# The End-to-End Physical AI Journey: From Simulation to Production (Presented by Wandelbots)

## Abstract

Key takeaways Hear directly from Schaeffler about their journey from traditional automation to a software-defined manufacturing strategy for the era of physical AI. Learn why future production environments require a technology-agnostic automation strategy capable of orchestrating industrial robots, mobile robots, and future humanoids within a unified automation ecosystem. Discover how Wandelbots NOVA provides the software-defined robotics platform that harmonizes diverse robot vendors and form factors, creating a common execution layer that connects AI, simulation, and real-world automation. Understand how manufacturers leverage digital twins, synthetic data, and NVIDIA AI technologies to accelerate robot training, improve adaptability, and move physical AI applications from simulation into production. Gain practical insights into the organizational and technology shifts required to scale physical AI from isolated pilots to enterprise-wide, end-to-end manufacturing operations.

## AI Summary

_No AI summary available._

## Transcript

**[00:00:05 – 00:00:08]** Good afternoon, everyone, and welcome to our session

**[00:00:08 – 00:00:10]** about physical AI.

**[00:00:10 – 00:00:14]** I think all of you have experienced over the last two days what

**[00:00:14 – 00:00:18]** we all have in front of us if we are believing in the

**[00:00:18 – 00:00:20]** change technology can bring.

**[00:00:20 – 00:00:23]** And I think it's wonderful to see how the whole ecosystem is coming

**[00:00:23 – 00:00:28]** together to allow you to understand what can be reality already today,

**[00:00:28 – 00:00:32]** because physical AI is not anymore just a hype or a trend topic.

**[00:00:32 – 00:00:36]** It has become reality and I'm very very happy to share

**[00:00:36 – 00:00:40]** the stage today with our colleagues from Schaeffler who are going

**[00:00:40 – 00:00:44]** to give you an introduction of how they have experienced their journey

**[00:00:44 – 00:00:46]** and what they are going to do next.

**[00:00:46 – 00:00:50]** We are a company which is focusing on establishing a

**[00:00:50 – 00:00:54]** software environment which enables physical AI for all

**[00:00:54 – 00:00:58]** machines in automation from all the different types of robots

**[00:00:58 – 00:01:02]** you are going to experience in the future because we strongly believe

**[00:01:02 – 00:01:07]** it's going to be a zoo of robotics you are going to have to manage.

**[00:01:07 – 00:01:08]** But it's not just the robots.

**[00:01:08 – 00:01:10]** It's as well the machines around it.

**[00:01:10 – 00:01:13]** It's the sensors, the different vision systems, everything

**[00:01:14 – 00:01:15]** you have to think of.

**[00:01:15 – 00:01:19]** And we are laying a foundation to create a standard, a standard

**[00:01:19 – 00:01:24]** which is allowing you to not just bring physical AI in one single use

**[00:01:24 – 00:01:28]** case, but actually in all of your different automation use cases,

**[00:01:28 – 00:01:32]** beyond just a single location, for a global standard, for

**[00:01:32 – 00:01:35]** a global governance of physical AI.

**[00:01:35 – 00:01:42]** For us you have to consider that type of technology enhancement

**[00:01:42 – 00:01:45]** from the whole life cycle of automation projects, from

**[00:01:45 – 00:01:49]** when you are starting to envision how automation can actually

**[00:01:49 – 00:01:53]** be applied in your environments to how you are going to operate

**[00:01:53 – 00:01:55]** it into your day-to-day life.

**[00:01:55 – 00:01:59]** And we as a company are bringing that software layer which

**[00:01:59 – 00:02:01]** is allowing that all of those different technology worlds

**[00:02:01 – 00:02:05]** can communicate with each other and can be leveraged with each other.

**[00:02:05 – 00:02:08]** But it's not just us making physical AI a reality.

**[00:02:08 – 00:02:11]** It has to have as well a deep integration in

**[00:02:11 – 00:02:13]** simulation platforms.

**[00:02:13 – 00:02:18]** And we are going to see what is possible based on NVIDIA Omniverse

**[00:02:18 – 00:02:20]** environments like iSEC SIM.

**[00:02:20 – 00:02:23]** But simulation alone is not the answer either.

