The Manufacturing Executive
The Manufacturing Executive

Episode · 1 year ago

Custom Machine Vision & Robotics Solutions w/ Jonathan Berte

ABOUT THIS EPISODE

Mass production alone is not enough. We all want customized products.

Twenty years ago, it was easy. Educate your workers about your single unique product. In a world driven to customization, however, no worker can learn 400 different products. No *human* worker, that is.

But a digital worker might. Could robotic camera vision streams be the solution?

In today's episode, I talk about building custom machine vision and robotics solutions with Jonathan Berte, Founder at Robovision.ai.

Here's what Jonathan and I discussed:

  1. His personal journey into the vision component of robotics
  2. How AI's vision is advancing right now
  3. The most important applications of AI in the manufacturing process

To ensure that you never miss an episode of The Manufacturing Executive, subscribe on Apple Podcasts, or Spotify, or here.

Listening on a desktop & can’t see the links? Just search for The Manufacturing Executive in your favorite podcast player.

That this moment mass production alone is not enough. We all want customized products and this customization is just hiring the manufacturing complexity and you can and just educate the worker with a variety of four hundred different types of products that were just one unique type twenty years ago. So it's just much more easy to put these flows into republic camera vision streams. Welcome to the manufacturing executive podcast, where we explore the strategies and experiences that are driving midsize manufacturers forward. Here you'll discover new insights from passionate manufacturing leaders who have compelling stories to share about their successes and struggles, and you'll learn from B tob sales and marketing experts about how to apply actionable business development strategies inside your business. Let's get into the show. Welcome to another episode of the Manufacturing Executive podcast. I'm Joe Sullivan, your host and a CO founder of the Industrial Marketing Agency guerrilla seventy six. If you ask my four year old son Jack about robots, he'll go running for his optimus prime figure or one of the other transformers that are scattered across my house by the end of every day. When I hear the word robot, I buy default think of big yellow robotic arms on an assembly line moving a hundred miles an hour. My guess today, however, is an expert in a different part of the industrial robot and that's the eyes. So let me take a moment to introduce him. Jonathan Birte is the mastermind and driving force behind Robo Vision, the collaborative platform for those looking to lead the artificial intelligence, or AI revolution. With a background as an applied physics engineer specialized in image processing, Jonathan built up a strong reputation building custom machine vision and robotic solutions in the first years of robovisions existence. The robot robovision Ai Software is what it is today in large part due to Jonathan building machine vision applications the hard way. Jonathan now passionately leads Robo Vision to new heights every day. Jonathan, welcome to the show. Thank you sure, welcome, glad to be here. Yeah, it's great to have you. This is such an interesting topic and you know, I talked to a lot of people who touch robotics and automation from a variety of perspectives, but you're kind of in this this corner related to vision. And you got some deep expertise there and I'm so I'm very curious to learn from you and sort of expand my own knowledge as you talk about, you know, your expertise. Okay, so you know, I guess like, like I said, Jonathan, to kick it off here, I know you're in this world of vision, the vision component of Ai, and I'm just curious if you could start by telling our audience a little bit about your personal journey that led you into that Ntche and then inspired you the found Robo Vision. Well, that's a good...

...question. Show it all started about twenty five years ago when I started experimenting with cameras and when I realized that is migration from analog cameras to digital cameras had also some a great potential, and I started basically dreaming about the potential when I was at University of syric Switzerland, yet programming about that could look behind the back of a Croupier in a casino. So just to follow the game, and that's where I realized that not only these gimmicky applications would be cool, but also image processing and everything related to getting value from image dreams cameras also in production conditions would at some point in time be very disruptive and I believe that early on I started slowly as a consultant and the automotive market was really the careful first steps. We all know that the industrial space is very conservative, very robustness driven, typical component lifetimes of five to seven years. So it was not an easy gaining the beginning, especially because of PLC's yet a standard as seven seamens and Bradley, the usual suspects, Omron being there, and this camera ecosystem was kind of a new kid on the block, but also a difficult new kid on the block because it was always the last component in an automation project and the most tricky one because you didn't have the production data yet. So that's where I felt the heart. Way would industrial conditions are back then. I really had a tough time those first year's two thousand and four, two thousand and five, and convincing people that there could be potential in this camera application. But there was simply not so much support. There was no where, no large code bases out there. There was the term a I was was really not used at the time and it was seen as an extra component to appealc so every budget was like too much, and can't you reduce the budget somehow with less features? But in the end it was a very challenging environment. That up until one points when things started to change, when digital cameras game more and more commoditized and and this whole ai revolution basically starting off in at Stanford, California, and with faith a Lee imagenet started to take off a people started talking about deep learning, and that's when I deeply understood that the biggest instruction of all was not the camera itself, looking at production conditions and quality control production line. Now it was the algorithm creation. It was around two thousand and ten, two thousand and eleven, when I basically started all over again. Instead, it, like my business models, not scalable. I need to do something about the...

