About the talk
Lab to Live: Superhuman vision meets superhuman strength: Applying AI visual processing to industrial kitting – Cogniac AI and Bobcat
Manufacturing requires precision and speed at scale. Bobcat has to deliver equipment with specific kitting using pieces fitted exactly right, every time. We delve into how Cogniac’s innovative AI platform keeps watch and catches errors in real time at Bobcat’s warehouse to streamline the assembly line and reduce financial and human risk.
Chuck Myers - CEO - Cogniac.AI
Joel Honeyman - Vice President of Global Innovation - Doosan Bobcat North America
Amy Wang - Co-Founder & VP Systems - Cogniac.AI (Moderator)
With 20+ years of experience as a Chief Executive Officer, I have been integral to operational processes concerning profitable investments, as well as acquisitions for startups & multi-billion-dollar organizations. It is my passion to consistently drive growth through global leadership & corporate restructuring, allowing for seamless M&A processes. As a leader, I have managed global expansion & financial management initiatives across my diverse career. I am consistently reaching revenue increases through strategic portfolio management, which has been evident through recognition as well.View the profile
I'm a truck Meijer on the CEO of cognac. Cognac is the visual AI platform that we try to fix the conference? Quite. Well. We really describe it as solving superhuman vision problems. I'm joined with the Joel Honeyman. Who's the VP of innovation at doosan Bobcat? A multibillion-dollar, global conglomerate? And dr. Amy Wang, who's one of the founders of Cognito acusa? A PhD and electrical engineering. So why don't we start with this Amy? Why don't you go ahead and introduce yourself
to everyone. I met me one and I grew up in China when to 200 diversity, right after college. I came to the states and got my PhD from Columbia University. I started my career in the life. I industry. One of the most exciting thing I did was that Ruckus Wireless. High performance wi-fi system that is powered by an Innovative machine. Learning solution is probably one of the first successful applications for machine learning in the real world. And a truck. Has I met my co-founder? Bill Tish.
We shared a passion for training complex Technologies into simple. You supposed to Lucien. So after Ruckus, when IPO building, I started copying with a vision to make a, i usable and accessible to more businesses and I'm the workaholic and I'm I only Hobbies my seven-year-old daughter who could be barging in a minute and I told her not to do so and I'm very honored to be here at contact stage and I'm super excited to talk about how magic happens when superhuman vision and he's superhuman strength.
Scripture, Joel the stage is yours. Hey, good afternoon everyone. I'm Joel Honeyman. I had a global Innovation here for doosan Bobcat based in Fargo. North Dakota Bobcat. We are the global leader and compact construction equipment. And we rebuild the tens of thousands of machines. Every year for our customers to get their jobs, done better and more productively, and that includes loaders, and excavators, and all kinds of all kinds of different schools that go along with it and bobcats. We obviously, we manufacture this equipment's and we're a
large, global company, and me and my team and Innovation, we seek out all kinds of different Technology Solutions that help us to work better and faster in that also includes in our plants. And so, that's why we're here today, and I'm proud to join the time. He act on the stage to talk about the work. The cognac is done with Bobcat to, to help us. Products better. So thanks Jewel. Really appreciate. You both being here and taking the time. Now that we met our subject matter expert. Let's just dive into why we're here in our Enterprise, AI Enterprise applications and
how we really solve real-world problems. So in a cockney accent, mission is to provide Enterprise Solutions using Innovative AI platform. So no code solution. Our recent partnership with doosan has demonstrated really via the efficacy of implementation in the Enterprise use of machine learning. Think it's a pretty novel approach, especially from the kid in application, which is why were thrilled to be joining with Joel today. So I'd like to start with an overview of industrial kidding. In the complexities surrounding these. Joel, you know, if you
could describe what is industrial kidding and why do these complexities matter and kind of what you were trying to sell? Yeah, I think Chuck, you know industrial kidding is kind of the classic manufacturers dilemma issue. We're going to call it but it's part of a process that we build complex machines and those machines for car parts. And so you have to put the parts together in a certain in a certain fashion and have the right parts to be able to complete the product at the end and Automotive all kinds of Manufacturers. I like us you similar processes and more specifically
we broke down this kidding process from from where it was previously which was we would deliver these parts line side at our main plants when we did the assembly, which got to be very complex with all the options and configurations. So now we move that getting off site into a different facility and we, we do the chitting specifically for each individual machine. So each serial-numbered machine has a unique it now with gets holes in it, that are required to complete it and so we can create this kit when you Put the parts in the kid and I'll get in the stash here in a little bit. But it's a,
