#83 Talking about Deep-learning vision, automated picking and placing w/ Herbert ten Have (Fizyr)

Welcome back to another episode of the Process and Automation Podcast with the Automation Guys. Yeah, today we have another interview episode. Interview episodes are very popular with our community, and today I have the pleasure to welcome Herbert ten Have from Visia, and today we will be talking about deep learning and automation. And yeah, welcome Herbert, welcome to the show.

Thanks for having me, it's my pleasure.

Perfect. Maybe we just dive straight in. Tell us a little bit about what you're doing as a founder, or have done as a founder at Visia. And yeah, what you also do otherwise.

Okay, Fazir started by a professor of the Delft University as a spin-off, and they were building robot cells, and that was already 9 years ago. And at some point they participated also in the Amazon Picking Challenge, and it was in Germany at that time. And then to their surprise, they won both challenges, picking and stowing. So they knew they were some kind of special. However, they didn't find the business model yet. And so I got in contact and they asked me and I joined 6.5 years ago as an investor and also as a CEO. And then within a few months we pivoted the company. We stopped doing robotics because we believe everybody can do robotics, like driving, like making cars. It's just from A to B and just a lot of mechanics and also support installations and a lot of liabilities you have if you do hardware. So we wanted to be computer vision, and so we are a computer vision company that made one product for picking and stacking placement in logistics. So it's from item picking, could be apparel, clothing, so groceries, And when it's packaged into a box, it becomes a partial or a bag. So we do partial induction, depalletization of boxes and bags, as well as the newest since one half year, trailer, loose load trailer unloading. So that's in a nutshell what we do.

Yeah, that's really interesting. So yeah, usually we are speaking mainly with sort of software automation, software robotics. And it's really nice to have you on today to talk about how all these computer vision technology we might have maybe used in other sort of digital processes, how they actually are used in the real world. Really exciting. So yeah, between Arno and myself, we have over sort of 20 years of experience in that space, and we're talking with lots of companies like yours. Yeah, maybe more in the sort of the digital area and workflow area. Where do you see the competitive edge for Vizier in the market?

Good question. We've been looking for that, of course, as well, because there is a lot of competition. So most of our competition is from the US. There's one or two from China and Japan, but the rest is all from the US, heavily funded. And so it was not easy for us to find a spot that we could really leverage the uniqueness where we are. And so we've done some time on that. So it's— I can often compare it to self-driving cars. So everybody can go, can just make a self-driving system for highway, something simple. But ever when you turn off the highway, it gets difficult. The same with logistics. So if you want to pick up just a simple box or just a simple item, everybody can do that. So sometimes you don't even— you can program it heuristically. You don't need a neural network to find the object. However, in logistics, it's take a parcel, DHL, UPS, Federal Express, etc. You cannot program how a parcel looks like. If you drop, if you take a bag and you drop a bag 1,000 times and you take 1,000 different size images, then the robot or the computer vision will see 1,000 different bags. And that's, that's, so you need to train a neural network on that. So, and where we distinguish from others is where there's sometimes the boxes are very closely stacked, so it looks like one big box, or due to tape or folding, it looks like 4 boxes, for instance, or 2 boxes. Or when there are reflective, then if you put a light on it, it looks into a mirror, you get a lot of light back. Or when it's black absorbing, or it's when it's the same color on top of each other, close to each other, black on black, white on white, or even transparent like clothing. There are in polybags, transparent polybags, So the closing might be in one part of the polybag, but the polybag is bigger. So in order for a robot to go in and to apply suction, so to simulate it, to pick it up, you need to be very sure and robust in your computer vision to know where to apply suction with the suction cups. Before you know it, you pick multiple items, or You pick a part of the item and then it drops off. So that's where we are special. That's where we believe we've done something unique and it's been, we get it back from our clients. They are telling us this is something where you were absolutely excelling.