**[00:02:23 – 00:02:26]** It's a very, very crucial part of our future.

**[00:02:26 – 00:02:30]** But it needs to be enhanced as well with artificial intelligence.

**[00:02:30 – 00:02:33]** And that artificial intelligence needs to be thought as well,

**[00:02:33 – 00:02:36]** the whole through along the automation lifecycle.

**[00:02:36 – 00:02:41]** So AI models like Groot, embedding them from the planning up

**[00:02:41 – 00:02:45]** to the day two operations, and really interacting in a continuous

**[00:02:45 – 00:02:51]** loop of data is making physical AI a reality because only this way you

**[00:02:51 – 00:02:55]** are going to have the foundation to understand what happens

**[00:02:55 – 00:03:00]** in your physical world and allow it to think about it and make sense

**[00:03:00 – 00:03:04]** out of it in the digital world and bring it back and vice versa.

**[00:03:04 – 00:03:09]** For us, physical AI is not just a production topic.

**[00:03:09 – 00:03:12]** It is a topic which we have to think through your whole

**[00:03:12 – 00:03:15]** business units of your value chain.

**[00:03:15 – 00:03:18]** If you have laboratories, they can be automated.

**[00:03:18 – 00:03:21]** If you have production, of course, that's where we all come from.

**[00:03:21 – 00:03:25]** That is where automation really became powerful with robotics.

**[00:03:25 – 00:03:27]** But it's as well the topics of intralogistics,

**[00:03:27 – 00:03:29]** warehousing, you name it.

**[00:03:29 – 00:03:33]** And maybe beyond what you are finding in your walls

**[00:03:33 – 00:03:38]** of a factory, because if you have connectivity, you can

**[00:03:38 – 00:03:41]** bring the robots anywhere and they can as well leverage physical

**[00:03:41 – 00:03:45]** AI in open space environments.

**[00:03:45 – 00:03:48]** I'm very happy to not just talk about what we are doing and

**[00:03:48 – 00:03:52]** give you an understanding of, hey, that's Wandelbots, we are great,

**[00:03:52 – 00:03:56]** because we don't mean anything if companies are not leveraging

**[00:03:56 – 00:03:58]** the technology in the right way.

**[00:03:58 – 00:04:02]** So I'm happy to introduce you to Sebastian from Schaeffler,

**[00:04:02 – 00:04:04]** who's going to share what they have accomplished over

**[00:04:04 – 00:04:08]** the last years, what a technology partnership means with Wandelbots,

**[00:04:08 – 00:04:12]** and what can be as well, yeah, make possible if you believe

**[00:04:12 – 00:04:14]** in the future of technologies.

**[00:04:14 – 00:04:15]** So, thank you.

**[00:04:15 – 00:04:18]** Yeah, thank you, Katerina.

**[00:04:19 – 00:04:22]** Yeah, so thank you very much.

**[00:04:22 – 00:04:25]** I think some of you know from Schaeffler, we want to play

**[00:04:25 – 00:04:28]** a very important role in the field of physical AI.

**[00:04:28 – 00:04:33]** On the one hand, we are a supplier of components for humanoids.

**[00:04:33 – 00:04:36]** On the other hand, we want to deploy physical AI in our

**[00:04:36 – 00:04:38]** more than 100 plants.

**[00:04:38 – 00:04:41]** And I would like to show you a sneak peek how we already

**[00:04:41 – 00:04:44]** use physical AI in our environment.

**[00:04:44 – 00:04:50]** AI starts in the virtual environment, not in the

**[00:04:50 – 00:04:55]** real environment; therefore, process simulation, how we plan

**[00:04:55 – 00:04:59]** a line, how we simulate a line, will start in the virtual world.

**[00:04:59 – 00:05:04]** So we can already doing layout planning, optimize footprint

**[00:05:04 – 00:05:10]** planning, and also to decide which are the right robots

**[00:05:10 – 00:05:14]** for the right use case already in a virtual environment.

**[00:05:14 – 00:05:18]** We can simulate the cycle time, we can simulate the

**[00:05:18 – 00:05:24]** reachability of the robot and this makes us very fast in order

**[00:05:24 – 00:05:26]** to decide what is the right setup.

**[00:05:26 – 00:05:28]** But what is now new?

**[00:05:28 – 00:05:32]** And you know, physical AI means also that we're considering

**[00:05:32 – 00:05:36]** already physical conditions in the virtual environment.

**[00:05:36 – 00:05:41]** And we can now start to train robots to doing really

**[00:05:41 – 00:05:43]** heavy things, complex things already, everybody, in a

**[00:05:43 – 00:05:45]** virtual environment.

