フラットな地球：エンジニアリングツールは消えるのか？

This podcast is presented by NXAI, your partner for Time Series Foundation Models and Physical AI. Hello, everybody, and welcome to a new episode of our Industrial AI Podcast. My name is Robert Weber, and it's a pleasure to talk to Peter Sieber. Good morning, Robert. You are completely alive Very much alive What's happening? Yeah, it's AI on the volcano week Right? So we record on Monday This episode is published on Thursday And normally Thursday early in the morning And yeah, we have plenty of interesting groups again A lot of participants Festo, KSB, you name it And Vitron and we are very happy to be invited by vitro and I think it will be a great event in Parkstein

on the volcano the very first listeners may not be you the Europeans but maybe you the Japanese or the Singaporean or whoever is first in this world because we make it go live I believe one minute past in the early morning but at the very time we will have already met on the volcano and we will start going into our sessions. But by then, what are we discussing today? What have you been doing? You had a very busy week as well last week, right? I had a very busy week. It was a, let's say, robotics-focused week, right? Let's start maybe with a little bit of a biased news, but because Seb and his team have published Kron LSTM. And you saw that? and everybody is asking me, what is this Kron LSTM? What is that? Yeah, it's a good question. What is Kron LSTM? I asked Sepp, and then he said, yeah, technically Kron LSTM uses the Kronika structured information

to update and control its recurrent memory state. Okay, what does it mean, Sepp? Did he say Kronik, like the time Kronos? Kron LSTM. Yeah, yeah, yeah. But he said, like, did he refer to, like, Kronos, like the time? No. No, he refers to Kronecker. And unlikely, he said, so, Zeb, can you please explain what this is at the end? And he said, yeah, unlikely traditional LSTMs or DeltaNet architectures, Krone LSTM introduces a Kronecker Delta rule. And then now it's a quote that selectively updates individual key values associations instead of modifying broader parts of the memory, right? and this is reducing significantly the inference with previously stored information, enabling more stable memory updates, better long context retention, and efficient recurrent processing. So this is the idea, and for sure, Seb is focusing there the whole topic when it comes to memory and robotics.

Yeah. All right. Okay. Okay. So that's what we're going to see, which is, I assume you're going to be doing another kind of announcement, which has all to do with his memory thing. And you say he's going to base that maybe on this Kronos or this, oh, let me see, Korn. Korn. Yeah, because I was going to ask, because in German, the word Korn, K-R-O-N, has a meaning. Crown. And maybe that is suggested that it should have. It should. No, it's Kronika. In the second announcement, I think a lot of people knew that already, that NXAI is working together with Nora, building on memory tasks, episodic memory, working memory for your friends' humanoid robots, Peter. So, humanoids and arm robots and all kinds of robots and robots with wheels and what else. Not only focusing on humanoid robots, but all kinds of robots. Okay, good. So I guess the first announcement is for those of you listeners that like to go deep or deeper.

I think I assume that there's going to be, not knowing each of you listeners, there's going to be a certain share. Maybe the major share says, okay, thank you very much. I'm very much interested in what it can do for me. How do I handle it? You know, how can I get it? All the stuff like that. Show me the results. can I use it all the good things like that and then many of you listeners like maybe myself will then leave going into the details, into the mathematical let's call them the mathematical details to others, but that's not like making one thing big or small it's just like some of us go deep and find new things, which is maybe we talk about that again today as well, maybe I have this quote again from. Well, do it at the very end. But I got a lot of feedback through the last episode, because there we had also this episode a little bit not industrial AI focused episode, more on

the research or more on the scientific side. But I got a lot of feedback there from people and say, it's great to have such kind of episodes all six, eight weeks to get a little bit of inspiration, what's going on when it comes to research, when it comes to other domains, right? and that's what we always say as long as you are interested in the things that we bring you and they are not directly related to your job maybe by listening to them driving in your car sitting in the public metro, tram, whatever you're doing you think oh no, what if I would apply what I just heard in a different area to my own job so it was a very interesting episode Lopez-Günter and Gonzalo very technically very deep when it comes to mathematics, but I think it was every, I don't know, eight, 12 weeks we can do an episode like this because we got a lot of feedback on that. What do you have, Peter?

Oh, I'm going to make a bomb again. Message of the day is software is dead. Maybe we shall cut in a minute of silence here. Can we do that? Yeah, sure. Are you joking? Yeah, but it is serious business. It really is. Now, so I am not saying this, and I mean software applications, right, in the end, because what is software? I'm not saying this because Brandon Thomas, he's a developer of ArtCraft, a suite of open source creative tools, and he meant to replicate Adobe applications, right, for free, and he has now announced apps to replace Microsoft Office Tools, Excel, PowerPoint, Word. We've all been using them. Some of us still do. Others do not. etc. I'm not. I'm saying this because especially in our world of computer edit X, X for design, you know, I happened to have studied computer edit architectural design engineering the last two weeks. It's unbelievable. An explosion, I must say explosion of AI based engineering announcements.

And I was preparing this because I was traveling just a day ago. And yesterday, you know, sitting in the car, driving in the CV, I was again looking at LinkedIn. And there were another 10 announcements. It's completely crazy. So I'm going to share some of them and then go a little bit more detail. And then that's the main topic of today. So in the first section then is software is that engineering. So there is Badr Buafir. He says, quote, unquote, three months ago, I shared a demo of AI generating a standard 3D model in Aviva. We recall Aviva, we talked about the last time as well, from a single engineering drawing. Today, AI can build a fully detailed centrifugal pump inside Aviva using 18 engineering drawings with a level of detail comparable to Katia or SolidWorks. Just the first, just the first quote. I'm going to give two or three quotes. Then there's Ahaoze Zhang. He says, or she, I don't know.

Last month, GPT-6 Astra Claude Opus 55 reached 0.96 on BenchCAD 1.0, raising the question, is AI close to solving CAD and 3D modeling? So we have already been building a harder challenge. Today we're introducing BenchCAD 2.0, where the best model scores is only 0.59. So, we talked about, you know, the benchmarks, you know, if I become so good, moving towards 1.0, you know, they have to produce new ones. Spectral Labs is another one. To train AI models that automate CAT, we had to build our own CAT software suited to machine intelligence. Now we have a fresh opportunity to enjoy, that's a rather interesting perspective, the art of making things. And I will come back at the very end to this thought once again by disencumbering engineers from all that is rage-inducing about design software. So their takeaway is CAT built for agents steered by engineers.

