Inside humanoid robotics: software and semiconductors podcast transcript
Dale Tutt: Welcome to the Industry Forward Podcast, the show where we examine industry and technology trends with the help of various experts from around Siemens and beyond. My name is Dale Tutt. I’m the Group Vice President of Industries at Siemens Digital Industry Software and your host for today’s podcast.
We are often asked why now with respect to the explosion and companies that are developing humanoid robots. And a primary driver has been the convergence of miniaturized electronics, software-defined products, and semiconductors, plus new battery technology that allows for greater mobility for these robots. And for today’s podcast, we are welcoming Michael Munsey to the show as we discuss the critical role that software-defined products and semiconductors are playing in engineering humanoid robots. So hi, Michael. Please introduce yourself to the audience.
Michael Munsey: Hello, everyone. I’m Michael Munsey. I’m the Vice President of the Semiconductor Industry here at Siemens. And besides being actively involved in all things semiconductors, I often get involved in the adjacencies of semiconductors, such as software-defined products, humanoid robots, which we’ll discuss also, that are themselves software-defined products as well. I’ve been in this industry for over thirty-five years.
Dale Tutt: Alright, well, welcome to the show, Michael, and I am really looking forward to the discussion today that is really going to cover the inner workings of the humanoid robots as tools. delving into the importance of their software architecture and their semiconductor devices. So one of the things that we’ve discussed on some previous episodes of the Industry Forward podcast, especially with Rahul Garg a few episodes ago, that we were talking about like, why now? What’s going on with humanoid robots? And I think it’s always good to set the context that we’re starting to see humanoid robots show up in many places where into environments that are designed for humans.
So if you’re in a manufacturing floor that the products like the car and then the production processes, the assembly of the car, has been designed around people that if you want to put automation in those spaces, that it’s easier to put humanoid robot into that space. But yeah, we’ve been talking about humanoids for a long time. You’ve been on cartoons and they’ve been in movies and we’ve never really seen them come forward, but now we’re starting to see a lot of companies working on them. And as I said earlier, it’s the convergence of the new technologies around semiconductors and the software and electronics and the new sensors, all the sensors that are available with physical AI.
And a big part of that, and also in batteries, just gives them mobility and some endurance. But I think a big part of this is going to be what’s going to really help them come out is the software. We’ve talked about in some other episodes about the training of the humanoids, that they are delivered with basically an operating system, and this is how you walk and this is how you grab things, and that’s about 40% of the software. And then there’s another 40% that’s specific training, and then another 20% that’s really unique to the specific operations that you’re asking them to do.
So software is huge, and semiconductor is a really big part of this. So I think, you know, when we talk about software-defined vehicles and software-defined products, and other industries, many would go to describe humanoids as a software-defined product. So Michael, what are your thoughts on this?
Michael Munsey: Let’s start with what a software-defined product is first, right? Most people think, well, my laptop or my iPhone’s a software-defined product because it runs software. And that’s, of course, a very big piece of it, right? Anything that runs any type of software applications are, by default, software-defined products. But it’s what’s underneath the hood that’s probably even more important, right? You have inside these products a semiconductor, because the semiconductor runs the software, but it also controls the operations of the entire product. And there are different levels of software that control how the actual semiconductor devices and other electronic devices inside the product actually operate. And these types of things tend to also be software upgradable.
So if you think about your mobile phone, there’s a microphone on there, and that microphone has active noise canceling. Well, when the vendor releases new versions of software, they will often improve the algorithms to do the noise canceling. So that microphone actually gets better over time. So the microphone itself also is part of a larger system that’s software-defined, but it itself is software-defined and also software upgradable.
So when we talk about software-defined products, we need to really understand it’s more than just running software. It’s about controlling the operation of the devices as well via software and being able to improve the product over time via software. And that’s exactly where humanoids fit in, right? Because, yes, eventually you’re going to go onto Amazon, order your humanoid, and it’s going to show up in the box, you go up in the box, and there’s some base level capability in there, right? But the manufacturer has no idea what it’s going to be used for ultimately.
So there’s going to be some amount of programming. And because of that use model that you’re effectively going to use that humanoid robot for, you may want to eventually improve the actual electromechanical capabilities and the motion and what it does over time. And therefore, you have to make sure that those processors that are inside that humanoid have the ability to actually improve over time, also to improve the overall operation of the humanoid.
Dale Tutt: I always think software-defined products is a very interesting topic, and I think sometimes people tend to take it for granted. And you think about your car that’s getting over the air updates, and all of a sudden you have new functionality, and they didn’t have to physically change the car. Our car, we recently got a recall notice. And in the past, you’d have to take it in, and they’d have to replace some of the hardware, maybe do some wiring changes.
But now they’re able to just load the software remotely, and it resolves the notice and the recall notification, and so I think for a lot of people when they think about software-defined products, it is easy to think about, well, it’s running software, but no, it’s actually not. It’s how you’re enabling new capabilities and new features in the long run. I think humanoids, obviously, exactly this, because you’re going to deliver hundreds of these into a factory.
And when you want to give it a new capability, you’re not going to want to spend a bunch of time going out into some guy with a little toolkit on the shop floor and replacing parts on the humanoid. You’re going to want to just download new software to it. And so, to me, is the epitome of that definition. So do you have any additional thoughts on that, how this definition fits the humanoid robots, at least in that context of what I was just describing there?
