip and someone else, but they were still pricing them at 12 grand, I think 12 to 15 or something in that ballpark.

Dale Deering: [00:17:21] You will see correction in this side of the business to build product. I think there is the market itself and the numbers. But you’ll see it trending down.

Winn Hardin: [00:17:32] Beautiful. Beautiful. Well, that’s how those things go, right? Component costs go down, up to distribution, installations, and units.

David Dechow: [00:17:40] Well, one of the things that has always driven component costs down is demand and usage. And that’s part of the problem here, and I love to hear the discussions about 3D, and time of flight of course being a 3D technology, in that as the marketplace finds more and more applications and is more and more comfortable with implementing that sort of technology in applications, it stands to reason that the costs are going to come down. SWIR is kind of a mysterious technology right now for many. Hyperspectral certainly is kind of a mysterious technology, so it’s not as widely implemented. I’m excited. I think 3D, SWIR, hyperspectral, including time of flight in 3D, are things that are going to continue to be more widely implemented, and then we’ll have cheaper costs.

Winn Hardin: [00:18:38] Yeah, there you go.

Jimmy Carroll: [00:18:39] Well, especially with the release of the newer Sony chips. Those are less expensive, aren’t they? You’re going to make these cameras more accessible to people than ever before.

David Dechow: [00:18:49] The Sony chip, the DepthSense that you’re talking about and similar chips that they’ve released have opened up recognition of the products and made it more of a commodity to create the items. It’s no longer very difficult to create a time of flight camera because you implement the Sony chip. Still, I think the market has to embrace that. And we as technicians have to be able to specify it better into the marketplace, or I should say more often into the marketplace.

Winn Hardin: [00:19:25] So, Dale, I got to ask you, we haven’t gotten into line scan. Teledyne being one of the leading suppliers of line scanning cameras in the world.  What’s hot in there? What’s beyond TDI and super high speed? Where are we going?

Dale Deering: [00:19:36] Well, we’ve gone the next step, actually, in super high speed. And we now have a new camera that runs at 1 megahertz line rate. And these kind of line rates are being requested by several different industries. And where does it stop? I think there’s always that push for lower cost. And if you’re trying to inspect very small feature size, you have to inspect a large area, well, you need something that’s very fast or it takes a long time. So I think that’s one aspect. The other aspect is we’ve talked about having different inspections, different angles of inspection, different characteristics or filters that people are looking at. The intention really is to get more detail about what I’m inspecting so I can tell, for example, if it’s a defect. Maybe I can also tell the material that it is. So I know what part of my process I have to fix. For example, in glass manufacturing, if you’re starting to see particulates and it could be the refractory materials that are actually breaking down and you need to replace something upstream. So I think it becomes, as customers find that, “Oh, imaging is really good to improve our yield,” now they want more detail about what exactly do I need to improve so that I can fix my process. So it adds further cost improvement to their process, and we talked about this before, as they find more applications for, “Okay, I’m doing this imaging now. And now I realize I can also do this.” So line scan also because you’re taking the data line by line, the other advantage is you can start to process that data faster. So you don’t have to wait until you get a frame and then you transfer that frame and then you process it on a frame. And customers are starting to realize, oh, it could be an advantage. I can actually process faster if I look at a thin rectangle, let’s say, of image data. And for some products, for example, if you’re sorting nuts or seeds, you don’t necessarily need a whole frame of data before you can actually start your processing. So you can actually speed up the applications by doing that. So that’s been a win for some customers as well.

Winn Hardin: [00:21:53] And that kind of touches a little bit on what you were talking about earlier, Jason. With the smart cameras bringing in additional processing.

Jason Covar: [00:21:59] Correct. I think there’s two parts. I think the chips, you’ll start to see more specific data driven from the chip level, but then from the compute side as well. Obviously as we’re getting to GPUs and we start to work in processing, we’re able to process more, closer to the problem. And so I think that’s another area that we’ll see as well.

Winn Hardin: [00:22:20] Absolutely.

Jimmy Carroll: [00:22:22] Just out of curiosity, Dale, and David I guess this would be a good question for you, actually Jason, all three of you guys, but what are some of those applications? When you think of line scan imaging and web inspection. It’s textiles. And you mentioned glass. What are some of the applications where they’re demanding higher speeds and new capabilities, just out of curiosity?