Wow. I think that's a very nice thought. I think it's a very nice thought. I'm not saying it's true, but I do feel that that is the direction where things are moving. Maybe, yeah, I can't just tell all of them. And, you know, here's a guy who is building his own engineering system. And then, of course, we have the guy you did an interview with, our Davey, Davey De Meijer from, I believe, Belgium, right? And he is doing, he's building plants overnight. Every morning when you look into his stream, he built a yogurt plant. Not he, it's Claude. He says, Claude, design a biofuel plant from scratch with six fermenters. That's all he says. And then he wakes up in the morning and there's his biofuel plant. It's amazing. There's a yogurt plant. There's a chocolate plant. That's it. And then Claude decides on the plot size. That's what he says. Designs the PFDs, not PDFs like Adobe.

I had to look it up. It's the process flow diagrams. The P and ID is piping. and it's we, you and I, I mean, I started this a little bit, very close, but anybody can look at it and you see a plant. The engineer, the persons who have been doing this for 50 years, they go into the detail. But, I mean, as you know, as you recall, you know, Davey's a smart guy, right? He's not just showing, he's not showing off. He's so happy to share what today is possible. So I'm going to round this off by going a little bit deeper. Derek Alexander Molitor, he says the same. Engineering is becoming a vendor. Engineering AI, sorry, is becoming a vendor. He says over the last months, you know, every major engineering tool vendor has started to move faster. Everybody's doing differently. Just to call them out, we have in our brain who we're talking about. They are Siemens, PTC, Dassault, MassWorks, Cadence, Synopsys, Ansys, Autodesk.

Those represent probably 90% of the market together, right? They all do it differently. Now, I'm not going to go further here, but he says the tool is no longer only the machine, the engineer, right? It becomes the assistant. In the end, he's saying, you know, in a few years, engineering tools will look very different from what we use today. Now, and that's what I'm going to give my reply to is say, agree, Derek, and that's the only thing that we know for sure in a few years. engineering tools will look very different than what we use today, right? So in my humble opinion, and you, dear listener, by now know what I'm going to be saying. What do you think I'm going to be saying, Robert? Software is dead. I'm not sure. I really believe the software is dead. This engineering software, I don't say is to that today, but it will be dying very, very soon. The future value, I'm so much convinced, is going to be created by context data.

and that's the thing I want to do, and I need to talk to you about that later on. I want to do an interview on very specifically context data and AI. For that reason, I believe the winners are going to be one of three parties. Maybe there's a fourth or fifth party. So number one, of course, the engineering tool vendors, I'm not saying that they cannot change everything. They will and they are changing everything. They can buy AI capability from, for example, AI platform vendors. Number two, the AI platform vendors, they can buy engineering context data. You know, they can buy industrial engineering companies. You know, let's think about what AWS is doing. Forgot how they call it. What's that? Yeah, that's exactly that. Number three is the customers, the industrial organizations that so far have bought these tools from the engineering tool vendors. And now they buy AI from the AI platform vendors and develop their own solutions.

Number four, open source. Number five, there's many, many different ones. And I'm going to close this first section off with saying, and by the way, the incumbents are forcing their customers out. By example of Autodesk, they started charging when software pulls design data out of its cloud. So as long as you work with the Autodesk tools and you access your work, it's your IP, right, then you're okay. If you're taking your designs out and you're putting it into, for example, your local AI, you're paying for it. And that's where the community is getting completely crazy about, right? What do they think that they have the right to the IP of the customer? So I'm going to leave it there. It's so big what's happening in this. And this is only one section. As I said, I started off with the guy who does the Adobe Excel PowerPoint,

and we are here now concentrating on the engineering, and it's really huge what's happening there. But let me make a reference to our friend Boris Scharinger, because he already mentioned this in his book, right? He said the whole engineering process will be totally different in the next two, three, five years, and he sees a huge potential when it comes to AI in engineering, right? And this was his first, let's say, sweet spot when it comes to industrial AI from Boris' perspective was maybe not optimizing shop floors, more optimizing engineering processes. And here we are, right? And it's so easy to open a startup going into this domain, right? So this is also – because you can see every week three, five, six, seven, eight, nine, ten new startups focusing on engineering topics, right? And it's very difficult to understand what is the difference between these startups at the end.

At the end, it's about the context data maybe. What is your USP exactly? Do you have this context data running the best CAD models in the market? I don't know. But this is very interesting because we see a lot of startups in this field. We see a lot of money going into this field. Let's see what is the outcome at the end. I'm very looking forward to this. Yeah, and this is not bashing the engineering company. Not at all. Any of them are doing, well, some of them are doing weird things. As I just said, I don't think they should do that because why would they? And what I was saying here was quotes that I've picked up, so I'm not bashing those either. But they are the incumbents, and there's a huge potential. And if you say startup, many of you listening have already thought of doing exactly this, maybe a year ago. Maybe you're in the very middle of doing it. And then, of course, the question is, what is it you're going to do knowing that you,

and not only the listeners of our podcast, that you are together with a thousand other people thinking about something similar in the same area. That is not to say don't do it. Do it, but keep in your brain that any of you engineers can today, say, design a biofuel plant from scratch with six fermenters. Any of you can do that. So that must be your starting point. At least that one. Maybe there's even a better one. And I'm just using this quote, so to say, from Davey. And that is what is possible today. So if you can't go beyond that, I would almost say then, you know, think about it, because that is what is possible today. And I'm not, again, of course, it's going to be the shitstorm, you know. But then I'm not going to refer back to Davey for that reason. But I think he's going to say the same. You know, what is happening today gives him, I don't know, if it's 60% or 95%. and the missing, the 4% or 5% in relation to what we can do today

is going to come over the next couple of weeks or months, right? There's a German saying, I have another news, right? So there's a German saying, ein gutes Näschen haben. Good nose. Do you have a good nose? Yes, have a good nose. It's maybe not the right translation. Maybe have some feeling for what's coming. You have a good eye for opportunities maybe or something like this. I think three years ago I met a guy on Twitter now called Axe. His name was Mr. Schmidt, right? And he was at Mila in this time in Montreal. And I did a recording with him on climate change and AI and stuff like that. And then I met him two years later or one year later in Paris. with his own company. And this company was called Antalpik. And they did a little bit the same what Max Welling is doing with his company.

