Custom Meta Tags

We Talk IoT podcast ep92 title (LC)

AI Gets Physical: Live From the Floor of electronica 2026

We Talk IoT - Episode 92

We Talk IoT Introduction and embedded podcast episode 92 (LC)

We leave the studio to report live from Avnet City and Avnet Silica’s stand at electronica 2026 in Munich.

One theme ran through almost every conversation: artificial intelligence is leaving the data centre and moving into the machines themselves. Engineering students from TH Rosenheim explain how they are teaching a humanoid robot to assemble parts – and to dance the “Schuhplattler”. At the stand’s demos, Avnet Silica’s experts show what local AI can already do. A low-power Windows IoT on Arm system keeps cool on hot, dusty construction sites. A robot arm with 3D cameras sorts bricks onto pallets, a small model of a warehouse. An old-style phone booth runs several AI models on a single AMD box and prints visitors into the year they dial, without sending a byte to the cloud. And a 50-TOPS AI module drawing under four watts retrofits a legacy industrial PC to spot falls in real time.

To close, Thomas Foj, who runs supplier management, technical solutions and digital innovation at Avnet Silica, names the two forces shaping the industry: AI moving from experiments into real applications, and a surge in data that makes memory the question of the fair. His most compelling use case is trusted physical AI in factory automation and robotics. By the next electronica in 2028, he expects AI to travel from the cloud to the edge, and on into physical machines.

Summary of this week's episode

  • 00:00 - Welcome to Avnet City and Avnet Silica at electronica 2026
  • 01:19 - Yann Stridde: Windows IoT on Arm for hot, dusty environments
  • 03:40 - Thomas Eder: the local AI-powered robotic gift assistant
  • 06:03 - Alex Hyun, DEEPX: vision AI for robotics and factory automation
  • 07:39 - Lisa Rees, Avnet Silica: physical AI in one little box
  • 08:30 - Christoph Rommel, BayStartUP: where startups meet large organisations
  • 09:00 - TH Rosenheim: teaching a humanoid robot to work and to dance
  • 11:44 - Michaël Uyttersprot: an AI time machine in a phone booth
  • 16:53 - Martin Grossen: an AI engine for legacy hardware
  • 23:05 - Trusted physical AI: the most compelling use case
  • 24:01 - Where the industry heads before electronica 2028
  • 25:13 - Wrap-up: what electronica 2026 told us

Related links

Podcast overview (GBL)

See all episodes

From revolutionising water conservation to building smarter cities, each episode of the We Talk IoT podcast brings you the latest intriguing developments in IoT from a range of verticals and topics. Hosted by Stefanie Ruth Heyduck.

Stefanie Ruth Heyduck

Get in touch (CO)

Liked this episode, have feedback or want to suggest a guest speaker?

GET IN TOUCH

We Talk IoT podcast episode 92 transcript (LC)

Transcript from episode sample

Ruth: Today we are not in the studio. I am standing in Hall B5 of the Munich Trade Fair Centre at electronica 2026. And for four days, about three thousand five hundred exhibitors from some sixty countries will fill 18 halls with chips, sensors, power modules and the machines that test them. One theme that runs through almost every aisle this year: artificial intelligence leaving the data centre and moving into the devices themselves.

Behind me is Avnet City. Here, a robot arm picks a gift for each visitor and an old phone booth paints you into the year you dial, and neither of them needs a cloud. Over the next half hour, we ask experts what excites them, look at the problems these demos solve, and ask where the industry goes next.

Welcome to We Talk IoT, where we explore the ideas and impact behind AI-driven tech of the future and how data creates real business opportunities. To stay ahead of the innovation curve, subscribe to our newsletters on the Avnet Silica website. I am your host, Ruth Heyduck.

Start of full transcript

Ruth: So I’m here at the Avnet Silica booth with, uh, Yann Stride. And you will explain to me what the Windows IoT on Arm use case is showing.

Yann: Yes. Hello, Ruth, and welcome. My name is Yann Stride. I’m business development manager at Avnet Silica. I’m responsible for embedded software, and particularly Microsoft and Windows IoT.

