"We have around 30 to 40 tests, like can we clean the surface? Does the robot emit particles? Is it biological resistant? How does it behave with temperature, for example, to make sure that the robot can be safely and cleanly operated in a clean room." - Werner Kraus
Werner Kraus leads robotics research at Fraunhofer IPA, where 1,000 engineers work on production systems and the unglamorous infrastructure that makes humanoids commercially viable: standards. Not the headline-grabbing demos, but the 30-40 tests required before a robot can enter a semiconductor clean room. The certification processes that determine if particle emissions from gear grease will contaminate pharmaceuticals. The biomechanical measurements proving that 500 Newtons of collision force is 3x too high for human safety.
Standards enable commercialization. Without them, you can't get insurance. You can't satisfy business buyers. You can't scale beyond pilot programs. Werner's team at Fraunhofer IPA does the methodical work of defining what "safe" and "clean" actually mean in measurable terms—then working with ISO committees and robot manufacturers to close the gaps between current performance and certification requirements.
I sat down with Werner to explore how clean room certification works across nine ISO classes, why the Unitree G1 currently achieves ISO Class 5 (semiconductor standard), what 500 Newtons of collision force actually means for human safety, and the surprising discovery his team made about data transmission to China when testing cybersecurity at the Humanoids Summit 2025, hosted and organized by ALM Ventures at the Computer History Museum in Mountain View, California.
Werner walks through the reality that robots themselves are contamination sources—emitting particles from wear, harboring dirt in crevices, and requiring steel surfaces instead of coatings in pharma environments. He explains why energy efficiency standards matter (the Unitree consumes 280W and requires battery swaps every 100 minutes), how ISO/TS 15066 sets different force limits for different body regions, and why his team consults with manufacturers on design changes to improve safety ratings rather than simply issuing pass/fail grades.
Clean room certification, collision biomechanics, ISO standards development, cybersecurity testing, energy consumption benchmarking—top concepts covered. But what struck me most was the systematic rigor: not rushing humanoids to market, but methodically defining the thresholds that protect both workers and product integrity. Testing until the measurements are repeatable. Publishing the data so everyone works from the same baseline.
That is a robot future built on engineering discipline rather than hype cycles.
Please join me in welcoming Werner Kraus to Turn the Lens, in collaboration with Humanoids Summit and ALM Ventures.
This interview is a collaboration between Turn the Lens and Humanoids Summit, and was conducted at the Humanoids Summit SV, Computer History Museum, Mountain View, California, December 2025. Humanoids Summit is organized and hosted by ALM Ventures.
Learn more about Humanoids Summit at www.humanoidssummit.com
Guest: Werner Kraus, Head of Robotics, Fraunhofer IPA | Host: Jeff Frick
Date: December 2025 | Location: Humanoids Summit SV, Computer History Museum
Duration: 11:39 | Series: Humanoids Summit 2025 (Episode 10 of 10)
Werner Kraus leads 1,000 engineers at Fraunhofer IPA working on the essential infrastructure enabling humanoid commercialization: standards. This conversation reveals the systematic testing and certification that transforms research prototypes into production-ready systems capable of working safely in clean room environments.
Key Findings from Unitree G1 Testing:
Fraunhofer IPA doesn't issue pass/fail judgments—they provide manufacturers with specific improvement recommendations through iterative testing.
Unitree G1 Test Results:
ISO/TS 15066 Biomechanical Limits: Based on University of Mainz research (100+ subjects, 29 body locations):
30 Gears = 30 Pinch Points: Beyond collision forces, articulation points create pinching hazards for fingers, clothing, or tools.
Solution Path: Werner emphasizes need for "new kind of sensors which keep the distance between human and the robot so that there's no collision appears at all"—moving from force limiting to collision prevention.
Standards Contribution: Findings feed into ISO working group on "dynamically stable robots" (met twice in 2025 at Boston Dynamics and PAL Robotics), developing standards specific to humanoid bipedal systems vs. traditional fixed-base robots.
The Counterintuitive Reality: Jeff's assumption: "If robots get into the clean room and they're clean, don't they stay clean?"