New materials, yeah. So research new materials, right? And last week, I saw news that NVIDIA is now investing in Antalpik. So, yeah, if you listen to the Industrial AI podcast and you want to invest in such companies, I think you listen first here. So we have a very good nation A very good nose You have a good nose Yeah And yeah very interesting Congrats Were you in the house Yeah Jensen was in the house Was in Berlin right with the chancellor You were not in the house. No, I was not in the house. I was sick last week a little bit, yeah. Jensen was in the house. GTC taking place in Berlin, yeah, as you say. Many, many announcements. Loads of them around robotics, right? Loads of them around making, you know, robotics. Yeah, agile robots, et cetera. A lot of robots. And, yeah, it's a big show, right, GTC? Maybe it's a small Hannover mess. You have a quote on that, right? Yeah, I have a quote from the ARC guy. Sorry, I forget your name now, but you had that thought.

You were going there. You said it's, yeah, were things moving, changing? And I was thinking of it, yeah, I don't know when, somewhere, sometime last night, thinking of here is one company, NVIDIA, saying, you know, we're going to come to town and, you know, we're going to invite you. I mean, they come with a specific topic. Maybe we're going to invite 10, 100 companies. I have no idea. And it's the complete ecosystem. Their ecosystem joins, comes together for what? Two days, something, I believe. Three days or four days. Three days, there you go. You were there last time, I believe. Was that the Paris one or what was that? The Paris one, yeah. Yeah, Paris, yeah. And that's amazing, right? I mean, it's also a little bit careful that one company has that is the world. Yeah, their ecosystem is a very big ecosystem, I would say. Exactly. But having said so, I mean, I haven't met Jensen,

but he seems to be the most modest guy in the world that you can leave this thing to having a huge saying, so to say, right? Yeah, and it's very interesting because in parallel to GTC, there was an investment conference in Germany. Oh, I saw that as well. Did you see that? Did you see the quote? Did you see what they were talking about? Yeah, exactly. So we have our chancellor, our chancellor, Mr. Matz, and at least the press release that I saw, the quote was that they were going to talk about AI in industry, industrial AI. And I thought, isn't that amazing? You and I started this topic seven, eight years ago. It's amazing. And the quote they used in German, KI in the industry, was the wording of the podcast that we started with eight years ago in the German language. Exactly. It's amazing.

And I want to come back to Hannover Messer because the USP of Hannover Messer was also always this political stuff on Monday morning or Sunday evening and stuff like that. And now they are combining GTC with a huge investment conference in Germany with a lot of celebrities, politicians. Mr. Mertz is doing a talk with Jensen on stage, etc. So be careful in Hannover, I would say. Be careful, because this is very huge, and NVIDIA makes no prisoners, right? Yeah, I didn't see the outcome. And of course, we must say, you know, and we don't want to be political. We never want to be. But where our chancellor is coming from, I mean, that's what he's been doing for what? I have no idea. 10, 20 years for investment. that's his topic I believe I started listening to it but I didn't have a good connection where I was at that time but I think it was amazing it's amazing for us it's amazing for the podcast

what we started and then I was thinking because we had in the beginning we were very very proud and I still am very proud that we were doing the autonomous driving as well and we had this amazing interview and stuff. And then I'm thinking back, you know, if we would have before then decided, which we never considered. No, we didn't. We never did. But we took it because we thought it was a very interesting, almost side topic, so to say. If we would have said, you know, we're going to be doing the autonomous driving podcast, well, we would not have been as successful as we are today, in all modesty, having said that at the same time. But I think, and I must give that to you, Robert. You had, and you have, an amazing Nesian. Nesian, good Nesian. Nose started with this very topic that everybody in the global language even is calling industrial. Yeah, I know. Thanks to Boris, right? So it was also Boris' idea. Most certainly, yeah. Peter, what answer do you have? I have two, three

very quick ones. I'm going to give two other topics on my hypothesis that software is that. So there is the new layers to the automation pyramid. I'm not sure you saw that one. So we had Knud. Knud is the other guy that you did an interview with maybe already a year ago, right? And he came back from which one? I think from one of the… Chicago? No, he went to the IMES somewhere talking about the events. And he says that the ISA-95, so that is the famous automation pyramid, right? It's got new layers. And he describes the different layers where we were. And I recall we had the different layers. You need to remind me again because I think you have not forgotten his name. The professor. Knut Lasse Lüt was the guy who shared. Yes, that's number one. And then you need to remind me of the other guy, the professor that I went to Japan with at that time. Mr. Industry.

Zülke. Professor Zylke, Professor Zylke, very sorry. I always keep on forgetting your name if you listen to us right now. But you have Robert, your agent. I very, very exactly recall. You just spoke on Festo. In the main hall of Festo, we were there with the OPC Foundation, and Professor Zylke, he announced OPC at, I believe he announced it at level 5 or 6 at that time. And everybody had then understood what OPC was really about. Now, the point is, Knudsi's eight layers, I'm not going to call them out, all of them are from the bottom. There's physical machines, and then there's software, and then there's somewhere, data ops and ERP. And then come new layers, knowledge graph, five, six protocols, MCP, A2A, MHS, and then the agents, number seven, and on top, the applications. So now, together with Jan Wohlkittel, who I will be doing an interview with very soon, right?

I do not agree that the pyramid is still the best way to represent, I must say. Maybe I talk to Knut again, and maybe you will talk to Knut again sometime as well, because it is his thought, right? Now, Jan says, because it says that the data flow is still one way, so from top to down through each layer and vice versa. but the discussions today show that data directly from equipment needs to be available in the top end. I'm going to close it off there. And I go a bit further. So that is Jan. And maybe we hear more about this when I do the interview with him. I go a bit further, say. In my humble opinion, the new world, let's say the new earth, is flat. Flat earth. You know, you remember when people thought you could fall off? Yeah, exactly. And as I recall, I was traveling. I did spend a lot of time on the sea. I did not smoke. But sometimes when you look across the sea, you can kind of feel, well, we know that it's kind of around shade.