Ruth: What use case are you showing to the exhibitors this year?

Yann: So here on the demo that we have here, it’s for applications where you have scenarios where customers are operating in very challenging environments. So environments like on the construction site where it’s getting hot, where you have a lot of dirt as well. So it’s pretty complicated environment and challenging environment, and where you need, so devices where, so that are running smoothly and that are not, uh, requiring lots of energy on it and that can, uh, stay cool. So you keep the, you keep them cool.

Ruth: Amazing. And what real world challenges does this application address?

Yann: It’s particularly the question of the performance per watt, meaning that you have devices who are really reading or running, uh, perfectly. The issue that they have, and the challenge that they have, is the device are getting hot. And on some cases, it’s really not permitted on some so challenging environments like I, uh, described, uh, before. So you need to keep the devices not cold, but so under 30 degrees maximum on the one side. On the other side, you need as well to have a performance per watt with really great because you have devices where you can’t have a high energy consumption on that, and it’s, uh, really exactly what this solution is addressing.

Ruth: And what would need to happen for this use case to be adopted at scale?

Yann: Just so the, the solution that you have here is already on the market. So customers can really so come to us and purchase, purchase a device. It’s already, it’s all ready. It’s running. It’s certified. It’s CRA compliant. Already ready.

Ruth: Terrific. Thank you very much, Yann. Have a great electronica.

Yann: Thank you, Ruth.

Ruth: So I am now at an attractor called the Local AI-Powered Robotic Gift Assistant, and next to me is Thomas Eder. Hi, Thomas. Who are you and what is it that you’re showing at electronica today?

Thomas Eder: Hi, Ruth. I’m Thomas. I’m here for the Robotic Gift Assistant, and here you can see a live application using different kind of cameras, a robotic arm, in order to, to do palletisation. So that’s the, the main use case for that. So, uh, Stereolabs and our partner Advantech, they provided us with those, the, the demo, and the demo itself sh- shows the, the use case of different kind of cameras for different kind of applications. So we have, uh, cameras for the robotic arm in order to control it. So we have, uh, cameras doing the, the virtual fence for security applications and that’s, that’s the main, uh, demo.

Ruth: And what other real-life use cases can I imagine? I’m thinking maybe logistics. Can you explain?

Thomas Eder: Uh, the real-life use case for that is really a, a use case in the warehouse, logistics application, palletisation. So for example, you have different kind of articles. So we have here blue bricks and red brick- bricks, and you want to do, place it on a different kind of pallet. Typically, without a camera, the robotic arm would not be able to do so. So attached with the Stereolabs camera, so it will be able to r- recognise the, the blue bricks and the red bricks, and you can place your order here at th- this station, and he will, uh, let’s say, sort it out for you and place it on this third pallet.

Ruth: Amazing. What would you need to scale this to a larger scale, to a big warehouse?

Thomas Eder: So this works perfectly here because you have the robotic arm coming from Universal Robots. You can really scale it up, so they have also models, this, this model is, is something 60 centimetres high, so you, up to 1.8 metres. So it’s really easy to scale. Also the cameras, you can use it also out in the field, so they are ready to use, so all the AI models are already deployed on that. So you for sure, you have to, let’s say, customise it a little bit, so to speak. Uh, you have all the building blocks here you need also to scale it in the field.

Ruth: Thank you very much.

Thomas Eder: Thanks a lot for the interest. Bye.

Ruth: So maybe, um, would you like to introduce yourself? Who are you and where are you from?

Alex: Mm. My name is Alex, Alex Hyun, so I’m working for DEEPX, which is the AI accelerator company, global top one company. And I’m from Korea. So I’m really exci- uh, enjoying the exhibition with all my partners and the customers together.

Ruth: Is it your first time at electronica?

Alex: Yeah, this is my first time for the, the electronica. So this electronica much, much interesting for me.

Ruth: What technology for you is at the moment creating the most excitement?

Alex: As I said, I’m working on the DEEPX side, which is AI accelerator using the, the vision, the AI. So that relate to the vision and of course the robotics application and then the factory automation. This kind of, uh, AI related, you know, the application are very, very interest for me.