...algorithm creation itself. That's when I went back to university. Personal friend of mine that was professor at the time a really cool machine learning lap that t headed and at the time some cool new guys were in this lab center, Dealman and aren't under art that are now like high profile deep mind engineers, and they explained me that's writing algorithmic software. who was coming to an end with the advent of deep learning, and that's why, around two thirteen, I really started off with robovision two zero. We had a I mean designing a platform that made algorithm creation in the visual domain automatic, and that's, yeah, that's the basis of the company. Basically, like you can create algorithms, whether it's quality control and the industry or covid detection in healthcare, you can automate this process, which is the biggest disruption of all. Joe, that's a really great background. It's it's such an interesting technological advancement. I'm just curious if you could talk a little bit about, you know, some of the specifics of how the vision component of Ai's advancing right now and like what's possible now that may not have been possible or even a few years ago. Well, basically, if you look at the fission problem, you typically have this aspect of an object or something that you need to detect or some some aspects or artifacts that you need to detect from the object, and you have background and it bust. Like fifteen years to ten to fifteen years ago, you had to engineer all of this background substraction for ground extraction, feature detection robustness. You had to do it at all by yourself, meaning that it was really you needed to hand pick the features, you needed to know all about the possibilities. So typically, a profile that had to write such an algorithm was already like PhD degree. I mean you needed like a PhD to be successful and then to create a robust image processing algorithm. Fifteen years ago now, this changed when researchers started to reverse engineer the visual pot way in the brain and they realized that, they started to realize some of the basic aspects of how a biological brain, a human brain, is processing visual information. And basically that was the advent of deep learning when actually the brain people started to reverse engineering the brain and and this feature detection that was like a manual task fifteen years ago became automate it. And it's, of course, a big game changer because those are the most expensive profiles in the industry, is the algorithm creators. So if you can just out to make the smartest people, all of a sudden you can just do many more use...

...cases and make many more algorithms, and that's the era we live in. It's the era of the comdetization of difficult algorithms, especially in the visual domain. What are some of the most important applications of AI and the manufacturing process from your perspective, of Jonathan, and also it's curious if you could maybe share a real life example, whether it's from Rebo Vision or, if you you know, from from a customer or just something tangible, to put some of this in contact for listeners. We all wear clothes, like rousers, yeah, anything. So these are processes that are going extremely fast. But if something goes wrong with the needles and you have this textile being woven and very high speeds, if one needle breaks or something, and if it's not like noticed early enough, you have like literally hundreds and hundreds of meters wasted. Nextile, the really needs to go, I mean down the drain, and this is where ai can be of othermost ex can we can really function well. It's like to detect these anomalies in involving textile and, as such, stop the production process if a needle breaks so not to waste any more material. That's a great example. So quality control, quality assurance would be sort of one place. Anywhere else that you done, you touch on object segmentation, like everything related to detecting an object, the pose of an object, for instance a glue process, which we did for the Audi factory in Brussels, like a many parts of a car being glued with high tech glues. That then, of course, you need to check if the glue has been put well on the material. And this kind of processes can be automated withs ai and a visual way. But also picking, like object picking, if you have like a conveyor belt or all of product needs to be packaged or need to be quality control before packaging, then you you typically need to know the position and the exact post so you can pick it up with the robot and put it in a box, and that's great. You know, one common theme that seems to emerge from week to week in my conversations of manufacturing leaders who are coming from all different kinds of backgrounds in the industrial sector, is this concept of the manufacturing labor shortage, or this problem, I should say. And when you and I talked a few weeks back in preparation for this discussion, you said something along the following lines. I'm probably going to butcher your quote, but that is how I wrote it down. You said something like if you want to double revenue. Finding the Labor is the biggest barrier and I'm just kind of curious to hear you talk a little bit about what you mean by that and how you see the role of ai alleviating that pain. Well, many of the for instance, quality control issues, are very tedious tasks and they need to be done literally twenty four hours a...