But the people, actually, there were 2,000 or 3,000 years ago that people thought so. And not only 500 years Galileo, that was a different topic I found out. But people thought you could fall off. But what I mean is when agents can hop across the layers as if they did not exist, we don't need them anymore. I really believe so. For me, the new world is flat. And that's the question I'm very, very much happy to have with Knut as well. So when he says we're going to have new layers, I say, no, agents don't need the layers. So why would we need them? So the title for this episode is Flat Earth. Flat Earth, yeah, right. Software is that Flat Earth. Flat Earth. I'm going to just quickly bring one more and then I'm going to be very positive again. So there. Anup Karumanci says, PLM is more than managing CAD files. It's how everything connects together.

Modern product lifecycle management brings engineering, manufacture. It's very similar. he has all these different it's not like the vertical but it's like horizontal right so one comes after the other you go back and forth it's the horizontal thing and of course I come I am the bad boy and say you know A. I will integrate I make him all superfluous that's again of course very bad and people don't always like that and I do very much understand that and that's why I will always say to each of you you know make sure look at your own job Look at what's happening to your job and change your own job. You know, don't wait until it's going to be changed. And having said so, I'm not the person who always has it right. You know, Robert is the guy who has it right. Robert is the guy with the nose. So what I was saying, and I didn't realize that, the first quote was about the vertical integration. I say no additional layers. I think everything is going to be flat.

You know, AI doesn't care that a sensor value is at the bottom and that, let's say, what we had called an app, an app value at the very top is probably the same. It doesn't care. You know, you're asking it to design your factory with six fermentators and it does it for you. And it will include whatever is there from the bottom sensor machines to the top, what today we call application. We don't really care. We call it application of software, I think. Right. Do you remember, I think, one and a half year ago, I told you that there's a new battleground when it comes to the new domain of models, right? And I said time series is a new battleground. And we can see this in these days because we got a lot of new models in the time series market. So we get something from Mistral. We have something new from Google, something new from Toto, Datadog, something new from LG, from banking accounts and banking companies.

It's huge what's happening in the time series market in the moment. So I see that really is a new battleground for companies, for models, not only LLMs, but also Time Series Foundation models. And it's huge. It's really huge in this time. Okay. Yeah. Kronos is there as well, right? Kronos, Kronos II, Mistral launched a new way of doing Time Series. Then we have TO Gamma, TO Better. So French companies, interesting, very interesting, the new battleground, I think. One new battleground. I was listening to Plato's Socrates this morning, and he did a reference to Kronos, which I think is one of the gods. He must have been the god of time, I guess, but I didn't go into the details there. Perfect. What do you have in the main part? Main part. What do I have? Should I not tell you? I'm going to tell you the other things later on. Very good. No, that's first. So I'm going to be talking to Torsten. I did already talk to Torsten, so I'm going to be talking to Torsten about connecting agents, AI agents, with industrial robotics.

Now, at that time that we spoke, a couple of weeks back, specifically by means of MHS, SRCI. Now, Torsten is the guy who, in his spare time, I believe, like three months, he just built a robot. Okay. He's going to explain what SRCI is, what MHS is. We have heard about that one. So those are the topics. And then in the end, he has a view on how to bring AI agents to the real world. And I must say it's almost like it shows how awfully fast our time is running. This is an interview from two weeks back. And here we're talking about all these layers and all these acronyms and where then, you know, just five minutes ago I said, well, are we still going to have them or not? And that's a topic that we're going to continue. I'm going to continue that with Jan as well. But for today, I hope you're going to be enjoying my talk with Thorsten Brach here. Okay, he's with Bosch.

Yeah, exactly. Bosch Manufacturing Solutions. Okay. And he's a firefighter. I see that on his. Oh, he is. Oberfeuerwehrmann he is. Oh, yeah. Very, very important. Very important. Thanks a lot. It's amazing that he, well, I'm not sure, I hope he hasn't had too many firefighting. With the agents. He was next to his job at Bosch and next to having built a robot completely by himself. Perfect, Peter. It was a pleasure. Thanks a lot. Thanks, Robert. And take care and we will see each other at AI on the volcano. In the next couple of days. I'm looking forward to you. Thank you very much, dear listeners. Hope to have you with us then soon. Thanks. Bye-bye. Bye-bye. My name is Peter Seberg, and I'm your host. And today I'm going to be talking to Thorsten Brach. He is with Bosch Manufacturing Solutions. And Thorsten and I are going to be talking today about connecting AI agents with industrial robotics,

specifically by means of, and now there's two acronyms, always need to be careful, MHS and SRCI. Hi, Thorsten. Hi, Peter. How are you doing? I'm doing good. And you? Oh, yeah. Well, no, yeah, I am. As I suggested before, as I'm based in Munich and as I'm part of a marching band, we were playing yesterday, but I'm perfectly happy and fine this morning. Thank you very much. So let's start with you, please, introducing yourself to our listeners. Yeah, I'm Thorsten. I'm a software engineer with more than 25 years of experience in automation and robotics. And yeah, it's also kind of a hobby. So I spent quite a bit of my free time working on my own project. But I think we'll get into it a little later. Yeah, sounds great. having your hobby as a job, isn't that the perfect thing that everybody would like to go for, right?

Yeah, that's great, but then it's also hard to find an end at a day, you know? Yeah, you never know, which is the same thing when, because I tell people all the time, it's the same, that's now, how long do we have mobile computing? I mean, how long have we been using notebooks and then later on, even worse, you know, our mobile. So as soon as that started, I guess, whatever, 20 years ago, 25 years ago, wherever we were, wherever we are, we can always work. And it's the same topic. Nevertheless, please tell us a little bit more about Bosch manufacturing solutions. Where does that, I don't know, company or department sit within the big world of Bosch? You may want to spend two or three words on Bosch as well. I think most listeners will have heard of them. And what industries, what product areas do you as a company serve? Yeah, Bosch Manufacturing Solutions is a Bosch internal special-proposed machinery division

with different locations across Europe, Asia, and North America. We are what you might call a full-liner. We have a lot of experience in areas such as battery, e-tri-sensor, electronics, and as well as hydrogen technologies. But we don't just work on internal projects. We also deliver projects for external customers. And I personally work in the department where we develop the NextSeed automation platform, which provides the foundation for most of these projects. Sounds great. Before we're going to move into today's topic, you, oh, that's an update. You just built a robot in, I think, around 100 days. We're exactly at the topic that we started in your free time. Tell us a little bit about that. What was the idea behind doing exactly that? Yes, thanks for asking, Peter. Yeah, you're right. I built this in around three months. And the idea behind is, yeah, we all see the impact of physical AI. And we see all the humanites coming and so on and so on.