Wow, that sounds amazing. That's amazing. Yeah, we always a very popular sort of I guess, theme of talking and talking points in the show is really to focus on, you know, the application of the technology and use cases for it. And I mean, you've explained that pretty well in that previous sort of segment. But just from another question from my side. So how would a kind of typical client requirement look like if they come to you? I know you've sort of just describe a bit. But if a client comes to you, you know, what is the sort of requirement, the use case, and then kind of the solution of your technology, you know, how does it fulfill that?

Okay, so we do a lot in that space, from item picking up to trailer unloading. Depalletization, mixed SKU or rainbow depalletization, is quite strong. So we do a lot of that. What clients tell us, they say, we have a harder problem, so it's, it's our case is harder, and we just tell them, okay, just take images of it. They can take it with a cell phone, they send us the images, and we run them in our network, and then we will provide them with the information out of our network. So completely untrained, we just get their images, we run them in our network, and we provide them back with what our neural network has seen. And that's very convincing. So that's mostly the first step. Then the second step is, so we always work with integrators, not with end users. So robotic integrators and system integrators, and they are using our software to produce a product picking cells. So the same as car industry, if we would be in a self-driving car, we would license out our software to BMW, Audi, Mercedes, Volkswagen, Porsche, and Volvo, etc. And in this case, we license out our software to a system integrator or to robotic integrator, and they build picking cells with it, mostly a lot of numbers. And so we help them installing our software, we help them optimizing, configuring and optimizing the picking cell, and at some point it's being installed at the client, and then we support them, of course. So we license out our software, that's the model.

Yeah, very interesting.

More automation entering our lives. I think every day nowadays we talk about automation, isn't it? So we talk about ChatGPT, we talk about this here on the podcast since it started to become very popular. We talk about intelligent automation, hyperautomation. And in your specific area, where do you see things developing for your customers, for the logistics companies? And where does Vizier fit in?

Okay, I often show a video to clients, it's online, it talks about, shows where in the world everybody is, if you order something online, is still being picked by humans. So the number of robotic shelves is still very, very limited. And so it's all humans picking your order, picking the items, and at some point putting them in a box or bag, and then also inducting them on the sorter. So a lot of steps where humans pick the items, put it in a box, becomes a parcel, and a parcel is also normally being handled by humans between 8 and 16 times. So a lot of manual labor, and that again could not be automated yet because of the variation of the items. So I see the future as where now, where some of the fulfillment centers can be automated, maybe not fully, but let's say to a big extent. And then a fulfillment center, instead of having a lot of humans, operators, order pickers, you would have a lot of robots and then maybe some people to control them. Compared with a data warehouse, a data center, there's always some people working to maintain some small adjustments and stuff. I think at some point when you control the goods stream You can automate a lot with robots and you will have some operators supervising it and dealing with the exceptions, doing some maintenance, repair and stuff. So it will not be completely lights-out factory, but we're getting the technologies there to get to make a huge jump.

So what do you think then in terms of the sort of currently we're doing this automated picking the palletization, that kind of stuff, percentages, where are we in comparison to sort of the human element?

I would say the same as self-driving cars.

Okay.

Early days, very early days. If you just take the number of parcels being shipped worldwide just per second, thousands of parcels, and just take the human labor to that, it will stay— there are not enough robots to do this. There are not engineers to build it, etc. There are not enough cameras to do this, so it will take quite some time. But again, technology is there, so it will happen. It's more a matter of capable companies to install them, as well as engineers and cameras and tools to have them. And sometimes the business case is not quite there yet, so maybe there just needs a bit higher wages or more scarcity of people And then at some point the companies don't want to take the risk of Corona anymore. They said, I want to automate this because if we cannot fulfill our orders, we're out of business. So we need to invest in robots.

Yeah, that makes sense. Yeah.