Dale Deering: [00:22:42] Well, one of the biggest ones right now, and we’re seeing it here in the US and a number of other countries, is the ability to inspect semiconductors. So today semiconductor production continues to increase. You know, we’re using semiconductors all over the place and more and more in different applications like electric cars. All these things demand semiconductors. So the inspection of those and the yield that you get from those is very dependent on inspection. And so when you’re looking at a semiconductor that today I think is that 3 nanometer and soon 2 nanometer type patterning and even the ones that aren’t at those advanced nodes, you need very small wavelengths of light and you need very small feature sizes. Well, when you’re inspecting in, let’s say, submicron, now you need very fast in order to process the 300 millimeter wafer. So these are the kinds of things that are driving speed and driving line scan in some of the advanced markets. The other ones are in medicine. And today there’s a lot of genetic material or proteomics or things that have fluorescence imaging. And when you’re looking at the cellular level and you’re trying to catch fluorescence from those cellular components, again you need very high magnification. And so in order to process that you need very high speed. And TDI is one of those technologies that because you have the multiple exposures, you can have a very small amount of light that’s fluorescing off the material, and you’re able to actually differentiate between the different components.

David Dechow: [00:24:20] Those are important tasks. But sometimes maybe some of the more mundane tasks have been improved by the higher-speed line scan as well. Line scan, of course, has long been a target or not only a target but an indicated technology for moving web applications in a variety of things. You mentioned printing, textiles, and so on. In some of these cases, the product has always been moving extremely fast. And in integrating even a line scan application, we find that we had to sample, we still had sample images. In other words, we’d sample for a short period of time and then process the image. At the higher rates, with the data coming in faster, we have more often been able to achieve continuous acquisition of the surface and not drop out. And that’s just one example. But certainly the higher-speed line scan is going to be a boon for many of these applications.

Winn Hardin: [00:25:42] Are the optics keeping up with all these sensor advances that we’re talking about? I mean, the larger the area, it becomes really tricky to design and develop. In some cases the optic may be the thing that breaks an application, right? Because it’s just so expensive compared to the sensor. And it seems very hard to fix that because it’s just basic physics, right? It is what it is.

Dale Deering: [00:26:05] Well, there’s a lot of metasurface optics and things that are coming today as well, which enable different ways to solve the problem.

Winn Hardin: [00:26:12] Awesome.

Dale Deering: [00:26:13] So yeah, for sure, changes in optics are very interesting.

Winn Hardin: [00:26:17] So we’ve talked a lot about illumination. I’m going to throw one back to the AI crowd here again. So I understand that there’s discussion about whether AI means illumination is no longer relevant. We don’t care about the shape of the photon. Is that what we can expect? Is AI going to solve all of our problems with lighting?

Jason Covar: [00:26:39] No, I think all those that have have worked with these projects and tried to solve these problems, you can’t totally get there when we talk about AI and creating these systems. And one thing I’ve seen on our side is they’re starting to use synthetic data to try to offset what couldn’t be done with cameras, just collecting a bunch of real-world data and trying to create applications. So, yeah, purely with just ambient light, that’s not a solution to get you all the way there, certainly.

David Dechow: [00:27:14] The real underlying point of competent illumination with machine vision, or if we’re going to call it computer vision, machine vision, AI vision, whatever we want to call it, is in a manufacturing environment to create reliability and repeatability. Ambient light, I really say this without equivocation, ambient light cannot do that in the manufacturing environment. So yeah, there was early on and there still are people saying today, “Oh, we have AI, we can do it all with ambient illumination.” You can, up to a certain point, but you can’t risk a manufacturing environment and the reliability and the precision of the inspection or the robotic guidance or whatever it is you’re doing. You can’t risk that in a manufacturing environment by not competently implementing suitable illumination. And even with AI, the illumination is there to try to create better contrast between the features you like and the features you don’t like. And ambient light just can’t do that and can’t do it reliably.

Winn Hardin: [00:28:37] If we fail to pick an apple on an apple tree, it’s not the end of the world. But if we produce a thousand widgets with any kind of investment cost . . .

Jason Covar: [00:28:46] I think that’s a good point too, from machine vision to computer vision and where we fall. But those where you can’t standardize lighting, very dynamic, and some of these innovative projects, again, going back, I think people are trying to get to that final piece with synthetic data. And I do think that’s pretty interesting. I think that’s the area that you can close that final gap.