And, yeah, I work in a traditional automation and robotics department. And I'm not so close on the front of all these physical AI things. but I think I will need the skills in the near future because the systems are coming. And yeah, so I decided to make a project. I built my own humanoid robot as an open source, so it's available on GitHub. Yeah, and start building the hardware and I already have it running, but with a traditional PLC inside, so all the AI magic is still missing. But it's exactly that I want to learn. I would go through all the topics which are needed to build such a system. And, yeah, I think it's a great playground for me to discover how things work. That's amazing. We're going to come back later. We're going to be talking a little bit more about Uenoise as well. We want to know your perspective on the basis of what we do talk about today. Basically, you know, your, and maybe let's start with that one, your experience in robotics.

You say you have, I think, you say 30 years as a software engineer. Now, please share with us a little bit more detail your past involvement, your experience with robots and robotics, because I do assume that a big share of our listeners will know and will have heard or will have been part of that world. But there is most certainly probably many people that are relatively new, specifically coming from the humanoid part. So tell us, share with us, what have been the types of robots, the brands that you've been working with, their capabilities and the responsibilities that you had in exactly that world? Yeah I think I already worked with all the major robot brands but I must explain I not on the robot side itself I mostly come from the PLC side I do the integration of the robots into our platform and write the function blocks And we have a

technology package, it's called integrated robotics, where we move the programming from the robot into the PLC so that the teach points are in the PLC, the tools, the coordinate systems, and so on. and then we enable the PLC programmer not only to program the PLC program but also program the robot application. And that makes things easier because on the one hand side, you have a kind of an abstraction layer so you can scale solutions across different robot vendors because if you program in the robot system, then you have to develop things multiple times for each robot brand. And if you move the programming to the PLC, then it's easier to implement the applications And in case of service, for example, then you don't need to decide, do I have to send a PLC programmer to the customer? Do I send a robot engineer to the customer? Or do I have to send both? In this case, the PLC programmer is enough. He can do both jobs. And you don't have to consider if the person has a KUKA experience or a Yaskawa experience, for example.

Exactly, and that is a problem because mostly your robot engineers only can support one or two or maybe three different robot brands, and then you need another expert. And with this approach, you don't need that deep vendor knowledge. We're in the middle of our topic, and you trigger specific thoughts, specific questions I have, But I'll ask them later. Remind me if I forget when we're going to be talking about just as one example of a communication architecture, OPC UA, because they have a very specific solution there as well. But we're not going to start there. We'll do that later. We're going to start with me asking you, as we're going to be talking about connecting, you know, AI agents with industrial robotics, which is maybe things that we're doing these days, specifically through MHS, SRCI. We're talking industrial robots and connecting them.

So before we get to the topic of today, maybe you want to give us kind of an introduction into kind of an overview of what is industrial communication. There is different protocols. What are the types of protocols? What can they do? What has been your experience in using them? How does the PLC communicate with the robot? There's many different languages. You know, there's Profibus, Profinet, others. I don't know. Maybe you have your own languages, so to say, that you are using. Yeah, mostly you have a field bus system between the PLC and the robot. Yeah, that depends on the brand of your PLC system. I would say in the Siemens world you always have the Profinet. In the Beckhofer world you have the Isacad. But there are a lot of other field bus systems as well. And that's a little bit of the field bus war, I would say, you still have. When did that take place?

That's meanwhile, because I was part of a provider of industrial communication solutions, maybe just after. I think when I met people, representatives of exactly the solutions you just mentioned, I did recall still the feeling that they didn't really want to talk together. That's maybe like 20 years ago, I guess, or the field buzz wars. Yeah, it's still active, the field buzz wars. But I hope that we will overcome this and we'll have one common solution in the future for interoperability between systems, vendors, and devices. Okay. Very good. Because we're moving into today's topic, we're talking other ways maybe, or maybe they sit on top. When we're talking MHS, SRCI, and I want to know from you first what they are, I believe they sit on top on a higher level, so to say, than the communication protocols, right? So tell us first, what does SRCI stand for and what does it do?

Yeah, SRCI stands for the Standard Robot Command Interface. And, yeah, that's a relatively new standard that aims to standardize the communication and the control of industrial robots across different robot vendors. And this has already been designed with future in mind. It's not limited to simple pick-and-place applications. It also supports more advanced applications such as conveyor tracking, force control, or trigger functions. It was originally developed by Siemens together with several major robot vendors. Later, it was moved under the umbrella of PI, Profibus, and Profinet International, which was a great move because it means that the standard is now managed by a neutral organization where it can maintain and further develop. But does it then mean, for example, that, and number one, was I correct in assuming that a standard like SRCI sits on a higher level than the communication protocol, in this case maybe Profibus, Profinet, And does it mean because they are in the same organization that typically the SRCI standard is being used by companies that at a lower level used in the past PROFIBUS,

but today maybe PROFINET, a communication standard? Yeah, you are right. The FIBUS doesn't matter. The protocol is based on, I would say, an unnumerous byte array. And it doesn't matter which protocol you use to exchange this byte array. There's a protocol inside And you have the mechanisms to pass this telegram And get the information out of it Okay, I understand that So that's the first one Next one, MHS What does MHS stand for? What does that do? Yeah, I would say it's better to start with MCP If this is okay for you Because then I can connect MCP to MHS Then I think it's easier to understand And, yeah, so MCP stands for the Model Context Protocol. Yeah, it's an open standard that allows AI agents to communicate with external tools and data sources in a standardized way. Yeah, for example, an AI agent can use MCP to access files, databases, or other applications without requiring custom integration for every system, I would say.