Ruth: Have you already seen something that inspires you, that you would like to take back to your business?

Alex: Of course, yes. Of course, yes. Yeah. This is the first day. And even though the first day, I saw there so many, you know, the opportunity to work with the us and then, and our partners and the customers. So that from this, uh, opportunity, not only for exhibition itself, but also having more communication and the meetings with the partners and the suppliers, even the companies together, which is more and more, you know, interest for me.

Ruth: Terrific. Thank you very much. I hope you have a great time at electronica.

Lisa: What I like is physical AI and, and AI at the e- And you know, what we’re seeing is like, you know, this AI here runs on one little box. You know, the fact that you can have everything, like, so condensed, and six months ago you couldn’t do what we’re doing in this one little box. It’s really, really-

Ruth: It’s incredible, isn’t it?

Lisa: I love it. And so, yes, I haven’t gone further afield to see it, but for me, like, and you know, I, I said to, um, Miguel, “Can you just put a message screen? It’s all happening in this little box.” It’s really cool. So I think that’s a nice one. And even in terms of this robotic arm here, it’s all local. Everything’s done locally, and I think local AI-

Ruth: Incredible how small these things are.

Lisa: …that’s robotics. Yeah, it’s kind of like where everything’s going now. You’ve got that security, you control it in your own environment. I think for me, that’s, that’s what I like. Um, maybe there’s more of it out, out in the rest of the world, I don’t know. But, um.

Ruth: That’s for our visitors to say then, yeah.

Lisa: Yeah, indeed. Cool.

Ruth: Tell me what excites you about, uh, electronica this year?

Christoph: AI, like everybody probably. Well, I’m interested in startups, so there are a few opportunities to see, to find startups. I’m interested in what large organisations do with, together with startups.

Ruth: Terrific. Thank you very much. And your name again was?

Christoph: My name is Christoph Rommel from the company BayStartUP, and we are supporting startups in Bavaria.

Ruth: Thank you very much. Next to me now is standing Roboter, a very special guest at the Avnet Silica booth.

Robot: Servus, ich bin der Roboter.

Ruth: And next to me are three engineers from the TH Rosenheim. D- did you build this?

TH Rosenheim student 1: We didn’t build him. We bought from a, a company, and we are working on software and teaching him how to do the things. Basically, we have the hardware, and now we are using this hardware, yeah, best way as possible. We started with some simple things, so we already teach him how to dance Macarena. Now we are doing the Schuhplattler, something a little harder, and-

Ruth: That’s a very b- difficult Bavarian folk dance, yeah?

TH Rosenheim student 1: W- I try it. I try it, yeah, but i- it’s, it’s hard. We need to, we need to t- to talk with a real dancer to have the data for this, but we’re working on. I think that in some weeks we will have it already. And then we have, uh, Luca. He’s working with BSH and, uh, industrial application of the robot, so something a little different scope. He can talk a little more about what he’s doing.

Luca: So, uh, we train the robot, so he be able to assemble pieces in a factory. So we can use the robot in fields where it’s not safe for humans, and it’s hard to teach because you need a lot of information. So we have the camera in the head of the robot, also cameras in the wrists, and also the information of all the motors of the arms. And we have, we take this information and train an AI model so the robot can be more smart to do that, to do this type of tasks.

Ruth: Very cool. And how happy are you with, uh, what he’s been showing, uh, the last few days?

TH Rosenheim student 1: Honestly, it’s really cool. It’s her first day here, but we were here, uh, yesterday, and it’s always really fun to show to the people what we are doing. It’s nice to see how they really like it, like to interact with the robot, to see, uh, something so, I would say, impressive, like just walking through here. I think that everyone thought that it’s, like, science fiction, and in less than five years it came out of the movies to the real life. So it’s nice to talk with the persons, to explain what we are doing. Yeah, it’s a really good experience overall.

Ruth: And what did, uh, surprise you the most?

TH Rosenheim student 2: I think it was how, how real he looks like, like, like, uh, us humans. Because we are here in the engineers. I’m doing … I studies in, uh, TH Rosenheim, Advanced Industrial Engineering. And, uh, in my life I just saw robots like this one we are seeing here, that they have just, um, an arm and it do tasks, very specific tasks. But then, uh, now we see these technologies into our daily life. If this, like, this could be a future for doing laundry. Washing our-

Ruth: That’s a very good idea.