Winn Hardin: [00:29:11] Gotcha. Dale, you look like you had a thought there, man.

Dale Deering: [00:29:13] Well, I was just saying, in some applications, like if you’re inspecting automotive parts, for example, I mean, they’re looking for single-digit PPMs. And so you need consistency as much as possible in order to get a consistent result. Now, it’s not to say that there aren’t advances that you can use, certainly in slower applications, that you can use ambient light. It really comes down to the physics of how many photons can you collect per unit of time. But it really kind of depends on what you’re trying to achieve.

Winn Hardin: [00:29:46] Yeah, as well as what’s at risk. It could be loss if it doesn’t come through properly.

Jimmy Carroll: [00:29:51] We’ve talked a lot about some of the latest advancements in vision, from AI and SWIR hyperspectral and 3D, but what other recent advancements out there in vision, and maybe it’s in vision standards, are you guys excited about right now?

David Dechow: [00:30:08] I think that the advances at the moment in the marketplace are subtle but important to the technologies. I personally don’t see, and I think we’ve heard a couple of possibilities of things that could be advances, but I don’t see something that is a game changer in the industry right now technologically. That said though, the subtle advances in usability in software, in the packaging of the components and packaging of the devices into more of a system-based or solution-based device or whatever, I think that those kind of subtle advances have perhaps not been advertised as much in the last five years, but those are the things that are really pushing machine vision forward I think. 3D imaging is one of those, and we’ve mentioned it already, but the 3D imaging devices that are available on the marketplace are totally different than what they were five to 10 years ago. You can pull a device right out of the box and set it up and have it deliver a calibrated image to a processing app in 30 minutes and be analyzing a 3D image. And that’s just one example, but that’s a strong example of how this subtle change in some of these existing technologies is moving forward.

Dale Deering: [00:31:50] And I would say maybe three elements. One is ease of use, as David mentioned, being able to use the technology faster, for the factory floor skill set to be able to pick up and actually implement that faster. And maybe that’s not new, but it certainly continues to accelerate. I think as these technologies come down in price and they get, as Jason mentioned, more integrated at the chip level, the overall prices come down, so that allows those technologies to be able to be used in more applications now instead of just sort of high-value applications, they can hit a much broader base of applications. And just the integration of features together. And one of the ones that I said is time of flight and thermal. So when you integrate these things together, you’re actually able to do something significantly more than either of them could do independently. And I think we’re going to see more of that as well in the coming year.

Jason Covar: [00:32:53] And I think to get the adoption that’s needed, I think that’s on all of us to not only make the technologies aware but make sure that they can coexist with one another, work well together, is very important for us as well, because if you have a very good technology that’s on its own that does not work with others, it doesn’t really have a ability to scale. So I think that’s actually something that for us is important. And I think the standards is key. We saw that with GigE Vision and USB Vision, what that does for people to be able to adopt these technologies quicker. So I think that’s actually a challenge for us as we get more innovative and get these newer technologies to keep up with that, so that the adoption rate can be equally as quick as our standard products today.

Winn Hardin: [00:33:44] When I hear you guys talking, and it makes me think about embedded vision systems, right? So many of these systems for the largest volume scale are going to be done by nontraditional machine vision companies. But I’ve always wondered what part do we play? What part does traditional machine vision play when transitioning this technology to the high volumes, whether it’s an automotive carmaker or John Deere or whatever we’re looking at?

Jason Covar: [00:34:10] So we’ve been working on those since 2019. And the reality is, it’s fragmented, because you have compute, you have an OS level, you have a different sensor. And so it’s very difficult, back to David’s point, to be able to package that, to put it in a position to be adopted is the challenge for all of us. And so we have to play together. And I think what often happens is people start to play generalist. And so then you kind of step out of your expertise to try to figure something out when it’s fragmented. And I think the reality is, if the specialists stick together and start to work together and collaborate, I think this is the way that you can truly push this into our industry. And I think that’s our opportunity.

Winn Hardin: [00:35:00] Gotcha. Well, that offers huge potential. I mean, the growth of embedded vision.

Jason Covar: [00:35:06] Definitely.