it's a incredible process. And then we deliver that exact kid at the right time with the right part to be able to unable us to build our thoratec time of our machines and Equipment every single day, back for me, know. What's what what's riskier of the kidding thing? If the parts are wrong in the kit. So, you know, what is in prior to having the cardiac solution. We were depending on on humans to pick the parts and to trust that they had the right parts in any kids, and just to give people
a little bit of context year. There's actually about 30 individuals, small, crates that come together to create one kit for a machine. And so if you want to play on all the parts, every single day were picking between 40 and 45000 Parts an individual day to be able to build our machines. And if you just think about it in terms of math mathematics, it's very hard for any one person or human being, no matter how. Well, they're trained to be able to pick every single part. Right? The first time, in many of these parts are small. Some of them look, similar to others. And so we had all kinds of
errors. And initially before cognac, was put into place are our success rate was about one out of three. Only one out of three kids was delivered accurately. And so in that case, what we did is we created line side in Ventura. And our factories to make up for the fact that two-thirds of the kids were delivered incorrectly, and then, it wasn't on any individuals that espouses. The parts is very hard for you, do for human interaction, to be able to do that in a very accurate scale. So it created a very inefficient process for us in our manufacturing. So that
leads to the next question, why not? Just throw more people at it. We throw a lot of people at it to do that. That's the problem. And it again you have new people in in, in training and educate, people are our employees and employers, we use for a third party that are doing this process, do a great job. And and so it's not on them. It's just the volume of it. So just to give everybody again. Some of the staff here is every 24 seconds. One of these drawers comes by to get filled with parts every 24 seconds. So I'm to be able to say that, you know in in human beings to be able to pick it,
right? But then you're going to inspect every one of those drawers every 24 seconds. It's physically impossible to be able to have to bring accuracy was just with, just human interaction. At that point. You know, what's in a traditional Machine Vision manufacturing system. Yeah, well, we we like the idea of using a high because you know, the the parts may change the kids may change how the parts are laid out. And so we needed something that learned that it learned the randomness and the different nuances that took place again,
thousand departs going across every day and they might be playing differently. They might be his position differently. And so we had to be able to account for that and you couldn't go back and have machine learning. Continue to your rigs, am in these models on their own. It had to be dynamic and real-time. Otherwise, it's just really want to provide value for us. So that's why we were. We like using an AI solution to be able to enable this over, you know, it a traditional kind
of vision system. So with that and you're talking about this really this after the efficiency in the Precision their required. So, you know, you choo, choo. The cardiac system. So Amy, maybe you can go into a little bit how the system works and and why it's all stools problems. Great. Thanks. Joe has very good understanding of how a eye works and then West advantage of the eyes. And so from a high-level cognac is a NOCO, the eye solution in Naples business like you some bobcat to automate their visual inspection in the best way to explain how cocky actually
work is Rizzo live demo. But before I go to the demo, just to recap the stage, as Joe mentioned, they have this Warehouse that has many of the parts that used to build his bobcat in the machine. And then, you know, where house, they have this conveyor about that. Doors go on top of it and people would have dropped the parts in those drawers and we had installed a camera and of this kidding lines. And the camera is plugged into watching a cast device. It's a super powerful computer with many people using it. We called echezeaux says the doors coming down this line and rs
flow to take an image of the Now let's talk. I let me share my screen and then go alive you. Very quickly what you think's best cognac apart from any other Vision AI system can go talk about that. What are you looking here is Bobcats tannin in Cockney, access stub. Let's take a look one of those kidding lines. Brian Wayne camera. And here are the images and you can see the images are coming to our system live and the most recent images are our few half an hour ago and
noticed that there are the contents of this drawers, very a lot and is Joe manchin, there are a lot of those different cards to make up this kid. So it's impossible for humans to to see which ones, which end to make sure the right kid is right partner in this kit should before, cognac what they had is, they would have a human operator, would be challenging, how many parts are in the drawer, to see whether expectacion. But if I drum Parts in this drawer, there's no way for them to know, it's to know about that.