...day. So if, at for in the night, you are a worker and you need to look at textile for hours in a row with big lamps shining in your face, it's just simply not going to work. So many of these people will get tired, they get it's difficult to keep focused. So at some point you will just waste material because it's not what humans are supposed to do, like looking at textile hours in a row. So if you don't have these profiles or if you cannot automate this process, then you will just just have more material waste. So you will have revenue. Yeah, you have more costs that. And also these profiles, especially in very industrialized countries like Belgium and Hollands, they're difficult to find. So you have this typical labor migration aspects. People coming from Eastern Europe and in our region. That's yeah, it's just very difficult to keep production running with labor shortage, especially in pandemic times when you have all of these regulations keeping distance. So it's it's a whole challenge to keep the factory running at the heat of the pandemic. and Are you finding from you know, talking to your own customers and other people in the manufacturing space, that the problem around Labor is that people just don't want to do those jobs? Exactly, exactly. Yeah, what you have is, I don't know if you have children, I've tree that one of the funniest movies they they like to watch. This is just like, I mean it's decades old, even more than a hundred years old. I suppose. It's Charlie Chaplin, and we all know this, this image of Charlie coming back from work and still having these kind of movement in his arms because he has been screwing all day, the same screw and and and our children just laugh with it and they say like Oh, that's just so funny. Was this like really happening or is this just a movie? No, it was really happening. And still there are a lot of jobs that are just very pinpointing one tusk. I respect all operators and all backgrounds, but there are plenty of other jobs to do. Then then just yea screwing one screw. That's just not a very humane thing to do. Forty hours a week. Yeah, that's that's an I love your Charlie, Charlie Chaplin example. There it's you know it, because it's it is true and it's and everything that I'm hearing is that, you know, the younger generation coming into manufacturing, that they just don't want to do these jobs. You know, the baby boomers are are exiting the workforce, the ones who were comfortable doing these jobs, maybe more than than the younger generations. And so who's going to do them? And the reality is there's a big opportunity in robotics to fill jobs that people, frankly, just don't want to do and as a result, maybe factors are kind of left, you know, stumped with what. How do we fill the fill the void? So makes sense. And that then also the aspect of more and more customization. Like...

...the Industrial Revolution has, of course come up with mass production, but at this moment mass production alone is not enough. We all want customized products and this customization is just hiring the manufacturing complexity and you can just educate the worker with a variety of four hundred different types of products at were just one unique type twenty years ago. So it's just much more easy to put these flows into republic camera vision streams. That's great. One takeaway I have here is I'm going to have to see if I can get my four year old and six year old watch or some Charlie Chaplin instead of just paw patrol. So we'll work on that. Jonathan, I've heard you talk a little bit about cloud versus edge technology as it relates to scaling your quality operations, and this is, admittedly, a knowledge gap for me. I don't know a whole lot about this and differentiating between cloud and edge technology and they're kind of buzz words to me and I'm wondering if you can enlighten me and maybe some of our listeners who maybe feel on the same thing, about sort of where this fits into the discussion that we're having here today. Well, it's very good question and some years ago cloud was seen contradictory to production management systems and quality control is systems, because everybody was responding along the same lines. Like what if you're into Internet connection just falls breaks down, and of course that is that can still be a possibility. But with five g with is lower latency networks, high bandwidth, with not only a fiber connected to your company but also, yeah, other kind of second choice or backup systems, cloud is really becoming of primordial importance in quality control because of the fact that you can just easily scale up. You can just have yet the scaling operations run by your cloud provider. In contradiction, if you have like an edge application, if you want to have have a double production or create a new production and you you you literally need to go and buy components, you need to have this this edge devices installed and calibrated. So the scalability is less with edge. But in some use cases, and we have industrial customers that work it air gapped environment, also because of production secrecy reasons and so on, then you of course need to resort to edge compute, and edge computer is basically an AI hardware device near to the production as site that is just having a very small loop image acquisition and fearence of the I model and operational results that will lead to a BLC like pushing out a product if it's below quality. So that that is basically the the reasoning between cloud versus edge. So cloud is your compute localized at a cloud provider, Azure, Microsoft or Google cloud...