Yeah. And I guess from our side as well, and we discussed this a lot on the podcast, is generally in automation is where do people get started? Because, you know, we deal with a lot of companies and there's always a starting point for these things. And it does feel sometimes there's a lot of convincing for people to say, well, you know, this automation thing is really cool. You know, if you don't do this now, it's going to be really a difficult future for you in 3 to 5 years' time. So I guess you probably get the same on your side as well when we speak to prospective customers. So what is your sort of strategy to get customers to embrace this technology and to really sort of open their minds to, you know, to tell them and invite them into this exciting world of what this technology can do for them?

Yeah, we work layered because we don't normally work direct except from a few of the biggest players in the world. Normally we work through integrators, so we help them build the product and they sell it to the end user. But of course we speak a lot with end users on exhibitions and shows. They often had some experience so far and which was not always good. So that's a bit of history which is also beneficial for us because then we can show why we are unique and why we can solve this kind of issues. But logistics is never an ideal world. So there's so much take parcels Anything can be a parcel. And we often take the joke like our grandma can wrap a guitar into newspaper and send it as a parcel. Yeah, how do you want a robot to pick this up? So yeah, it's funny, of course, but it's— there is always a long tail. And so it's part of the solution is not trying to automate everything like 100%, but assume at some point you don't want to pick it up. Let's say you see something leaking. You see fluid leaking, you don't want suction to pick it up because it will create a bigger mess. So at some point you need to know, okay, let's divert this partial and let the human take care of it. Or so it's, it's, uh, it's unique. You often look, have to look at it holistically, uh, so not trying to just replicate the human one by one by with a robot, but make looking more at the whole process, at the business case, and how in the mid to long term, you can optimize the use of robots and not optimize the use of humans anymore.

Yeah, I mean, we do this the same in software robotics as well, where we don't want to automate, and to be honest, it's impossible to automate everything 100% of the time. What we do want to do is we want to find a sort of middle ground where, let's say, 60% automation or even 70% in some cases and then 30% where the human is still in the loop. And it's the same principle whereby if it's something mundane that we can train a bot to do, we do that. If it's something a bit more complicated, you've used the example of a perhaps a package that's been damaged and, you know, factory warehouse picker needs to deal with that. That then gets taken out effectively out of the robot queue and can be dealt with by exception. So it's exciting to see that the two worlds are actually from a, you know, principally the types of things that we want to solve is very similar. I can imagine that the types of techniques that we use to solve those design principles is probably going to be very similar. So it's really, really interesting topic. And I can imagine, like you said, You know, with the online shopping growing year by year, I know for a fact there's Amazon parcels arriving at my house every day. And there's, you know, it's, you never know what happens behind the scenes there. And, you know, to my surprise, I didn't know, you know, I thought it was a lot more automated in these warehouses. So it's very, very insightful.

And even that is super low, isn't it? If you see the whole potential of online shopping or of all our shopping, so that the number of that is still super low, isn't it? In comparison, everyone thinks we are doing a lot, but in comparison, we're doing tiny figures. And if that goes or doubles, just doubles, I think, wow, there's definitely no way around your technology.

You have, if you enter a warehouse or a parcel distribution center, you have no idea how many people are working there. And not even just during the day, it's in the evening or the night or early in the morning. And just take a trailer unloading, loose load trailer unloading. Mostly, most trailers are just a thin plastic on the side or rubber or something like that. And just imagine the people are sitting on the ground picking the parcels. Middle in the summer or middle in the winter. If you stand there and so I often take pictures, of course not taking of humans on there, but it's nobody wants to do the work. Everybody wants to do the picking the parcels in the big hall inside, not outside where it's going to be very cold or very warm.

Yeah, yeah, no, you can understand and empathize with people like that. And just, just on the topic of, you know, kind of partial picking and this year. So is this it? I mean, I can imagine you guys have got ambitious ambitions for the future. Is there a sort of a next step where you can take this technology to? Because I know we talked a lot about partial and partial picking. And but is there a sort of a a future state of this where this can be installed in sort of robots. So it's not just sort of picking orders, but actually, you know, putting it in boxes and loading the trucks. I'd say, what's your sort of view on that?