Now, he's Cockney existence that we have trained, AI models to recognize all these parts. Now, let's take a look upon the image so that they are tomatoes processing, his images, and then we are predicting and these are two parts in his drawer. You might have noticed that some of this a parts are very similar and this is you see the longer version year and this is the shorter version. It's not hard to imagine that this will be very difficult for human eyes to differentiate this part. That still mention to the similar parts, are hard too hard to tell
by human and that is exactly the kind of thing that AI is good at the train animados can predict this repeatedly and consistently. Now back to the word Clouseau RH float device will process these images and then says that this model predicts you results to Bobcats business software. If there is a mismatch in his content, The Operators be informed immediately so they can fix the sticks, the kid. Spot. They won't shift around kid to the customers. And now, this is just a short overview off of how the cardiac system works in the
kidding application line, at sat at the Bobcat now. Let me stop sharing. Yeah, quick question. You know one of the one of the issues with artificial intelligence is always Benny has been model release a 9. O clock me. A kiss a pretty unique solution for releasing models. How many, you know, how many models AI models do you think you've trained in these kidnap location to love demo process? Is there straight for? But there is a magic, there's a copy of magic is that our
system is so intuitive and easy-to-use. That their warehouse operators can automate their inspections on their all. And they don't need to hire data scientist and they are the one actually training models on their own. We call this program with visual data. And so we are your system has completely automated many of those decisions that data science has to make a. Last time I checked, we have trained close to 1 million. Tomatoes for the kitty application alone and that's yeah. If I can't imagine if you need data science to do that. It was how many human
hours he needs to do it? She's at, but we don't stop there and we keep training models, the cast of things change and our motto is continuous improving. As we speak here, our systems working to improve the Bobcat applications. Great. So it really good job. Describing the system and thanks for showing the demo. So maybe you kind of just started to lead into that question, you know, just kind of to wrap up that conversation. Can you talk about what makes cognac more affected? You started to leave there for it's kind of an AI for a solution
to create set. Maybe if you could briefly kind of just give us, you know, what kind of implementation do you have in production today? Well, for one, you know, our assistants being used by many at our system BoRics in the real world and we have been in production with many of the Fortune 500 companies examples, for example, like BNSF railroad has been using coffee eggs for mission-critical applications. Like we all decide to Texans in your time and tracking sections. We be monitoring 22-minute wheels and thirty thousand miles track every single month and we have prevented many
derailments in the manufacturing industry menu for the big, automakers be using talking text for stamping. My inspections and we have installed like 30 high-resolution cameras to inspect millimeter size shoe size. In the next ten feet by 7 feet car. Door panel. Just think of this cost me in as trade. AI system at so many more eyeballs and human. Does it can look everywhere. It doesn't miss anything so it can operate as a level much higher than human do. That's why we called superhuman Mission. Nobody else is operating at this scale. That
is probably the largest visual inspection set up for a single part in the manufacturing industry and finally 442 Bobcat, kidding applications. We have trained the largest number of custom II models for single-use case, industrial, AI space. And we're so thrilled to be leading the way in solving all this difficult, real-world problems. And how do we do that? That's what you want. A Sprite makes a cockney. Accents them. I mean, you started in on, but can you summarize in a few words? What makes it the right fit for, for doosan? From your perspective? The right fit for
a doosan Bobcat Snead's and how that would compare to any of the prior operations. Well, we were looking for a solution to be able to improve your accuracy in this part of the Opera. Patience. We did a pilot with cognac R&D facility in in Bismarck North Dakota. And that pilot went. Well. I think the thing that that sold us was, you know, how easy it was to train, real talk about fifty or a hundred images that we provided and 90% very quickly. And then to Amy's point, the model continues to grow and