**[00:05:45 – 00:05:50]** And this enables us the collaboration with Wandelbots

**[00:05:50 – 00:05:54]** and also by using the toolchain of NVIDIA.

**[00:05:54 – 00:05:58]** And you know all these SIM to real gap, and trust me, this gap is

**[00:05:58 – 00:06:01]** getting closer, closer and closer.

**[00:06:01 – 00:06:02]** What does it mean?

**[00:06:02 – 00:06:05]** And you're experts; you see it immediately.

**[00:06:05 – 00:06:08]** But for somebody who does not know what it is, it's

**[00:06:08 – 00:06:13]** very hard to decide what is real and what is virtual.

**[00:06:13 – 00:06:18]** And we are now, with Wandelbots, put the first application

**[00:06:18 – 00:06:22]** in our production environment, which was completely simulated and

**[00:06:22 – 00:06:24]** trained in a virtual environment.

**[00:06:24 – 00:06:30]** It's about an inverter system where we assemble different components.

**[00:06:30 – 00:06:34]** And if you have a very closer look to the video, especially

**[00:06:34 – 00:06:39]** if we assemble now the components here, you can see the friction,

**[00:06:39 – 00:06:44]** you can see how to mount the pins on the tooling.

**[00:06:44 – 00:06:49]** And this is the enabler for us in order to consider the real

**[00:06:49 – 00:06:51]** application already in a virtual environment, because you all

**[00:06:51 – 00:06:54]** know where the real effort starts.

**[00:06:54 – 00:06:56]** It starts with the commissioning.

**[00:06:56 – 00:06:58]** And the commissioning is there where we generate

**[00:06:58 – 00:07:00]** the money and the efforts.

**[00:07:00 – 00:07:05]** And we want to reduce it by using this technology.

**[00:07:05 – 00:07:08]** So what we are currently seeing was everything more or less

**[00:07:08 – 00:07:12]** rule-based, but now we come in the direction to enable

**[00:07:12 – 00:07:14]** physical AI to develop skills.

**[00:07:14 – 00:07:18]** And we are using Wandelbots and also the toolchain for

**[00:07:18 – 00:07:24]** NVIDIA now to starting to develop the right skills for robotics.

**[00:07:24 – 00:07:29]** And here you see an example of how we train a dual arm system

**[00:07:29 – 00:07:31]** to handle a box by teleoperation.

**[00:07:32 – 00:07:35]** We can teleoperate the robot, but we can also teleoperate the

**[00:07:35 – 00:07:39]** virtual robot, so we can collect data in the virtual environment,

**[00:07:39 – 00:07:44]** we can collect data in the real environment, we can label this data

**[00:07:44 – 00:07:51]** and then decide how we apply this data to develop the right skills.

**[00:07:51 – 00:07:55]** And for us, we don't want to develop a skill for a dedicated use

**[00:07:55 – 00:07:57]** case for a dedicated embodiment.

**[00:07:57 – 00:08:01]** No, we want to have skills which we can orchestrate to

**[00:08:01 – 00:08:05]** different embodiments, because if I need a packaging workplace,

**[00:08:05 – 00:08:07]** I don't need a humanoid.

**[00:08:07 – 00:08:11]** But if I have 10 tech packaging workplaces, I need mobility.

**[00:08:11 – 00:08:15]** And then I want to develop, deploy, orchestrate the

**[00:08:15 – 00:08:17]** skill on our humanoid.

**[00:08:17 – 00:08:19]** And this is exactly what we are now doing.

**[00:08:19 – 00:08:22]** You see that what we did with the dual LAN system, we trained the

**[00:08:22 – 00:08:27]** skill and together with Wandelbots, we were able to orchestrate

**[00:08:27 – 00:08:31]** the skill on another embodiment, here on the example of a humanoid.

**[00:08:31 – 00:08:34]** And this is exactly the way we would like to go, to have

**[00:08:34 – 00:08:37]** an agnostic skill library and to deploy these kind of

**[00:08:37 – 00:08:42]** skills on different embodiments which fits best to our use cases.

**[00:08:44 – 00:08:45]** A short road map of what we are doing together with

**[00:08:46 – 00:08:48]** Wandelbots and NVIDIA here.

**[00:08:48 – 00:08:52]** First of all, we want to build up our foundation, our end-to-end

**[00:08:52 – 00:08:56]** physical AI pipeline, how we collect data, how we label

**[00:08:56 – 00:09:00]** data, how we use that data for model skill training,

**[00:09:00 – 00:09:02]** and how we orchestrate this data to the embodiments.