Yeah, actually, that's already in the product. So we— so the product exists of two parts. One is picking, so looking into an unknown environment and finding out where are the parcels. So segmentation, classification, grasp pose sequence, all that stuff. That's one part of it. And once it's attached to the gripper, then we can see the dimensions of the object, so XYZ. And then we can also look into the new environment where we have to place it, and then we can see where to place it in an optimum way. And we can even take an extra image after it to verify if everything went as expected.

Amazing. Absolutely amazing.

Perfect. Yeah, really cool. And so yeah, this is very technical as a topic, very exciting. Very often our listeners like to know the interview guests a bit more from, from the non-technical or non-business side. So that's why we like to ask you a couple of quick questions which are a bit different. Are you okay?

Yeah, I feel, um, trouble. Let's start, let's start.

All right, so, um, the first one is: who is your idol and why?

Okay, I'm gonna give you a name you will probably love, but, um, I think Bill Gates is an idol for me. And, uh, not for the fact, of course, what he did at Microsoft, but what he did with the Bill and Melinda Foundation. So it's based on effective altruism. So how we can improve the world. And the greatest thing with malaria, it's so the medicine, the way to solve malaria was I think 20 years ago was already sought out by science, but they took it to the market. They said, let's, if we spend $200 million, where can we have the biggest effect? So I think they're great examples for many people following there in their footsteps. How to, to be of value to the real total world.

Yeah, I guess that's, that's quite humanistic, isn't it? Where you can impact human beings' lives in a very practical way. And I think that's, that's a very good measurement or attribute of a human. We get all sorts of answers, but yeah, very, very, very, very interesting. We've never had Bill Gates before.

Yeah, this is exactly what Arno and myself, and probably you, I can imagine, because you're building these great things similar. So we, we use technology, we want to change the world, we want to make the world a bit better by using technology and applying automation and those kind of stuff. So hopefully what, what we do generally can contribute in a similar way. Just make the world a little bit better. Exactly.

So I'm going to sort of move on now to sort of your reading list and, you know, the sort of things that you're interested in, in broadening your mind. Um, so imagine you can get all the contents and knowledge of a book instantly. Well, which book would you choose?

Yeah, I'm an older man already, so when I was a student, it would have been Gödel, Escher, Bach. It's an amazing book if you read it. Yeah. Then in my mid-20s, I went reading all the books of, most of all the books of Krishnamurti. Really insightful and giving me more understanding of, yeah, a lot of things about the world. But now I've got a second job, is I'm a skipper and I live close to a nature park, the Biesbosch in the Netherlands. And so I take a lot of people on the boat, tourists, and show them the nature park, and often I don't know the bird or I don't know this. So I would say I would love to read a lot more about nature, maybe specific birds, or so to at least to be able to transfer that knowledge and insight, because it's so wonderful every time you hear something why the typical birds are special about from a species. So I think I would love to— nowadays I would love to know a lot more about nature.

Yeah, yeah, that's absolutely true.

Yeah, and I, I can, um, sort of identify with that because, um, you know, nature— well, I'm originally from Africa and there's, there's a lot of sort of nature in there surrounding you and, um, and it's, it's so nice to see nature in its sort of natural environment. It could be something for computer vision where you can create an app and you point it at the bird and it tells you exactly about it.

I'm using such an app. No, I, I'm using Birds AI. It's an app, uh, it's open source. And so, uh, you, you detect the sound and then it, uh, it gives them the names of the birds.

Yeah, amazing. Yeah.

And it's funny, if you talk through it, it will tell, um, human species as a bird. By the I took South African friends to the nature park and they were afraid of falling into the water where we went kayaking or something because of the croc. In South Africa, water is always dangerous, like either crocodiles or sharks, but it's not in the Netherlands. So it's not in the Netherlands. You can also swim there.