be able to still learn and and get better over time, which is another aspect of it a week. We just want to manufacture things like so we we don't want to deal as a me sad with data scientists and reprogramming. All kinds of things. We want to build equipment and do it as best as we can. And so Heaven, cognac, do that behind the scenes. However, it's done, you know, what was a great solution for us to be able to use and then to implement What led you to this change? Was it that got me after the
great marketing job selling you, or there's probably some history where you went out looking for, for a solution of your technology guide, your job and a bobcat is to lead Innovation. So not what what let you down this path. Well, we took a look at a number of vision inspection systems. We were looking for different types of Technologies for a manufacturing processes and just seeing what fit and we talked to some other companies. And we we met cognac through our relationship with plug-in play Silicon Valley, and got introduced to, to the cognac team there. And then, like I said, we do a
lot of Pilots, that's what me and my team does. And sore, like Kayla's pilot this and the pilot came along really quickly and our operations people who are Skeptics at heart. Just because if they have tough jobs said, hey, we see if we can make this work and they, they adopted it. In our in our facility outside Minneapolis Minnesota where we do that. I'll do this kidding process as good. A kid in basically palletized and shipped to a whole another state to be assembled. So if there's something wrong, it's a big problem. And just think I guess just to give you the stats today, cognac
is catching between 12 to 15, Miss picks in any shit, anyone shipped. And if you think about that, at 1, kids shows up Blind Side with one wrong part, that's one less machine week. We can build in a day or it's a delay or it's the only causes all kinds of different issues. So, to be able to catch stats. And so, our accuracy actually went from one and three to. Now, we're one and $20,000 is our air rate in our killing process. Enabling the cognac solution, along with our change management at are at our facility to be able to enable this.
What happens if you? You miss a part. I mean, obviously you still going to ship the tractor, but you know, that delay obviously has a cost. If you ever tried to monetize, I tried to calculate that monetization restaurant. Yeah, we haven't specifically cuz it's really hard to isolate. I guess, as long as we're not getting calls from your CFO saying we're missing our bill. That's the right metric for all of us in turn, leads to make sure that we're not where it were hitting, aren't you know, hitting her text. I'm in our numbers.
How would you describe that implementation process? I noticed, I've know you do a lots of companies. I've been in, you know, technology is great, the real world implementation and kind of get the getting that reorientation process with folks, on the factory floor is always as much problem frankly, as the technology itself. You know, what's that process been like, for you? Well if it takes by in so we got really good executive. Ian former VP of sourcing Mike Wood and his
team wear. This all reports up to Ian. So Mike Jorgensen who runs the facility in in Minneapolis for us. You know, he really he really took this on after the pilot. We showed the value of it and he said, hey, I think we can make this work and saw it rolled out in just one area of our plant. Where we we we actually we inspect incoming parts and make sure our supplier gave us the right parts. Cuz obviously, when you have thousands of parts that could be an issue to is just getting the right part from our supplier. So we actually do this unpack and repack process and cognac was
first put into place in that process as a way to test. Just are we seeing every part? How accurate are we is the model getting better? And so that was a way for us to start it and then we put it line side in The Killing process. That you, you kind of engage some of your actual vendors to include some of this, before those parts. Actually get shipped to you soon. So that there's not a, there's never a debate of whether they actually delivered, what they were supposed to deliver. Yeah, we sure could but we certainly love getting those images.
United States. As you guys describe, we get an image of of everything that comes across stand and we've used that numerous times, whether it's with our supplier, our own internal operations people to say, no, this is what the kid had in it. It had the right item is in it until we've used that or we went back to our supplier, to say here's the damage. These are the wrong Parts in the wrong box and it needs to be fixed. That's fantastic. Success in your partnership with cognee act.