Michael Munsey: Absolutely, because at least people that of our age, right, will remember like Rosie from the Jetsons, right? There’s, you know, the idea of what they want the humanoid robot to do, the cooking, the cleaning, right? Taking care of the kids and having a great sense of sarcasm, right? So, you know, that’s people’s like first impression, right? And the key thing here is that you’re balancing creating a product that is capable of doing many things without you even being able to perceive everything that the end user’s going to want to do, right? And the big issue around this is how do you make sure that what you’re delivering is going to ultimately be capable of what the end user wants it to do, right?
And in system-level design, there’s always been this discussion of with more and more software defining the user experience, defining the product, how do you create the compute platforms that ultimately are going to be able to run all this software. And one of the examples I always like to give is that if you’re on your cell phone and you drop a call or you’re on your laptop and you get the little spinning wheel, that’s an inconvenience, right? If you’re in your car and all of a sudden the software workload inside of the car locks up the processor, now you’re in a very dangerous situation where you could potentially kill somebody, right?
So the key thing here is how you make sure that those Compute platforms that are going to be embedded in these humanoid robots are going to be able to handle the software workload necessary. There’s no shortage of videos on the internet right now where somebody’s all set to show off their great new humanoid robot, and they’ve spent all the time on optimizing the software to do the task that they want to demonstrate. And then somebody accidentally bumps into the robot, and the robot loses all of its control and falls over, right? So besides getting a good laugh out of it, shows you can create so much software to make sure the thing is perfect at doing one thing. But then you lose the ability for it to do other things, like stand up or move forward or function in the way that you want to do.
And ultimately, these humanoids are going to be tasked with multiple things to do. We’ve already mastered the robot that does one thing, right, and one thing only on assembly lines. That’s not what you want a humanoid to do. You want the humanoids to be able to do multiple tasks. and learn from what they’re doing and optimize themselves and teach themselves. And so that’s going to be the key thing is what you need to be able to do to make sure that these humanoids have the right processing platform inside of them to be able to handle multiple tasks, upgrade easily, to make it more efficient, and be able to quickly switch between different tasks without worrying about just basic operations breaking down.
Dale Tutt: Yeah, and I think that’s a really good point, what you were talking about. And I love that example of the robot falling off the stage or falling on the stage. I would probably phrase this a little differently. You want to teach the humanoid robots how to do all the basic stuff, actually do the tasks that you want it to do. But it also has to start to develop like that underlaying, I’m going to say that like this awareness of the surroundings that okay, I’m waiting to do something, I’m waiting to do my task, but here’s something coming towards me, I need to step aside to prevent falling down on the stage or getting bumped over. Or if you or I are working in an environment and we see that there’s a power cord setting across the floor, and as we’re walking along, we know to step over it, not to step into it and then trip over it.
And so there’s that layer of Basic operation of how it actually interacts with the environment, independent of what it’s doing, and then to be able to prioritize, Oh, I’m falling down the stairs, so I should maybe let go of what I’m carrying. That’s going to take some learning in how we do this, and so before we move on to some further discussions… What are companies doing to help set that priority? And how does maybe AI and the learning that comes with that, how do you think that’s playing into this with the software? And you’ve got all this compute power and how is it making the decisions about like, I should worry about this and not about that? Does that make sense?
Michael Munsey: Yeah, and that makes perfect sense. The key thing there is being able to verify what you want the system to do long before you ever write the first line of software or write the first line of code to develop the processing platform, right? And so There’s a lot that could be learned from traditional systems level engineering when you think about it, because, you know, these are complex systems, humanoids, robots, right? And there’s a lot that we’ve learned from systems engineering, from, you know, building planes, building automobiles, building like advanced defense platforms, right? That could be applied to the same development of the humanoid robot, right?
The idea here is to make sure that you cover as many of the key important cases as possible, right? So that you could set the right guardrails and make the right decisions between what you want done in software and what you want done in hardware. More in software, the more powerful the hardware is going to need, right? But also the less reactive it’s going to be, right?
So you need to play off a lot of these, what I need done, equivalent to the human manual, dexterity is probably the word I’m looking for, right? That would be important, right? We have good methodologies for doing this now, but this is where AI starts to come in, because AI now will help us. realize that we might have missed something along the way, right? we might have missed an important corner case or edge scenario that should have been considered when we did the system-level analysis. And at the same time, help us weed out some of the more mundane things that we shouldn’t be concerned with. And probably most importantly, find the overlaps where, you know, we may be considering four very large use cases, but 80% of it is common.
And if 80% of it’s common that are not really for independent use cases, right? And the fact that in our EDA tools that we use to design the chips and in our software development platforms now, that there are more and more tools to help us not replace the way we do the design and the analysis and the verification, but actually on augment what we do and help us realize what we’ve missed or what oversights that there were, that that’s really helping us to now create these really complex systems right now, which we have to put into, ultimately, these humanoids.
Dale Tutt: What’s clear from this discussion is that humanoid robots are fundamentally software‑defined systems that are enabled by powerful semiconductor platforms and intelligent software architectures. Their success depends not just on teaching them how to perform tasks, but on giving them the computational foundation to prioritize, adapt and operate safely in dynamic, real‑world environments.
As humanoids take on more responsibility such as switching between tasks and responding to unexpected events, the role of systems engineering and robust compute platforms becomes even more critical.
Thanks for joining us on Industry Forward. If you enjoyed this conversation, be sure to subscribe for future episodes as we continue exploring the technologies driving the next wave of industrial innovation.
Siemens Digital Industries Software helps organizations of all sizes digitally transform using software, hardware and services from the Siemens Xcelerator business platform. Siemens’ software and the comprehensive digital twin enable companies to optimize their design, engineering and manufacturing processes to turn today’s ideas into the sustainable products of the future. From chips to entire systems, from product to process, across all industries. Siemens Digital Industries Software – Accelerating transformation.