Yeah. Just the worms, isn't it, in the ground.

I know the first time I've been to the Netherlands and there's plenty of water and you think, hmm, I better just read what's in this water before I go in these canal boats.

But no problem.

No problem. Can't imagine there's a big crocodile swimming in there, just tranquil.

In Amsterdam would be just bikes.

So, um, the next one we have is, what's the best advice you've ever received?

Yeah, I think that must be a personal one, but it's letting go. So, uh, don't worry about a lot of stuff, just let go. Uh, let somebody else, like, like now, let somebody else, uh, somebody else run my, my company. That I sort of took care of for many years. So it's a lot of things you just let go and then focus on things where you can add value to people and processes. And so I think it's that— I think it applies to me to let go.

Yeah, that's a very good advice. Yeah, nowadays with the busy world, I think we can all forget that very easily.

Yeah, so I'll tackle the last one if that's okay, Sascha.

Yeah, sure.

So Herbert, if you could be an Olympic athlete, what sport would you choose?

Short introduction, I did a lot of sports. I still do a lot of sports. So every morning I run about 7K, so marathon and stuff. So I did a lot of field hockey and snowboarding and telemark and you name it. I would say Slalom snowboarding. It's so nice, it's so natural, so hanging in the corner and dealing with the snow. It's all about— it's partially, it starts cognitive, so in your mind you have to control and understand what you're doing, all the type of movements. At some point, your body takes over. And so, like riding a bike. And I think that's an amazing sport. Yeah.

Yeah, very nice. Yeah, I'm snowboarding myself, but not, not slalom snowboarding, but I guess occasionally that happens, uh, that you do it anyway.

Yeah, yeah, I did it. I did some competition, but of course too late. I started snowboarding when I was 28, so no chance of being a competitor. But, uh, yeah, it's, um, it's, it's a very nice sport. I like it a lot.

Cool, perfect. Yeah, it has been a pleasure to have you on the show. Really nice to as well dive into the automation world outside of the sort of the back office within enterprises and insurances, financial services organizations. It's really nice to see similar sort of AI technologies are being used for that. And yeah, very exciting. And every now, every time now when I order a new Amazon package, think about Vizier.

Yeah, probably every— if let's say there are 3 items in that box, then probably each of the items will be picked by a different person, and then your parcel will probably be handled also by easily 8, 9 persons. So many people have dealt with your parcel, and if you order at 11 o'clock at night, it's being delivered in the morning. It must be during the night that people have been doing the work for you.

Yeah, yeah, definitely. Yeah, so we are so, so crazy these days, isn't it? So we want it tomorrow.

Yeah, that's my personal point. I would say we need to reconsider this kind of crazy economy, and if we can do it one day later, we can just have a lot of processes smoothing down and having people working during the day. They can be at their kids, at their spouse during the night. So I think we need to reconsider a lot of these things.

Yeah, indeed.

Yeah, thanks a lot for joining us. And yeah, one thing before we close the show, where would our listeners or how would our listeners get in touch with you the best way?

I'm on LinkedIn. I think that's easiest. Herbert 10 Haver, just send me a message and then we can talk. I'm in the process of, I'm helping other companies, so I'm a mentor for the last 7 years, helping startups and certainly in the logistics and automation domain, robotics domain globally. I can help maybe a little bit somewhere, certainly on the commercial part. So feel free to reach out, Herbert, 10 Haver.

Great, thanks a lot. Yeah, we will put your link and the link to the company in our show notes and then yeah, everyone can see that later on the automationguides.net. So yeah, thanks a lot again. And yeah, for everyone out there, so we will be back with another episode very soon. And until then, let's automate it. Unfortunately, that's it again with this episode of the Process and Automation Podcast. If you like this episode, Please give us a 5-star rating and don't forget to subscribe to this podcast so you don't miss any upcoming episode. We hope you will tune in next time and until then, let's automate it.