Winn Hardin: [00:35:07] Even if we’re not directly making those Caterpillar vehicles, actually you probably are making them, inspecting their welds and every other thing. But if Caterpillar is still making their own, just the acceptance and the general awareness and knowledge, it seems like creating, you were talking about how to make these systems more applicable for factory floor workers. And how do they intuit? I mean, the cellphone’s probably gone a long way down that path of getting people more familiar with imaging and how lighting affects it. And just finding patterns, taking your photo bomber out of your shot real quick.

Jason Covar: [00:35:46] Well, I think that’s also a challenge for us in imaging, right, that everybody assumes your phone, we talked about that earlier, it must be this quality and this is imaging. And I think that’s where we have to kind of correct course and really get people back to actual machine vision and computer vision requirements.

Winn Hardin: [00:36:09] So were you implying that people think that their phone should be able to be a machine vision system, essentially? Because I’ve heard that bandied around.

David Dechow: [00:36:16] I’ve seen that, yes.

Jason Covar: [00:36:17] Yeah. We’ve seen that and that machine vision systems should be like your iPhone. So you see both sides of it. 

David Dechow: [00:36:27] That goes back to the imaging though. The same thing is true always, is that you can’t hold up your phone and make it reliable and precise in terms of creating an image from which data can be extracted. It just doesn’t work. You could make it happen in some very constrained cases, but it just doesn’t work.

Jason Covar: [00:36:49] And I think as we have people coming into our industry because imaging is important,  particularly from the software side, that’s our responsibility to train them and truly let them know what the requirements are.

David Dechow: [00:37:04] Well, you mentioned embedded. I think embedded is an interesting thing in our industry. Maybe a little bit misunderstood. Sometimes I’m not sure what embedded is in our industry. But when we think about board-level cameras, system-on-chip, and so on, and I know both The Imaging Source and Teledyne have these kind of products, and I’ve used them myself, I think it’s very interesting to continue to try to find ways that embedded can be successful and be viable in manufacturing. One of the problems is that in order for it to be worth my while as an integrator to use embedded, I have to have a solution or a use case that would use many multiples of them, because the non-recurring engineering up front to integrate it into a coherent solution is quite a lot. And so you save money on the embedding, but you spend a lot more time in engineering. I’m excited about it. I’m working on a couple of embedded applications right now that I think have promise. And I think as people, again, come to understand it more, we’ll see more usage.

Winn Hardin: [00:38:27] I’m just glad that we’re able to participate more and more in that, especially, as you pointed out, high-volume, chip-on-board, or camera-on-board are making that happen.

Jason Covar: [00:38:37] The opportunity is big because it creates an efficiency. So you have kind of specialized hardware closer to the physics of problems that you can solve. So yeah, if we get it right, the potential is very big.

Jimmy Carroll: [00:38:51] Are there other technologies? So we’ve talked about these, and 3D is one. David, we’ve been talking about this for years, like, 3D now, you could say it’s kind of come of age and especially with things like lower-cost 3D, like with the availability of RealSense. Are there any other technologies or maybe it’s something we’ve already discussed, but are there any other technologies you could see as kind of being a breakthrough in the next couple of years becoming more readily available, like maybe SWIR, for example, with the with the new Sony sensors Like will SWIR start to finally gain traction in the market or anything else?

David Dechow: [00:39:31] Well, that’s a broad question.

Winn Hardin: [00:39:34] It’s kind of like, is there anything cool that we haven’t mentioned?

David Dechow: [00:39:38] There is a bunch of cool stuff, and there’s always cool stuff coming up. Well, let me mention this SWIR first. So Sony VSWIR. A great product. It does, once again, commoditize the SWIR cameras. There are limitations to that sensor and to a camera that uses that sensor, including its capability, signal noise, and also the limit of its wavelength response, where in some cases we want to go to upwards of 2100, 2500 nanometers in SWIR, and the Sony tops off at around 1700. But that doesn’t mean it shouldn’t be used. It just means that we have to know the technology, and in any of these cases, know the technology and be able to integrate it competently. So I do like those kind of product lines. But one other thing I’ll just make mention, when you ask about emerging and brand-new and weird technologies, one of the other things is that we see technologies out there, I see technologies out there that are really nice, nice technologies, but I call them technologies searching for a use case. They’re out there searching for a use case. One of these I love seeing when it first emerged from PROPHESEE is the event sensor. Great product. Now Sony has taken them under their wing as they have other technologies. There are some dedicated and good targeted use cases, but in general purpose machine vision, it’s still to me a technology seeking a use case. And we see that quite often in these really brand-new and cutting-edge technologies is they’re great. They look exciting. They really are very, very interesting. But they have to mature a little bit before we find as integrators and end users, before we find good use cases for those kind of things.