**[00:09:02 – 00:09:05]** On the next steps, coming from the rule-based approach,

**[00:09:05 – 00:09:09]** more to the skill-based approach, to have a skill library for

**[00:09:09 – 00:09:14]** our use cases along the complete value chain of our productions.

**[00:09:14 – 00:09:18]** At the end of the day, Physical AI, what counts for the plant

**[00:09:18 – 00:09:20]** managers, they look on their costs.

**[00:09:20 – 00:09:24]** And when we are able to reduce the cost by using Physical

**[00:09:24 – 00:09:26]** AI, then that's the best commitment that we get from them.

**[00:09:26 – 00:09:31]** And exactly what we do, we want to empower our people in the plants to

**[00:09:31 – 00:09:35]** work with this technology by using the Wandelbot platform, not to

**[00:09:35 – 00:09:39]** only be experts, that everybody of us are able to use the technology

**[00:09:39 – 00:09:47]** In order to open the complete And for that, I want to hand over again

**[00:09:47 – 00:09:50]** to you, Katharina, and we're really looking forward together with you

**[00:09:50 – 00:09:57]** to walk the path to make Physical AI successful in the industry.

**[00:10:01 – 00:10:05]** So you have seen here right now one of the leading companies

**[00:10:05 – 00:10:10]** which had the courage to bring new technology into their operation.

**[00:10:10 – 00:10:14]** But it's not just about integrating technology.

**[00:10:14 – 00:10:18]** When we are starting to really change and really drive that

**[00:10:18 – 00:10:23]** disruption, it's about to rethink how you are operating actually

**[00:10:23 – 00:10:25]** automation in your environment.

**[00:10:25 – 00:10:28]** Because the technology is here, but it's a huge change

**[00:10:28 – 00:10:32]** management project to really make the people which are

**[00:10:32 – 00:10:36]** working for you, which have worked for the past decades in a specific

**[00:10:36 – 00:10:40]** way, to allow them to understand what's possible if you are going

**[00:10:40 – 00:10:43]** more in a DevOps type of setup.

**[00:10:43 – 00:10:46]** Because we are moving away from the V-model, and if you

**[00:10:46 – 00:10:51]** want to make Physical AI a reality, it has to be DevOps-based.

**[00:10:51 – 00:10:55]** And in addition, it's not just that one perfect physical

**[00:10:55 – 00:10:58]** AI platform architecture which fits for everyone.

**[00:10:58 – 00:11:01]** You have to consider what components you are going to

**[00:11:01 – 00:11:05]** bring together, what AI models you are going to leverage,

**[00:11:05 – 00:11:08]** and how you make sure it's becoming your physical AI platform.

**[00:11:08 – 00:11:13]** And I'm really happy that you guys started that path because there are

**[00:11:13 – 00:11:14]** not so many blueprints out there.

**[00:11:14 – 00:11:17]** They are not the standards which we have grew up with

**[00:11:17 – 00:11:21]** in the IT landscape or OT landscape from the past decades.

**[00:11:21 – 00:11:24]** But there are companies like Schaeffler who are starting

**[00:11:24 – 00:11:27]** to walk that path without waiting for the perfect future,

**[00:11:27 – 00:11:31]** for the perfect setup, and just create their own proof points.

**[00:11:31 – 00:11:36]** And so I can just encourage every single one of you, the time is now.

**[00:11:37 – 00:11:40]** Physical AI is not anymore a hype or a trend topic.

**[00:11:40 – 00:11:44]** You can start leveraging those architectures, those technologies

**[00:11:44 – 00:11:48]** today and make sure that from an industrial perspective we are

**[00:11:48 – 00:11:50]** really becoming competitive again.

**[00:11:50 – 00:11:53]** Because we don't have to wait for another region, for another

**[00:11:53 – 00:11:55]** someone to create something big.

**[00:11:55 – 00:11:58]** We have that knowledge, we have the assets and we have

**[00:11:58 – 00:12:01]** the right ecosystem here at that location who are helping

**[00:12:01 – 00:12:05]** you to achieve what Schaeffler has already started to prove.

**[00:12:05 – 00:12:08]** So thank you very much for joining our session today, and if there

**[00:12:08 – 00:12:12]** are questions, we are available on the site, and wishing you

**[00:12:12 – 00:12:14]** a very nice remaining time at GTC.