Werner's response: "That's not, in fact, true."
Robots actively generate contamination:
ISO Clean Room Classification (ISO 14644): Nine classes from ISO Class 1 (cleanest, ≤10 particles/m³ at 0.1μm) to Class 9 (least stringent). Each class represents ~10x difference in particle count.
Unitree G1 Achievement: ISO Class 5 (semiconductor standard, ≤100,000 particles/m³ at 0.1μm)
Industries Requiring Clean Room Certification:
Recent Commercial Interest: XPENG-Bayer partnership on humanoids in pharmaceutical manufacturing
Fraunhofer IPA's 30-40 Test Protocol:
Vertical-Specific Requirements:
Iterative Improvement Process:
Werner's comment "it's a story for me, it's own basically" on edge geometry suggests deep research expertise—potential published papers on contamination harboring in robot design features.
Why It Matters for Commercial Deployment: Energy consumption determines operational economics, not just environmental impact.
Unitree G1 Current Performance:
Technology Trajectory:
Automotive Technology Spillover: Werner: "It benefits from cars and everything else in terms of the energy density."
Standards Development: Enables comparable metrics (Watts per kilogram, Watt-hours per task), design optimization targets, and operational cost prediction for purchasers.
The Testing: Jeff: "What is the biggest surprise you didn't expect?"
Werner chose to highlight cybersecurity findings.
Confirmed Discovery: "If you connect the G1 to the internet it start to send data to China. So it's giving a live signal."
Critical Unanswered Questions:
Deployment Security Implications:
Prohibited Use Cases (likely):
Potential Mitigations:
Legal/Regulatory Considerations:
Industry-Wide Context: Not unique to Unitree—similar scrutiny for DJI drones, Huawei telecom equipment, TikTok data sovereignty, Chinese automotive manufacturers.
Emerging Standards Needs:
Not Just Pass/Fail: "We are consulting humanoid robot manufacturers as to how they can make their robot safe, for example, that they change the design that it doesn't hurt so much anymore."
The Process:
Example—Clean Room Progression: Start at ISO Class 7 → recommendations on materials, geometry, seals → implement changes → retest, achieve Class 6 → continue toward Class 5
Industry Collaboration: Werner's colleagues contribute to ISO working groups meeting at Boston Dynamics (US) and PAL Robotics (Barcelona)—creating feedback loop where:
Benefits:
ISO/TS 15066:2016 - Collaborative Robots
ISO 10218-1 & 10218-2 - Industrial Robot Safety
ISO 14644 - Clean Room Classification
Emerging Standards Areas:
Force vs. Pressure:
Transient vs. Quasi-Static Contact:
Effective Mass: Total mass contributing to collision force = robot link + payload + end effector + reflected inertia from gear reduction. Often significantly higher than physical mass.
ISO Class Logarithmic Scale: Each class number = ~10x change in particle concentration. Going from Class 7→6 easier than 2→1 (exponentially increasing cost/complexity).
Network Air-Gapping: Physical isolation from internet—no wireless, no network cables. Prevents remote attacks and data exfiltration but eliminates remote monitoring/support.
Fraunhofer IPA (www.ipa.fraunhofer.de/en) Stuttgart, Germany | 1,000+ engineers | Production systems, manufacturing engineering, robotics Parent: Fraunhofer-Gesellschaft (Europe's largest application-oriented research)
ISO (www.iso.org) Geneva, Switzerland | 169 national standards bodies | 24,000+ international standards Relevant: ISO/TC 299 (Robotics), ISO/TC 209 (Cleanrooms)
Robot Manufacturers Referenced:
Industry Partnership:
Event Context:
Humanoids Summit 2025 Interview Series (Turn the Lens): This is episode 10 of 10, documenting the full spectrum of humanoid development:
Series Themes: AI/software, hardware, human factors, infrastructure, business—covering the inflection point from lab demonstrations to commercial viability.
Werner's Edge Geometry Research: His comment "it's a story for me, it's own basically" suggests published research. Search: "Werner Kraus" + "edge geometry", Fraunhofer IPA publications, ISO technical papers.