TH Rosenheim student 2: …do the dishes. Thank you. And, and that’s, that’s something very impressive to see.

Ruth: Cool. Thank you so much.

Ruth: So I am here now with Michaël Uyttersprot from Avnet Silica, and we are standing in front of a time machine. What’s up with, uh, this telephone booth, Michaël?

Michaël: Hello, Ruth, and thank you for having me here. So yeah, we have, uh, indeed, um, what we call a time machine, and, uh, with the time machine, this is like an old-style telephone box where you get in, and when you get in, you have… You can dial a number on an old-style telephone, and you will also see a kind of an avatar on a screen, and what it will do is it will ask you in which year you want to travel. So you dial the year or you say the year, and then you can, um, in fact, decide that you would like to go in that year and to have a picture or a photo that is printed for you, and you can take it at home.

Michaël: So this is in fact simply explained, uh, but of course there is a lot of, uh, technology behind. With this kind of demo, we want to explain to customers that you can implement AI on the edge because all those kind of models, there are several AI models that are used. Those are running on one single device, which is an AMD device, and there’s a complexity of course.

Michaël: So today you have a lot of solutions running in the clouds. You have ChatGPT or, or, or Claude Code, for example, or image generation models. So this is working very well in the cloud, but you need a lot of GPUs for that. This can run maybe on thousand C- uh, GPUs. What we do here is we run this on one single device, which is an AMD device, which includes also a processor and a GPU, and those different models, six different models, runs on it.

Michaël: So you have a language model. You have a vision language model to check if the person is a woman or a man, for example, in front of the camera. Uh, there is also like, uh, a kind of painter, so that creates the image that will be printed on, on the photo. There is also then the model that will check who is in front to take this person, so the face of the person to put it on the photo. So there is a lot of things behind the scene. Finally, then you see the person on the photo in the age that he decided to go.

Ruth: That sounds fun. And what is a real world use case this could be interesting for?

Michaël: Well, this is of course a kind of a game that we implemented. It’s an attractor. It’s very good for a show. The key thing is that this is running locally, so for privacy reasons, security, uh, real time, this is what we see a lot for use cases for customers today. And this can be in many kind of directions. Can be industrial, can be medical, can be consumer. You can imagine, for example, and I take, I take one use case, that you are in front of a shop. It can take a photo of you, and you decide then, for example, that you would like to have other clothes that you see in the shop. So you are dressed in the, in the clothes that you see in the shop, for example. What is interesting again is that this is all running locally, so nothing need to be sent to a third party company, so for privacy reasons, for example.

Michaël: And also this is very fast. You have the photo in 40, 50 seconds. So this is acceptable if someone is standing in front of a window of the shop, for example. Another use case can be just to have a kind of interactive system, like an avatar. It does not need to draw an image, for example, but that you talk with an avatar. Again, the same reason why it can be interesting, it’s because it’s running locally, no internet connection, for privacy reasons, for example. So those kind of, uh, use cases, this is what we see more and more also with customers. And especially if you take a look then how it is implemented, it is taking away all the infrastructure. It’s like a small mini PC you have, and even in the future it will run purely on a very small embedded board that you can easily implement in your own application. We are not there yet, but this is one of the next steps that we will take.

Ruth: And what do you need to scale this to a larger size?

Michaël: This processor is inside a, a, a box. So it’s a, a kind of mini PC box. It’s, uh, the size of a big book. And, uh, everything is in there, so you have the processor, you have also the memory. And in fact, the, the, the trick here to run it fluently is because it has unified memory, so that means it… With this device or on this box, we have 128 gigabyte RAM, and this RAM will be shared between the GPU and the CPU, because you need to have both to run the application. This make it fluent, and in fact, how we do it here with, with the different kind of AI models, up to seven or eight AI models that, that runs together also with an agent that controls everything, it takes about 85 gigabyte.

Ruth:Oh.