Look like for free for doosan Bobcat in that goes down the road here. Well, since we have this first instance, you know, I already put in, you know, into our system. It would be nice to expand it to some other applications that can use their manufacturing process. One. We're going to look at piloting is our end of line, inspections of finished product. Just make sure all the right things or are on the equipment that you couldn't has different kinds of options. Different kinds of decal with messaging, that's required depending on the country. The stage that might be in a so
ensuring that we are delivering the product 100%. Accurate has obviously, critically important. One area will look into to run a pilot on to put the system out their Instagram from from the vendor side. You know, what what do you think that long-term relationship look like looks like really from a technical technical approach with would not just Just doosan, Bobcat fool with, you know, your other customers. Text Joseph is actually a true story that might
Jurgensen that the supply to manage it as a Bismarck. And he told us was the one up by his Downstream customer. I told him that a bishop, the wrong parts to him. And then he went look up in Connie has to stand to see that. You are an auto parts in there and he called the guy out saying, hey, I have a photo of that drawer. I sent you all the parts that you there. And here it is. The best of, you know, one of the major differentiator what we have is this manage the database. If you are a business leader, you're using cognex. This and then you will have 24/7 access what's
happening your business at your fingertips, but if you don't have cardiac system you if you want information, your people might have to go down the line to check whether they saved images. As a matter of fact that in many companies in the industrial World, they never saved images and the inspections are secured a bat and they toss away the images and even saved images with just like a bunch of other men while that's like Kyle chunks and most of those are not useful but cocky, I will give you like a
beautifully all file folders that you can look for what you're looking for. Very, very easily. And I think it's very important for visual information to a visual animated data to be managed. We can use it later. That's the key. Differentiator, as companies going into this computer vision age, when you look at it, this, the overall future on me. Where do you see? Forget right now, but, where are we five years from now, you know, what's the What's the big paradigm shift? That's going to happen,
you know, in this nation area to where Vision AI really makes epically. Direct me to pay. So I I think we are still in early adopter stage of Israel and many of our customers like Bob do some bobcat. There are The Visionaries in their Industries and they clearly see the value of this technology has their expanding, their use case and Technology adoption curves, but given the nature of this AI technology, where the same algorithm actually work to solve many
business problems, like line side, effect detection and retire at all. These things makes that exponential growth part of the S. Curve, is very steep. It'll be feted by customers like you some bobcat, how fast you can, she laid all those images and utilize them. Apps. You can be more like f a l. L e d shaped as you'll start slow because the complexity of AI, technology makes adoption little bit harder, but after that will be vertical line across the tremendous batteries provide. Amy, that that was great description and really appreciate it. So
Joel, I think we're kind of running to do a bit of the Q&A here, and I have some questions that have popped up that have been somewhat directed to you. Why don't we talk about this one? It's a partnership. Can you connect? Talk us through the pilot and Discovery process, you know, what are some of the data that you have been wanting this process and you know why you're on this journey and and then there's a follow-up question on that. I don't want to give you too many once, but how can you use social media in data to build that trust with your
customers along a long line? Well, as far as the process, you know, it worked really well, you know, we were introduced to find you act. Like I said throughout through plug-and-play and it we put together a brief profile and it was less than 60 days and real to do that. At one of our main facilities provided and system, and see if it work. And then it was a matter up. Again, up selling meal, back to the right email executive team to be able to say there's valuable position in here
to do that. But, I'm a big believer in Tyler. That's the only one thing I would mention this. A lot of these systems have to be able to integrate in with Legacy Manufacturing Systems. And that was one thing that cognac is able to do Good to be able to have that data feeds come into our own systems, which is no tall task for easy. Task for anyone listening that's dealt with an integration of a system. Broadly is AI being used its Bobcat or maybe in the in the bigger parent
doosan today. Are you don't we have similar projects in other areas? I think we're just starting. I think Amy is correct. I think AI. True AI when we were all true. AI is this is just starting to happen. I think there's some things that might be called a eye that maybe aren't out there if we want that opinion, but I think for big companies like ours, you got to start with, some seems like very simple tasks are problems, but they're really not. These are great problems to solve. They don't sound very special but they really are to us. And I think that's where it's going to. Start,
is just starting some core processes, that a company and and work your way up and and not try to solve everything at once. I think you can do any people and companies try to solve everything or solve the hardest problem. First, and I don't think that's the best way to approach it to my team every day. Stop picking the hardest one. It's a little bit. Not the topic, but there's a lot of people that are watching this that are interested in entering and of a i and in data analytics today is there is a career as a career move
and you've been doing this for a while and obviously have quite the academic background with it. What, what advice can you give those folks? Work work work on solving real-world problems. Not trying to develop another Akron. Sister like 1% better make you love the state. Identifying star r, r a i better look at the complex as we were so visual. And that's that's that's how you applied, make a are accessible, and then they kind usable. And I think if you work in that area and if they do that kind of thing, so if you make your life so much more fulfilling his way. I can change it to work
and you want to be part of that. B Tech questions to a me to with with a no Code system, you know, it quickly. What does onboarding look like for the company from your perspective? So we are at very a holistic GI Solution Center around visual data and what continent is very intuitive and responsive to changes because we monitor changes. We surfaced jeans to users all centered around the visual data to the boarding process. You will customer with connector image source, in this case with a camera into Cockney,
accents them and we have a very intuitive user interface with guy to use her step by step of the create applications incognita system and label a few images. And after that are AI, creating a engine takes over an automatic trans models. Monitors changes Izzy. Most human intelligence me when you did that, how the whole experience works with And that's why I caught me and works in real world because we adopt we monitored adapt to changes for a little my influence in here
because I get asked this question all the time, you know, there's always say you say, I by nature is a probabilistic solution, you know, it always approaches to 100% and in some cases, and jewels case there are things are exactly a hundred percent. But a lot of times you really just trying to move the needle in a manufacturing space, where a human may be 75% accurate, you're trying to move that up into the high nineties. That makes a big difference. So, where do you think if, if AI is approaching that 99% and me, where do you think the next step is? Where do you think that that goes?