Unitree Cybersecurity Deep Dive: What data is transmitted? Can it be disabled? Comparative analysis of Chinese vs. Western manufacturers' telemetry practices.
ISO Working Group Documentation: Meeting summaries, working drafts, technical requirements under debate, timeline to published standard.
High-Density Batteries at Summit: Which companies demonstrated? Energy density figures? Commercial availability timeline?
Clean Room Market Analysis: Total addressable market, automation penetration rates, ROI calculations, XPENG-Bayer partnership details.
Werner Kraus: Clean Room Humanoids, Got Particles? Get Certified | Turn the Lens with Jeff Frick Ep54
English Transcript
© Copyright 2026 Menlo Creek Media, LLC, All Rights Reserved
--
Interview Introduction
Jeff Frick:
Hey, Jeff Frick here. Coming to you from Intel Headquarters RNB, the Robert Noyce building in Santa Clara, California. It's a full circle moment. It's where I started my career in tech way back in 1997.
You know we just had the Super Bowl just down the street a few days ago at Levis Stadium, and the hot Super Bowl ad of 1997 was the Intel bunny people. And they were folks dressed up in their clean suits and they had all different colored clean suits, pink and blue and green sparkly. And they built a whole ad campaign around the clean suits. And in fact, I think I bought some for Christmas back in 97. We'll have to go back and check the photos. It was a really hot item in the employee store back in the day.
It makes a lot of sense here as we release the last of our ten interviews that we did at Humanoids Summit because we had Werner Kraus on. He's the head of robotics from Fraunhofer IPA and he talked about standards and tests and really doing the things that's going to make it capable for robots to start to enter the world of clean room manufacturing.
And I thought, well, you know, if they get into the clean room and they're clean, don't they stay clean if they don't have to leave and go home and eat dinner and see the family every night? And that's not, in fact, true. There's all types of contaminants that can come from the robot itself and cause a problem in a clean room environment. And those are some of the certifications he talked about, you know, potentially stuff coming off the gear with the grease. He talked about, you know, wear parts might have emit certain particles. He talked about cracks and crevices within the outer shape of the robot itself that might harbor you know a higher likelihood of dirt or things getting you know, stuck or hidden in there, even talked about like, what is the surface and what are the surface materials and what are the attributes of those materials for the robots in clean room manufacturing. And the applications are wide. Obviously semi manufacturing which is why I'm still here in front of Intel is an obvious one. Another one is pharma, a growing opportunity for robots in clean room environments. And the third one which I didn't think of right away, but he mentioned is solar panels. So a lot of opportunities for robots to expand in more labor pools where there's a shortage of labor.
As we've heard over and over and over again, standards, getting to standards, coming to standards, adapting to standards is really what enables the broad commercialization. And you got organizations like Fraunhofer IPA that are doing the dirty work to sit down and actually define these things and write them down so that everybody knows what they should and shouldn't do. It's a great episode. I hope you've enjoyed all these episodes from Humanoids Summit. I certainly did. I learned a ton with every single one. I hope you did too. Thanks for watching. Thanks for listening on the podcast. We'll see you next time. Take care. Bye bye.
–
Main Interview Cold Open
Fraunhofer IPA sounds like a beer.
Yes. Yeah.
And you're from Germany. Do they have IPAs in Germany or is that against the German purity law?
Exactly. Against the German law.
I don't like IPA anyway.
Jeff Frick:
Hey welcome back everybody. Jeff Frick here, coming to you from the Computer History Museum in Mountain View, California for the second Humanoids Summit. It's actually the third one. They had one in the summertime in London [2025] but this is the second one here in Mountain View. And it's amazing how fast the space is changing. And the excitement in the air. I think it's tripled in the number of people. And they announced they're headed to Japan next summer [Tokyo, May 2026]. So a lot of exciting stuff happening in robotics. So many factors that are still being sorted out and worked out. And one of them is standards. And you've got to have standards because ultimately you're going to have to have insurance and you're going to have to have business people buy off up on this stuff. So you got to have standards, and we're excited to have our next guest who's got a thousand people working just on defining all the different types of standards that they might have to worry about. So welcoming in he's Werner Kraus, Head of Robotics for Fraunhofer IPA. Werner great to see you.