Winn Hardin: [00:41:53] Chicken and egg.

David Dechow: [00:41:54] Chicken and egg.

Winn Hardin: [00:41:55] Can’t develop applications till you’ve got the hardware.

Jason Covar: [00:41:56] I think the software part is also important. A lot of times we talk about these technologies and the chips and the data that comes from them. But it’s non-trivial with a lot of these once you get the data, are you actually able to utilize it? And so I think that’s an important part of this as well. As these technologies come out that the software is not forgotten.

Winn Hardin: [00:42:24] Yeah. At least the good news there is, there’s a whole lot more coders out there than there are FPGA designers or whatever else we’re looking for.

David Dechow: [00:42:34] But the marketplace, just as a quick pitch here, the marketplace needs more machine vision engineers. No question. We need more machine vision engineers who can look at a problem, create a design, and implement a design either in code or in configurable software. We really need more engineers.

Winn Hardin: [00:42:55] That gets back to training, you know. And you don’t see a whole lot of college curriculum where you’ve got computer vision built into it. More so now than than ever before. I know A3 is, of course, working on that. And I think, David, you’ve spent most of your life training people in the machine vision space. But I agree.  I flashed to Coursera and AI. We just need that ability to push more out to younger people and make it alive. And we see that in every automation space. I mean there’s programs like, what is it, Champion? For teaching on CNC work. It’s a rather long acronym. Don’t ask me what each letter stands for. So any other last thoughts, Jim?

Jimmy Carroll: [00:43:45] No, not really. I mean, my fun one always to ask is, do you have any any predictions for the next couple of years? We’ve sort of covered that. But if anybody does have any, especially if they’re bold, I’d love to hear those.

David Dechow: [00:43:56] It’s easy to make bold predictions when you’re not required to be right.

Winn Hardin: [00:44:04] Those are the best kind.

David Dechow: [00:44:07] Futurists have it great. Futurists originally predicted we’d be exploring space by 2020. But we’re not that much exploring space. Futurists have it great. They don’t ever have to be right. So we could certainly give you a bunch of predictions.

Winn Hardin: [00:44:29] Because if you’re right every time, you can’t write a new book every three years.

David Dechow: [00:44:32] That’s right.

Dale Deering: [00:44:33] You have to talk to the science fiction writers.

Jason Covar: [00:44:35] I think you’ll see collaborations at the specialist level, with equal parts in and equal parts out. That’s my prediction. There’s a lot of collaboration with a lot of companies, but I think you’ll see a true equal parts in, equal parts out, specialists getting together to push things forward. I think you’ll see this in the next few years.

Winn Hardin: [00:45:01] More collaboration. Would you say that’s a change? I mean has it always been that way?

Jason Covar: [00:45:06] Well, I think there’s different layers of collaboration, where I can take your component and make it work with my component. But what if both lanes are developing a technology together with their level of expertise for the common good of pushing things forward, because typically there’s always a separation. I think you’ll start to see this. The market, it’s challenging now in certain parts of the world, but I think this will drive that initiative.

Winn Hardin: [00:45:36] Indeed. Any final thoughts, Dale? I mean, you work for one of the largest machine vision companies on the planet. And you’re connected to a lot of other amazing technology companies.

Dale Deering: [00:45:47] Well, and I guess that is one advantage within Teledyne because we have so much technology generation. It’s a frequent occurrence to work together and be able to say, okay, actually, we can borrow some of this technology and we can work together with another group, and we are currently doing that in a number of technologies, not all of which we are ready to promote.

Winn Hardin: [00:46:13] He’s not dropping any Easter eggs today. Well, gentlemen, thank you so much for taking your time today. I know it’s been a busy day. We’re getting ready for the A3 Business Forum to kick off tomorrow morning. So it’s such a pleasure to have you on. I really appreciate it.

Jimmy Carroll: [00:46:28] If anybody has any questions for any of these guys, feel free to reach out to us at Manufacturing-matters.com. And thanks for tuning in.

Winn Hardin: [00:46:37] Until next time.