You can understand MCP as a universal connector between agents and software applications, I would say. Very good, and it was introduced by Entropic What, a year ago, I guess, or maybe longer? Yes, exactly And we already see the huge potential of this protocol Especially if it comes to making interaction with software more efficient And AI agents can automate tasks Which would otherwise require a lot of manual work So then we come to MHS And there is where the things get really interesting I mean, MHS stands for the model hardware standard, and that's the PONDO of MCP. While MCP is more for connect the agents to the digital world, MHS is to connect the agents to the physical world. Okay. Very good. So we have a little bit of a perspective here. We have at a certain level the communication protocols, and then at the higher level now we're talking,

well, maybe one question is, but we'll do that later. It's like what is maybe interesting, I don't know, your perspective on that is that as long as we've been talking about the communication standards, the companies typically that would bring these standards, you share two which are more European, European at least originated, Backhoff, Siemens. There's others. There's specific ones for the Japanese market, for Rockwell, the American market. They were typically automation companies. And the two ones that you now talked about that we are going to go into detail for, and then SRCI, I think you were part of that one. Now we're going to move into that one. But the MCP and MHS now have been brought into the market by companies that are not from an automation. space, right? Yeah, that's right. So, we're going to see what that means maybe later on. Now, you maintain an open source implementation of SRCI for, and I'm going to just

spell this one out, IEC 61131-3 environments. And, of course, I'm going to ask you what that means. With the SRCI core profile already tested with robots from multiple manufacturers. Now, those listeners that come from the automation space that can call out this IEC standard by heart. For those that are not, please tell us what that standard stands for. Yeah, this IEC 61131 is an international standard for programming PLCs. It standardized programming language, program organization, data types, variables, and execution concepts so that PLC programming follows a common structure across different manufacturers. That's the idea behind the IEC. Okay, very good, because we're going to be finishing, that's at least what I have in mind, with a big question that I have. Yeah, if we're still, you know, how far we humans,

we the people that have been coding exactly along the lines of this way of coding, programming PLCs, if we humans are going to continue to do that. But we keep that thought for the end. Now, you have maintained this open source implementation. Tell us a little bit more about that, and maybe you want to move on already to how this standard you are now applying MHS, what this is all about. Yeah, okay. I would start explaining why I implemented this interface as an open source. As I already mentioned, from my perspective, SSCI is a game changer because it makes things easier if it comes to robot integration. And because this protocol was introduced by Siemens, the Simatic POT system was the first POT system which had an integration for SSCI. And I was completely excited about this interface and was waiting for SSCI implementation for backoff and for coders, this space POT systems, but nothing happened.

And at the beginning, I was checking their websites every month, then only every three months, and at the end, only every six months. And at some point, I was... Why is there nothing there? Yeah, exactly, exactly. And at some point, I was really disappointed that there is no implementation available. And so I decided to implement SRCI interface by myself. So what does it mean? Maybe share what is the big picture of providing these different solutions, of you making that solution available. What does that mean? What is now different from what can you or what can somebody else, because you make it open source, what can you now do? And maybe you try to give us a use case, an example of what you did yesterday, you know, the last 30 years, step one, two, three, and how you do the same thing or a similar thing today with different steps. Yeah, I think the problem in the industry is that every vendor has its own proprietary interface.

If you have to integrate in robot, you have to read all the manuals and do all the integration work to integrate the robot system. And, yeah, if you have a standard interface like SACI, you just have to maintain one single interface which fits for all the robot brands available in the market. Not all, but we see that more and more robot vendors are joining the SACI working group. So maybe in the near future, we will have all major brands inside. And that makes things a lot easier for the applications because they just have to include just one single interface and then they can work with different robot brands. Now, okay, I think that's already giving the answer to the question I have had is like how does MHS work? How does MHS make these hardware or implement this hardware standard, so to say?

So we're typically MHS. So we have 10 robots, industrial robots. We start with them standing in front of us. And what would MHS do? MHS would go to brand number one, let's say it's a KUKA. And what would MHS do? Would maybe read all exactly these manuals, as an example, and then decide, okay, I've read everything. I know everything about how to communicate with the KUKA. and now I provide you, you which is then, tell me, it's an agent or it's a human, but now you can just, you know, you can just communicate with them. And it will do the same with number brand two and number three. Is that, and then the following question is like, how are you helping MHS by providing it with the, oh, I forgot my acronym, SRCI interface, right? Yeah, I think our industry is very slow. So I think if you would wait until every robot brand has an implementation for MHS, I think then the standard does not play any role anymore.

So my idea was just take an existing solution like SACI and make like an adapter and bridge an MHS to SACI bridge. Then you don't need to implement MHS directly on the robot system. Then you just have this adapter and you are able to connect all SACI capable robots with MHS. So that's basically the idea behind. I understand. Now, maybe this is the point where, before I'm going to forget, I'm going to be asking you, and there is a disclaimer, you know, those of you that do not know, Robert and I, we also co-produced the OPC Foundation podcast. It's just a question. It could be with any of the other ones, but specifically, and I've referred to it here a couple of times, that OPC UA maybe is the, and they don't want to be called like a communication protocol. They see themselves like as an architecture with security embedded, et cetera, et cetera.

But the most important thing always has been this information model. And this information model, at least as I've understood it for the last couple of years maybe, also always had this potential for doing for AI, for standards, for artificial intelligence, Meanwhile, super intelligence doing something with the machines, with the solutions we have. Now, I know specifically that these information models, they exist for specific subsets. You know, whatever kind of machines the companies come together, they produce this information model. And one of those exists, I know, for robots. Now, where is that overlap? Are you aware of that one? and where's the overlap or where is your solution different from the one that is maybe being provided by, as an example, OPC Foundation? Yeah, I know what you mean. That's the companion specification, I think. Yeah, I'm not so deep inside what's going on in this direction.

Yeah, I think we are more in the, how can I say, in the practical way we communicate via the field buses, with the systems to have the high performance, the real-time behavior. I think this OPC UA will not have that capability with future. Maybe this OPC UA FX and TSN, maybe then in the future it's possible to have also the performance with OPC UA. And I'm very sure I think the standards will grow together in the future. Okay. Okay. Phil, can you give us an example of what you, you know, typically what you were doing a year ago or two or three, four or five years ago when you wanted to start communicating with one of these robots that we talk about, right? What did you have to do? How long did that take? And then today, today you still want to do the same. You or anybody else listening and interested, what do they do instead today?