Werner Kraus:
Thank you, Jeff, for having me.
Jeff Frick:
Absolutely. So you said you have like a thousand people just working on standards. It's a big, big problem. How are you organizing the kind of the classifications of different standards?
Werner Kraus:
Yes. We have 1,000 people working at Fraunhofer IPA for production, engineering and manufacturing. So while we are implementing new solutions on the shop floor and we are also creating the standards, for example in the area of clean room, AI certification, and also safety standards where we are closely collaborating with ISO and also making the investigations in our labs, basically.
Jeff Frick:
Okay. So one of the things I learned last year at this conference was collisions. Basically when a robot hits something it's called a collision. And we don't want a robot knocking grandma down. We know humanoids are going to be in homes and out working with people versus an industrial robot that's behind the safety. So with collisions you showed me some stuff before we got on. You don't only think about collisions in terms of the contact, but the angle of the contact where it hits in the body. And you've got a whole different set of rules depending on even what part of the body before you get into the forces. Share a little bit about how you're sorting this and organizing it and how it's going to make humanoids ultimately safer to be near people.
Werner Kraus:
Yeah, definitely. It's a big topic as the ultimate goal of a humanoid robot is it's working close together with us humans. And, so we have a Unitree G1 and checked all kind of safety standards we know from the collaborative robot, but also from service robots. So, for example the robot has thirty gears and joints. It may also pinch people if you get in your finger. So there's also a question about the radius for example. The locomotion one is quite well. So it also on uneven places it does not fall. But we also made then also some collision tests for example so the when the robot is moving down it can easily reach 500 Newton of forces. And this is much, too much when we compare it to the current ISO standard and also when the robot is colliding while moving we also see that the collision forces are quite high and that a robot which is only 1.35 meter high [53” or 4’ 5”] and 35 kilograms [77 lbs]. So there needs also and it clearly it sees that we need some more safety methods in the humanoids.
Jeff Frick:
So is the goal that those standards then will be implemented as a standard. For instance, you can certify that your robot when it has a collision will not have more than whatever the designated amount of forces and then how is that implemented in all these different robotic form factors?
Werner Kraus:
Yes. Good question. On the one hand side, we are contributing this, findings also to the current ISO working group on dynamically stable robot, where a colleague of mine is also contributing. They met two time this year already at Boston Dynamics and PAL Robotics in Barcelona. On the other hand side we are consulting humanoid robot manufacturers as to how they can make their robot safe, for example, that they change the design that it doesn't hurt so much anymore. But ultimately, I guess we need really some new kind of sensors which keep the distance between human and the robot so that there's no collision appears at all.
Jeff Frick:
Yeah. Now, you showed me before we got on some of the categories. One of the ones I would've physical safety is an obvious one.
Werner Kraus:
Yeah.
Jeff Frick:
but, Cleanliness.
Werner Kraus:
Yeah.
Jeff Frick:
So what are you judging robots on cleanliness? What are the factors and why is that important?
Werner Kraus:
Yes. That's we need cleanliness when we are operating robots for example in clean rooms for chip production. When we are manufacturing solar panels also in the pharmaceutical industries and just recently in China, XPENG revealed a partnership with Bayer the pharmaceutical company, on humanoids in the pharma industry. And there we have to check and ensure that the robot doesn't emit particles because otherwise these particles from the grease or from wear would come into the pharmaceuticals for example. And therefore we have around 30 to 40 tests, like can we clean the surface? Does the robot emit particles? Is it biological resistant? How does it behave with temperature, for example, to make sure that the robot can be safely and cleanly operated in a clean room.
Jeff Frick: Is the threshold of engineering standards so way, way higher for something like that than something that doesn't have to, I mean when you told me like I'm like well of course it's clean it could be clean and they can keep it in the clean room. It doesn't have to go out to its family at the end of the day. And as you said it's also emitting and creating its own problems.