Do you make a system that makes many? Many of the applications real-world get to 99%. Now some of that you're not in the manufacturing warm and that their production line. You can control us of things. We can probably get you 99% very easily but liking Bobcat situations where house is real world thing and her lots of the work and make assistant that be able to control. Those changes, are our work with those changes to get those application. Those are like much larger percent of the world problems to get 99%. I think I'll be very good space to work on to make
a. I more usable to the world will be many other applications to be to to, to work on too. But I think we should focus on making more applications. Get you 99%. I have the honor of somebody asking me a question. What is the future of cognac? Data, and I like to describe we, we we like to believe the cognac really provides a system of record and I think Joel's description of being able to go back and even validate kits, is a pretty important is a pretty
important phenomenon and you really look at it from a lot of our customers. How can I go back? Whether I'm in automotive company and I want to validate that I did. A bit of a project, is my fenders are supplying the right part. How can I go back and find all the metadata? We see that in the real world, where everything is very safety-critical and gets reported to the federal government. We need to know when those imps incident, occurs in a really, how all of this information was gathered in the first place, so you can go back and make judgments fit to change the
directions. I think his AI gets better and better. We really do. Cognac is becoming a system of record as much as you would think. About work or sap is your ground troops system of record of your structured data. We think that called me, I could be in the ground crew system of record for your visual data. So, I hope that that answers the question, why it's nice to go down to that. Sure. It has the most interesting to me about AI in supply chain and Logistics. Is this, the power of networking site, right? As check was saying, a system of record, but now imagine probably down
the road that's company standardized to technology to inspect Parts, like using coffee. I know, you said we expecting many times that each things change hands. Now, if you can imagine to attach a digital manifest, which is a visual information on items and get an updated throughout his life. And that's all that information can be used to, optimize the whole supply chain and that would be the power off this system of record. And I'm very looking forward to be part of that at the drought. What kind of solution for the whole industry? That was a great, a great at, appreciate it. And
enjoy your our Innovation guy? I'm going to let you kind of finish up here. We're getting some sometime wrap-up time here now. So I would turn it over to you. As you're the guy that sees Innovation across lots of different platforms. Not just a I so I'll turn it over to you, dude. Kind of wrap things up. Well, working in the spaces, lot of fun. We working a lot of different kinds of Technologies, whether its autonomy, or Digital Science, and or automation or in this case your visual inspection visual records, as you said, Chuck. And and it's it's an evolving space. But
I think what we're seeing is is there are these our real-world applications of this technology out there and cognac. His is obviously a great example of that because we've implemented at bobcat and so we see we see others out there and some Different areas, but also I seeing this kind of technology evolve. And like I said, in other applications within our business, it would be a great next step and anything that helps make our our production easier, but ultimately, makes our customers were satisfied. Give them a better product. That's what we're here for is to
give the the best Bottom Line Products or customers. And if I can help us deliver that, then that's the solution that were going to put into place. Appreciate you guys taking the time for this. I personally obviously have a vested interest. I love the cognac story and appreciate all the time and effort and I hope it was super educational for the audience out there.
Buy this talk
Buy this video
Our other topics
With ConferenceCast.tv, you get access to our library of the world's best conference talks.