And how long does that take? So what is the difference that you are providing them with? Yeah, in the past, you had to read all the manuals of each robot vendor. Then you have to define an interface because you just have some inputs and outputs which are exchanged between the systems, but there is no standard on these bits on the field bus. So you have to define your handshake flex between the systems. You have to define your values you want to exchange. Then you have to program a kind of handshake in the robot program, and you have to program your PLC function block with a handshake inside, where you're then able to trigger a robot program on the robot side, and then you have to program everything in the robot program. I think that is what is done in the past. And that would take, depending on whatever your job at hand is, but that would take a couple of weeks, a couple of months, I don't know. Yeah, this would take a couple of weeks, for sure, a couple of weeks.

So, and then, what do you do today? What can you do today? Yeah, today it's very easy. You install a technology package on the robot for SSCI. Then you download the SSCI open source project. If you work with Codesis or Trinket-based PLC system, or if you have a Thematic PLC, you can use the function block from Thematic. and you can directly program your robot in the PLC without defining programs, without defining interfaces, without defining which values we exchange. Everything is predefined in the standard, and you can directly use it and directly access your robot. Now you still programming but you programming at a higher level So is it more like what you had to do in the past after studying all the manuals and you stick to the IEC 6.1, no, that's wrong, whatever that standard is, a very specific standard you exactly need to abide by the rules of how that standard is written.

when you say are you today when you say programming do you say there is some kind of solution that you have I assume you have some kind of application that does something that needs a robot and that means it's going to give signals to the robot and drops to move to the left pick up a box put it down, pick up the red bin whatever is the new way of programming at this higher level of, you know, do things. And in the past, it was about, you know, more cryptic sticking to the rules of the language of the robot. Is there a difference in how you, in what it is that you program these days? Yeah, I would say in the past, all the robot language of the major brands, there are very old programming languages because the base of the robots software was developed

I would say in the 1980s around. That's the time that I learned probably. So they're probably in Fortran or Pascal I have no idea what they were. And because the vendors would be downwards compatible, they was not able to modify the language a lot that was a problem or is a problem in the programming languages of the robots. If you move it to the PLC, as I mentioned, then the things get a little bit easier, but the EC61131 is also not in modern programming language like Python where we already have object orientation inside, but it's still a hard work to program in the PLC and I think in the future it will be hard to get people who want to do that. I understand. Now, our topic is that we're talking about connecting agents with industrial robotics. And I think in the meantime, I have discussed this a couple of times,

at least once with Robert in our news section, and I didn't really know what it is that I wanted to discuss, but in the end I thought what is happening is that we've been, let's say for 50 or maybe 70 years, as long as we have industrial automation and also in general coding that we have, and you give the best example with this IEC standard, we need whoever wanted to build some kind of industrial automation solution or coding, you know, whatever the language was, you have to stick to these languages. You had a thought or you get a job, and you think, okay, now I'm going to start programming by using the languages that you learned. Now, my hypothesis is that what we're doing, what you are doing, as one example, is that we're skipping. So we're moving up again another layer. We talked about the communication standards at a certain layer. Maybe there are – I think there's also global systems for discussing these layers.

So let's say they sit at layer three. And then on top of that, we now are talking about a new layer that didn't exist before. so is that how do you see that working for yourself for the people in your community for the listeners are agents but also humans are they gonna be programming quote unquote the robots in the future by talking their own language so german or dutch or french or whatever chinese by saying go to the left, go to the right, whatever, rather than talking the cryptic languages of the standards that we have been using 50 years. What do you think? Yeah, I think in future it will be like to describe so that you just say, for example, pick the red cube and the robot will pick the red cube. But I think inside the software we will still have the normal deterministic programming as before. I'm with you, but who takes care of that?

And again, I have been slaughtered before. One, two years ago, we talked about this kind of topic when it was about general coding. So I'm the biggest friend of people who are capable of doing whatever it takes right now. But it's maybe a thought, a positive, that the agents, that agents will take care of it. So software take care of exactly these, as you call, deterministic layers. And of course, today we're not discussing that. I believe you cannot, from today to tomorrow, just build your factory and have it completely done this way without any human being capable of understanding any of the things happening at that layer. We're talking, you know, I'm suggesting these things may happen in the future. And the thought is, you know, what does that mean that at a higher level, a person who has less understanding of that layer can still be very useful at the higher layer?

And maybe he or she needs some person at the lower level who can go there if necessary. I'm not sure if that makes sense. Yeah, I think that's a real problem in the future. if all the software is AI generated and if something breaks nobody is able to understand how the systems work and I think that's a problem we have to solve that we are but is that really? is that really so? the example I always give and I have this friend and his friend is 80 more years old so he learned programming before me He was more like the 60s or 70s, I guess. And he always gave this example that when he was programming, he was programming on the chip level, and he had to move registers from left to right. I have no clue what that means, but I can imagine. It was something like, you know, making space somewhere on the chip free for something else.

He had to do that himself. I assume that you have never done that. I've never done that. But the point I'm making is, you know, this is being taken care of by, I assume, some software layer on top of the chip, right? So the chip manufacturer, whatever that chip is. And I'm not saying that for that reason we can forget all and then we've built so many layers on top of it. But nevertheless, I'm not going to try to put any words in your mouth here. at least I would agree that if we're going to be moving that direction I would agree that we need certain ways of confirming that whatever the agents may be doing at a lower level makes sense. Now still we haven't talked about the agents kind of yet. Do you have specific thoughts there? I mean can you share a project? Can you share specific things that you are doing with agents connecting to industrial robots.

Yeah, I'm experimenting with it. I can share an example and proof of concept. In our next heat automation platform, it's possible that you can release functions to serve party applications via REST API. And, yeah, I was experimenting with this. I released some functions of my machine. This was the robot. There was a predefined function for pick apart, check apart, and place apart. And then, yeah, I used fast MCP because MHS was not announced. I used fast MCP and took this REST API and provided the capabilities of this REST API via MCP to an AI agent. And then I was able to tell the AI agent, so pick the part and check the part and place the part. and then the agent was able to execute the sequences. So is it like an interpreter maybe? I don't know. I mean, so you're sitting there and how do you do it? You write it, but you could speak maybe as well.