Michaël: So there is still some room left, but we see that this is sufficient for the application we, uh, we have here on the, on the booth.

Ruth: Terrific. Thank you very much. I’m standing here with Martin Grossen from Avnet Silica in front of a display saying AI on the Edge. Hi, Martin. H- how are you? How is your electronica going?

Martin: Hi, Ruth. Thank you. Very, very well. So I’m very excited, uh, to be here. Also showing the showcase here together with Microsoft IoT. Yes, my name is Martin Grossen. I’m the director for solutions and software at Avnet Silica, and this week at electronica, we are showing some demos together with Microsoft IoT.

Martin: The title here for this demonstration we are standing just, uh, in front of is Edge AI for IoT. And AI, of course, is a big topic, uh, for OT devices and IoT devices. We have a, a great story to tell here because not every system in the field is ready for AI. And, yeah, there are some, some requirements to have the local AI on the edge devices to make free calculations, um, and make predictions, uh, out of video streams. To submit the video streams to a data centre, it’s too expensive, makes no sense, so we need some AI engines locally.

Martin: We showcase here a, a pretty old system with AMD, with a Ryzen 2000 system that is, uh, ver- very well known in the industry and, uh, also used in millions. But the problem is there is no AI engine inside. So with, uh, M.2 module, we added the AI engine from DEEPX. Now we are able to have a local, pretty strong AI engine with 50 TOPS, and to do the predictions, and then only submits the required data over the internet over the eUICC from, uh, BICS that we have in our portfolio as well.

Ruth: And we are standing in front of a display that is filming us with a camera, and, uh, there are lots of lines and boxes and real-time things happening, and a very small motherboard with a tiny chip on it. Can you explain what, uh, you’re showcasing here?

Martin: Absolutely. So on this DEEPX AI, uh, engine, we run now a model that’s called pose detector. And as you can see, it detects all the people, the, the pose of the people, if they are standing, if they are sitting, and we will get an immediate alert when somebody’s falling down on the floor. So when there’s a, a detection on floor, then we have the alert immediately over the internet on our mobile phones. And you see, okay, when something happening, then we get immediate, uh, the alert.

Martin: This of course, we have a lot of different applications, not only for this, but also, uh, you can hear it also in, in the background, we have a dancing robot here. But also when you have the robot in healthcare, for example, then such an AI model makes absolutely sense to detect when somebody really, uh, fell down to, to alert, uh, somebody. And so this is, uh, only one, one of, uh, of millions applications that we, we can hear, because we can do face detection, we can do object detection, we can do segmentation, uh, with this AI engine. So all is open.

Ruth: Yeah. And the system is really busy because we are on the, uh, electronica booth, and there’s lots of people going around, moving all the time. Some are sitting also, and lots of things happening. But, uh, yeah, the system is just running locally and doing this in real time. So the, the challenge this, uh, solution is overcoming is that you can use it on legacy hardware, and, uh, it doesn’t need much power, I guess.

Martin: Exactly. So this model is very low power, but very effective for 50 teraops. So, so it’s, uh, it’s really below, I think, three or four watts, so that we can add… And also very cost-effective. And to add to existing actual platform, this makes absolute sense also cost-wise.

Ruth: And this also ties in, into the, into the next showcase. We have already talked about the, the one that’s in the cranes. That’s basically then another use case for this.

Martin: Exactly. But with the new platforms, of course, we have normally AI engines already inside the CPU. They have strong AI engines like the Qualcomm one, uh, you talked about, or one that, uh, Michaël showed with the, with this, uh, the telephone box, if we go to the future. Also, it, it runs AI locally, purely locally, no internet connection. And this is what we see from the industry. We don’t want to submit all the data to the internet. We also, from the data privacy side, uh, especially when you are in the healthcare sector, you don’t want to submit everything. You need to do it locally, uh, in a, in a safe environment. You need also the speed. Sometimes when you have the Liebherr cranes, yeah, you need the reaction time. Immediate, not dependent on a, on a connectivity somewhere to a data centre.

Ruth: Thank you so much, Martin. It was great meeting you again.

Martin: Absolutely. Thank you, Ruth.