Werner Kraus: Yeah.
Jeff Frick:
for some of these issues can come. is that much, much harder in the clean room environment than say our everyday things that we see over here in the expo hall, is so would be a really special purpose build.
Werner Kraus:
Yes, definitely. As well you have to really look into which vertical you want to go and then also the surface and the covering of the robot will change. And so for example that the coating is not dissolved for example. So it will be have a steel surface is one example. We also have to look into the edges. Are there any corners where some dust could be inside? It's a story for me, it's own basically.
Jeff Frick:
Right. And is that a go no-go decision on whether the robot passes for a clean room or is there gradients of okay or is it 'No' you cannot go in or you can go it.
Werner Kraus:
There are actually nine ISO classes as well from 1 to 9.
Jeff Frick:
Okay.
Werner Kraus:
One is the cleanest worldwide. And the first finding showed that there's something like ISO five would be the Unitree which is a standard also in semiconductor industries, for example. And it's really a typical process that we are, checking the robot and giving then advice how to improve the design of the robot to reach a higher cleanliness class.
Jeff Frick:
Yeah. Interesting. Another bucket is energy efficiency. So why energy efficiency? It seems like that would be very highly variable depending on the form factor and the objective. So is it like energy density. What are you coming up with standards on energy?
Werner Kraus:
Yeah. Yeah, also the energy consumption is of highly relevant when we are thinking of operation in production because it says, okay, by when I have to change charge the battery or to change or swap the battery. And we have found out that the Unitree has around 280W is the energy consumption. and this turns out to be something in a normal scenario that every one hour forty you have to swap the battery. But also on the Summit we have seen some further developments so high density batteries. So I guess four hours apart already possible. Maybe also eight hours a complete shift.
Jeff Frick:
Right, right.
Werner Kraus:
Yeah.
Jeff Frick:
Yeah, I mean, because the battery is a whole different technology, right?
Werner Kraus:
Yeah.
Jeff Frick:
It benefits from cars and everything else in terms of the energy density.
Werner Kraus:
Exactly.
Jeff Frick:
Yeah. So, what is the biggest surprise as you've been going through this on some of these standards that you didn't expect?
Werner Kraus:
Actually, we also looked into one point regarding the cyber security. There's also some rumors that the Unitree is sending data out and we also could see that if you connect the G1 to the internet it start to send data to China. So it's giving a live signal. So it was true.
Jeff Frick:
Yeah. Not just a rumor.
Werner Kraus:
Yeah, yeah. indeed.
Jeff Frick:
All right Werner, well it's very exciting. It's just amazing on how many different levels you know you're defining all these different standards. It's very, very complex and very, very detailed to make sure that we get these things in a position where we can have them out and about with people.
Werner Kraus:
Yes, exactly. Thank you Jeff.
Jeff Frick:
All right. Thank you. He's Werner, I'm Jeff, you're watching Humanoids Summit. Thanks for watching. Catch you next time. Take care.
—--
COLD CLOSE
Alright, clear
Awesome.
Thank you.
Thank you.
Werner Kraus: Clean Room Humanoids, Got Particles? Get Certified | Turn the Lens with Jeff Frick Ep54
English Transcript
© Copyright 2026 Menlo Creek Media, LLC, All Rights Reserved
Jeff Frick has helped tens of thousands of executives share their story.
Disclaimer and Disclosure
All products, product names, companies, logos, names, brands, service names, technologies, trademarks, registered trademarks, and registered trademarks (collectively, *identifiers) are the property of their respective owners. All *identifiers used are for identification and illustrative purposes only.
Use of these *identifiers does not imply endorsement.
Other trademarks are trade names that may be used in this document to refer to either the entities claiming the marks and/or names of their products and are the property of their respective owners.
We disclaim proprietary interest in the marks and names of others.
No representation is made or warranty given as to their content.
The user assumes all risks of use.
© Copyright 2026 Menlo Creek Media, LLC, All Rights Reserved