...or Amazon web services, whereas your local edge is device such as a an Nvidia Jetson, like a small GPU, which is very near to the production where it happens, some cases even above the product, the conveyor belt, looking at the products and and giving a direct output to the PLC to steer some operation. So both have you use cases. Cloud is more scalable. It's also easier to update because basically it's just your glib provider updating to new gebeus or something. So the game is very depending on the use case. Joe, that's great. I appreciate you kind of given me some more insight into that topic. Well, Jonathan, you you and I sort of met via the industry four point no group that is sort of taking root on clubhouse right now. I keep talking about clubhouse, I feel like in every every episode of this podcast, because it's just becoming, you know, a really interesting place and I think the there's some pretty cool things happening in the manufacturing sector. There I was. I thought I'd just kind of asked you what what do you see in on that platform in terms of conversations that are happening around ai and robotics and just kind of curious how that's sort of influencing you and the potential you see there. I just love it because of the direct correct you can really make a connection with, yeah, some other AI entrepreneur or somebody responsible for production and go into these conversations. It's a really good tool to to learn about to world, how other people and other teams are solving problems in other continents, to sink and give each other advice. I think it's the next big thing in social media, but I'm biased because I'm kind of a bit addicted to Glob House right now. I can understand how it would happen. It's that I've participated in as an observer, largely in that particular group, the industry, for a point on, mostly just to sort of see what people are talking about and learn and enhance my understanding of some of these topics, and then I've been, you know, sort of running some of my own conversations more related to a marketing and sales and in the manufacturing space, but it's pretty interesting. There's a lot of smart people. You know, it's I don't want to say I'm surprised that the manufacturing sector is there, but I am a little surprised at how much it's taking root and so curious to see kind of where things go on that front. Yeah, and also to form new relationship, to talk to potential partners. It's just so much more easy than a gold email just to have a conversation, just talking formal and afterwards about it is very specific project, something for sure to actually hearing someone's voice rather than just reading words on a screen. Right. Yes, Great. Well, Jonathan, is there anything that I did not ask you today that you'd like to touch on while I've got you here? Not Really, I mean we didn't really a lot is changing,...

...also in the North America with Boston Dynamics and all of these new robotic frameworks, Berkeley based. So I do think that flexible manufacturing is really taking off if robots are much more easy to program and with if you combine the learning with advanced through publics. I see, yeah, a lot of difficult tasks like welding and my like molding and C and see manufacturing being replaced by smart robotics. Yeah, it's such an interesting space to be watching right now. Recently, on one of these episodes ahead Ryan Lillibridge from mission design and automation on here and he was he brought up the fact that robots have been around now for something like sixty years, which is which kind of I almost had asked, what do you say sixty, or do you mean twenty or thirty, because I you know it, you almost don't realize it, but some of the advancements that are happening right now are really kind of there's just so many cutting edge things. I mean you've talked quite a bit here today about, you know where vision fits into that, but you know, from robots as a service to being able to know the advance when it's and cobots and robots working alongside people in safe environments are becoming more or safe, whereas they, you know, robots were always traditionally caged and you know, I think it's just it's really interesting how fast things are changing. Absolutely and I yeah, I'm really looking forward to this future were robots are even becoming more smarter than they are today. Very passionate about technological advancements. Great. Well, Jonathan, this is a really great conversation. I appreciate you doing this today. You welcome to was a really fun to do. Great and can you tell our audience about where they can get in touch with you and also where they can learn or about robovision? Yeah, the best way to get in touch with me is via Linkedin, just Jonathan Bertie and you will find my profile, or to follow the robovision dolts a I linkedin page or just a website ruble vision dealt ai see what use cases we have, how easy it is to create new algorithms with our platform and how we can solve your next challenge in manufacturing. So it was great talking on this platform. I'm really looking forward and bringing to you some listens learned that we've learned in manufacturing halls here with Ai Republics in Europe and and open the American market for advanced AI technology. Great will, Johnathan. Once again, really appreciate you doing this today and as for the rest of you, I hope to catch you on the next episode of the Manufacturing Executive. You've been listening to the manufacturing executive podcast to ensure that you never miss an episode. Subscribe to the show in your favorite podcast player. If you'd like to learn more about industrial marketing and...

...sales strategy, you'll find an ever expanding collection of articles, videos, guides and tools specifically for bdb manufacturers at Gorilla Seventy sixcom learn thank you so much for listening. Until next time,.

In-Stream Audio Search

NEW

Search across all episodes within this podcast

Episodes (113)