So what is the receiver? Is the receiver an LLM? Or so you say, you know, move to the right. Who is the receiver of this command, move to the right? Yeah, I talked to the agent, but in the background, the agent used the MCP to address my REST API and then it was calling a deterministic function which was already programmed on the PLC. Yeah, right, okay. So the execution, as you say, still happens deterministically, right? Yeah, but I think that is something we will see in future if the AI agent just reacts as an orchestrator of predefined functions. This has the advantage that you are able to have interlock mechanisms so then you can take care of that the agent cannot break things because the sequence is predefined. But I think then we cut the capabilities of the AI. So I think it's very hard to find something in between.

What do we want AI to be capable of doing? At the very moment, there's discussions at the highest political levels globally what it is that we want AI to be capable of doing, right? Okay, understand, but maybe still, so when we talk agent, can you be a little bit more precise? Can you share with the type of agents, maybe a name? So are we talking, you know, foundation models? Are they also maybe open source? Could they also be closed source? what is this thing that I have typically always been calling my personal assistant because that's what I believe in. I believe, you know, even in a couple of years from now, I'm just going to be saying, telling things to this agent, assistant, and receiving things from my assistant doesn't matter in my private life or in my work life as well. So what is that?

Can you share one, two, three names of tools or platforms? I'm sorry. Yeah, my example, it was very easy. I just used my VS Code, and in the VS Code, I used a chat, which was connected to an LLM. And I talked to them and said, okay, please, robot, pick the part. And then the agent was able to discover the MCP server and to execute this command. Very good. Let's come to a close by talking about what we started in the beginning. You built your own humanoid robot, I understand. So we've learned where you've been coming from, industrial robotics. You're building. You say you need in your free time, you need to move into physical AI because that's where kind of everybody within our space is moving. So with the experience that you have had, but also looking into or combining it with the solutions that you have been building

with your colleagues for 30 years, You know the industrial space very well. And then what is then the use, the view, you believe, of humanoid, of humanoid robotics, of the different modularities? Are these then the walking like a human type of robotics? Are they going to be in wheels? How do you see that space coming from the industrial robotics? Where do you see solutions there that make sense? Yeah, I think first we will see wheel-based robots in the industrial environment because, yeah, it's easier to have the safety, to fulfill the safety with a stable wheel-based system instead of a balancing system like a leg-based humanoid robot. I think that's very clear to me. Maybe later in future we also see systems with legs, but I think the first luminoid introduction will be this with wheels.

And that absolutely makes sense because our factories are built for forklifts, for AGVs and so on. And so the environment is there so they can directly interact in this environment. So I think that's basically the first thing to do. So you mentioned the forklifts, the AGVs. Yes, I see exactly, well, at least, let's say, in the newer, larger factories with a typical shining concrete floor, right? You know, we sometimes say it's so clean you could eat from it, right? And those factories that we know, and not maybe the small ones that have come out of age and are very old and maybe where that doesn't work. But you mentioned the AGVs, the forklifts. Of course, the providers of those logistics solutions, they come to market by putting AI solutions, agent solutions, onto their machines, onto their vehicles is the better word as well.

That's, of course, one solution. How do you see that? in relation to the companies, in the meantime, we have how many? I don't know, not only 100, maybe 500 companies across the world that all come with their humanoid solutions. Do you see one with a better advantage than the other one? Who do you think is going to be the first major players to be successful in that world? I think what we will see first, because the humanoids are currently do not have the safety, I think that we will see first, yeah, air robots with a cobot on it. Because that is something that already is safe and which you can directly bring into the production. I think this is, for now, that is the solution. But I'm very sure that we also will see the humanoids in the production. And we already see them in test environments. And I think it's very important that we now starting with trying things out with the humanoids,

because I think in two, three or five years, the humanoids will come, and then we need all the experience, how the things work, what are good use cases, what are bad use cases, how to solve problems, how to reteach or relearn algorithms. And we need this knowledge to be competitive in the future. And maybe we're soon going to see your robot in an industrial environment, Torsen. Oh, I think my robot is not capable to move in the industrial environment because it's built from cheap components because it was a hobby project. Yeah, it looks like an industrial one because I have all the emergency stops and things which are familiar in the automation world. That's what I recognize, yeah, right. But I think I would never be able to interact in the industrial. But I hope maybe at some time it can bring me a beer.

The most important thing, bringing a beer to a man. Nevertheless, this is your version number one. I think it's a wonderful thought. I said a similar thing a week or two ago for any of you listeners. You know, I think you and then doing this thing even in your spare time with having, you know, the knowledge of 30 years being a software engineer in an industrial environment and then having this drive, this feeling, I want to do this and do it in your personal time. I think it's a wonderful thing. And, you know, I think you're kind of driving your own future by doing that. Right. and I think that's a wonderful idea. Maybe by closing off, Torsten, tell us about a typical day in the life of Torsten Brach. That's a difficult question because there is no typical day in my job.

But yeah, I would say that's also the nice thing in my job. So it's not everything the same. It's not boring. For example, at the beginning of this week, I was sitting in the car and was driving to a customer because I had a problem with the interface between the PLC and the robot because there was a new firmware available and there was breaking changes. So, yeah, I immediately had to sit in the car and move there. So that was one topic of my job. But then the other day, I sit in the office and programming some components for our framework. And the next day, I'm discussing solutions with customers. So, yeah, it's a wide variety of tasks to do. And, yeah, it's a lot of fun. And so I really enjoy my job. And as you know, I also take a little bit of my job home and do the things I kind of do in my business hour. I do the things in my free time. Perfect. Thank you very much, Thorsten. Yes, we can hear your drive to be doing all these wonderful things.

Thank you very much for having shared your experience. I think it's very important that we, in this industrial world in which we are, that we think of the potential of doing new things with agents, with AI, maybe with certain types of humanoids as we hear from you. And you say, yes, they are, they will come. Not any of them, but specific types first and maybe the other ones later. thank you very much for having shared that with us and looking forward to see robot version 2 or 3 by Thorsten Brach making it open source for everybody to use thank you very much thanks Peter, thanks for having me bye bye you