Ruth: I’m sitting here with Thomas Foj and I’m really very excited to hear what your conclusion of this year’s electronica is. Thomas, maybe would you like the chance to introduce yourself?

Thomas: My name is Thomas Foj. I’m in charge of supplier management, technical solutions and digital innovations, and I’m running this within Avnet Silica.

Ruth: And how has your electronica been so far?

Thomas: Been a great show. Uh, we are here, uh, now for the second day. Uh, yesterday was a very busy day. We had, uh, great interactions with customers, suppliers, and with our partners. And looking forward now for the remaining three days, uh, where we’re gonna be here in Munich.

Ruth: And, uh, which challenges are driving investment and innovation at the moment across industries?

Thomas: If we consider the circumstance today in the market, there are two major drivers which are shaping the industry. The first one is the quick adaption of AI across the industry, that they are moving from tests and experiments more into the area of real applications, which is good because this brings us now forward where we were looking for. But the second one is all the strong growth of data, which ultimately brings us to the circumstance around memory. And this kind of challenge creates obviously a big point for many companies in the entire ecosystem. And the biggest question which we have today at the fair is how long the cycle will last.

Ruth: And what is the most compelling use case you have seen so far at electronica?

Thomas: So for me, the most compelling use case which, uh, we have seen so far, which I have seen so far, is, uh, trusted physical AI, specifically in combination of factory automation and, uh, robotics.

Thomas: We are moving from simply processing, uh, of the data into a complex environment of, uh, the AI systems, which the systems are understanding obviously the ecosystem around decision-making, interacting with the real world. And, uh, this is obviously a, a complex element, specifically when we talk about trust. And the systems need to be secure, reliable, and all that interact in real time, and this is a, uh, great opportunity for the future, but we need obviously to sort it out and, uh, bring the right technologies and the right solutions out for it.

Ruth: And electronica will be returning in two years. Where’s the industry heading, in your opinion?

Thomas: From my perspective, uh, AI will be more integrated into real-time applications, uh, which we see in the market today. So from data centre, from the cloud side, uh, into industrial automation, robots, cobots, even f- uh, predictive maintenance will become, uh, a new era, or will be a new era, uh, when it comes to the future applications. So in other words, we are talking about from cloud AI to edge AI, and finally to physical AI, where we see the transition. This demand will, uh, move, uh, forward with, uh, more processing. We will see obviously more memory in the components, uh, energy efficient power, and this all obviously surrounded with, uh, the ecosystem, uh, requirements from a hardware and software perspective. So from our perspective, uh, the winners, uh, will be in the future the ones who are understanding customer demands and scaling this into sustainable business moving forward.

Ruth: Thank you so much for your time, Thomas. I hope you enjoy the rest of your days at electronica.

Thomas: Thank you very much.

Ruth: That was electronica 2026. Four days, 18 halls, and one question we heard at almost every stand. How do we make AI useful where machines actually work: on the factory floor, in the field, on the road? AI, physical AI, this has been on the tip of every tongue I spoke to these past days. The next electronica opens in 2028, and we will check then which of today’s predictions really held up.

Ruth: Thank you to everyone who stopped between meetings to talk to us, and to the team at Avnet City. Thank you for listening to We Talk IoT. Stay curious and keep innovating.

Ruth: This was Avnet Silica’s We Talk IoT. If you enjoyed this episode, please subscribe and leave a rating. Talk to you soon.

About - individual pages (MM)

About the We Talk IoT Podcast

We Talk IoT is an IoT and smart industry podcast that keeps you up to date with major developments in the world of the internet of things, IIoT, artificial intelligence, and cognitive computing. Our guests are leading industry experts, business professionals, and experienced journalists as they discuss some of today’s hottest tech topics and how they can help boost your bottom line. 

From revolutionising water conservation to building smarter cities, each episode of the We Talk IoT podcast brings you the latest intriguing developments in IoT from a range of verticals and topics.
 
You can listen to the latest episodes right here on this page, or you can follow our IoT podcast anywhere you would usually listen to your podcasts. Follow the We Talk IoT podcast on the following streaming providers where you’ll be notified of all the latest episodes: