Expert interview: when are humanoid robots ready for the factory floor?

0:09 Kevin Maximilian Kron: Humanoid robots are not ready to be deployed today. I think if we would have the perfect humanoid today on the hardware and software side, most companies will still take three to five years to really roll out the technology. The humanoid will not be the best machine for every single use case. Spot is basically an autonomous data collection platform. In the past, as Boston Dynamics, we did a few mistakes releasing Spots maybe too early to too many customers. The usual customer on our Spot has an ROI of like 18 months roughly. At some point, they're not willing to babysit every single project. They expect it to just work.

0:47 Interviewer: Spot or Atlas: which would you take home for a weekend?

0:50 Kevin Maximilian Kron: Atlas.

0:51 Interviewer: Four fingers or five?

0:52 Kevin Maximilian Kron: If I'm targeting the household, five looks good. If I want a reliable product, I would choose four.

0:57 Interviewer: Humanoids on a European production line by 2028: myth or reality?

1:00 Kevin Maximilian Kron: That's a myth.

1:00 Interviewer: Physical AI in 2026, in one word?

1:02 Kevin Maximilian Kron: Just deployable.

1:08 Interviewer: What do European customers ask about first, and how has that changed since 2021?

1:12 Kevin Maximilian Kron: Yeah, that's a good question. We had, I think, a bit of a paradigm shift. Back in the day, around 2021, it was more about feasibility. Can the robot walk? Can it navigate autonomously? Can it handle the terrain? How often does it get stuck? You had these basic questions where you needed a proof of technology to actually show that this technology could be applied.

1:37 Kevin Maximilian Kron: Now, today, it's a bit different. Today we have a few thousand Spots in the field actually performing work. And so these early-on questions, it's easy now to point to these references of customers scaling, customers proving that it can work, that there is enough business case behind operations. So it's more about integration, security and ROI. So we went through this shift that the customers are now asking rather: Can we integrate that into our CMMS system or into our Maximo system, for example? And that's about the technological feasibility.

2:18 Interviewer: What are customers using Spot for?

2:23 Kevin Maximilian Kron: So Spot is basically an autonomous data collection platform. We're walking from point A to B, we're collecting data, and then we're looking to give as many insights from this data to our customers. So we are doing industrial inspection. We are validating if the customer has any compressed air leakages or gas leakages. We are looking at pumps, motors and bearings failing, and we are reducing the cost for the maintenance of those assets, while also increasing the uptimes and decreasing downtimes. So it's a mix about labor savings, it's about energy savings, optimizing operations. So it's a whole bunch of different things, and every customer looks at both different. For one customer, it doesn't really matter that the fire exit is blocked; another customer also gives this a dollar value. So it's very diverse, and we're actually doing a lot of work to help customers finding the right ROI for their own facility.

3:22 Kevin Maximilian Kron: And with many customers engaging with us back in the day, like in 2021, '22, '23, where they bought robots, our robots were magical walking robots. They went from A to B, but we were missing these capabilities that we have today. And so today a customer will have a totally different experience and a totally different product than three or four years ago. And so we spend also a bunch of time to engage with former customers, people who went on a journey years ago, to actually give them all the value that they maybe hoped for in the past as well.

4:02 Interviewer: What can Spot do today that wasn't possible two years ago?

4:07 Kevin Maximilian Kron: So I would highlight two of my favorite features. One of those is something comparable to a Google Street View of a facility. So you have the robot walk autonomously through a facility. The robot creates a 360-degree image every two meters roughly, and then you can basically check out how did it look like yesterday, a month ago, before the incident. And we paired that with a large language model. So you're just asking a query to your LLM and asking: Was there a leakage somewhere on the ground? Was there a person in an area where no person should be? What's the gauge value reading? So these kind of inspections were just not possible before this entire AI boom.

4:47 Kevin Maximilian Kron: And I think the other one, just recently off the press: We are starting out to do also motion amplification. So we are basically running through a facility, we are recording a video of an asset, and this video will have all the motions amplified. So we basically know if, for example, a screw is loose or a bolt is loose. And so I think these are magical capabilities that are superhuman, that a person by himself cannot do, right? And so we went away from just doing what a person can do to actually deliver more insights of a facility, and making sure that the value is a lot higher than it was before.

5:34 Interviewer: If Atlas were available today, which factory tasks would come first?

5:39 Kevin Maximilian Kron: I think if we would have the perfect humanoid today on the hardware and software side, most companies will still take three to five years to really roll out the technology, because it's not just technology. It's change management, it's compliance, it's security. For most companies, the safety officer has never seen a legged robot, so how should we evaluate an autonomous legged humanoid? So it's a journey that we are taking customers together with us. If a customer will start in three or five years, they will still need these three to five years in order to actually get those deployed.

6:17 Kevin Maximilian Kron: And we are a bit more pragmatic: we have customers who are starting today with Spot. Spot is quite similar: it's also a moving autonomous mobile robot and legged robot. But we are also starting in the next year to do things like roadshows, skill developments, bringing robots into customer facilities. Europe, it will be a bit later compared to the US, just also because of logistics, and issues, or not issues, but you have import-export topics that are quite more difficult. For example, if a hardware would break in Europe, it takes more to fix the hardware in Europe and have the replacement parts here. But with customers, so basically, on one side, we are aiming to be in Europe sometime end of next year. But in the meanwhile we're preparing those customers and then find the right use cases, checking: can they actually scale? What's the impact of these use cases? If it's a use case where you're replacing half an FTE, it's probably not something that you want to automate right now and focus on right now.

7:31 Kevin Maximilian Kron: And one of the topics also: we are trying to move the technology not only into a lab, but actually also factory floors. And it's quite different deploying robotics in the lab versus a factory floor. If I'm deploying a robot in the lab, all these safety, IT topics can get excluded. I don't need an operational robot, and you are missing out on many lessons learned. And so we are trying to be as close to factory floor and shop floor as possible.

8:03 Interviewer: What is Atlas doing in Hyundai's factories today?

8:08 Kevin Maximilian Kron: So the tasks we're focusing on right now in automotive are sequencing, machine tending. So we are mostly in the logistics area within the facilities. We will be moving also into areas of, for example, general assembly. But general assembly, you have takt times, you have more constraints. If the robot fails, you will need more time, or you will lose product and all that. And so some of the areas we're initially focusing on, but will eventually move towards.

8:38 Kevin Maximilian Kron: And so within the Hyundai facilities, we are able to use Hyundai as our customer zero. We are able to make sure the hardware is mature before you're moving it into other customer sites, so that the customer, in the end, does not feel like we are developing hardware on their backs, but that we are coming in and are not doing a science project, but bringing in an operational robot.

9:08 Interviewer: Stretch at Lidl: what took it from pilot to full deployment?

9:12 Kevin Maximilian Kron: I mean, on our scale, you basically have Spot, that is like two or three years ahead of Stretch, and you may have Stretch, that's maybe two or three years ahead of Atlas, right? And with Stretch and Spot, everything's about customer experience. How much friction do these products create? How much value do they enable? How often do they break? What's the intervention times? And so many customers, depending on at what stage they're usually investing new technology, at some point they're not willing to babysit every single project. They expect it to just work. And we are getting there. We are there with Spot, where Spot just works. Stretch also made, in the last years, tremendous steps into this direction. And that is also what's missing for Atlas, where humanoids will go towards the next few years, but where we basically need a few more breakthroughs to get there.

10:18 Kevin Maximilian Kron: And in the meantime, exactly those customers that are scaling are the ones where they see enough value, they see they do not need to babysit the technology, where they align their use case and business case. And also, it's also about sponsoring, right? If we are going to a facility, into a single facility, and I'm talking with the management, it's a very different discussion than when we align with our C-level and C-suite. So if you build the strategic partnership, you're aware technology might fail, that you might need to find a different facility, but not every single facility is suited, and you basically go in with joint expectations. A lot of those things will go wrong.

11:05 Interviewer: What makes Stretch different from other case-handling robots?

11:10 Kevin Maximilian Kron: So for Stretch, I think there's various aspects. When we first, before we built Stretch, we published videos of Atlas, and customers approached us and were like: We need this humanoid for our packaging operations. We told the customers: wait, wait, wait, the humanoid will not be the best machine for every single use case. Let us build you something that's working today or in the near future, and that will enable this use case. And that's how Stretch was born.

11:38 Kevin Maximilian Kron: With Stretch, I mean, it's basically a two- or three-click solution. You drive the robot in front of the docking plate, you press two buttons, and the robot unloads the entire container by itself. You have a telescopic conveyor that's connected to the system. The telescopic conveyor drives into the truck, into the container, when the robot drives in there, and drives out when the robot tells it to. And so it's very low friction, it's very easy to use, and it actually takes away, in these environments, the heavy burden from the operators. So we have operators that are not showing up, because they have packages coming in, with a container coming in where every package weighs 20-plus kilograms, and they are unloading the first 50 and are not showing up anymore. And so that's the reality of the field. And by enabling exactly these workers to utilize the technology to do less straining tasks, you are creating a lot of happy advocates for the technology.

12:44 Kevin Maximilian Kron: And many customers, it does not end with the truck unloading, but you also always have processes downstream, where many companies are also looking to us to help them in the future. Even if Stretch cannot handle that today, we are able to have the conversations with them how Atlas will enable anything else that they need in the future. So it's very much an end-to-end conversation with us and our customers.

13:14 Interviewer: Spot or Atlas: which would you take home for a weekend?

13:16 Kevin Maximilian Kron: Atlas.

13:17 Interviewer: Four fingers or five?

13:18 Kevin Maximilian Kron: If I'm targeting the household, five looks good. If I want a reliable product, I would choose four.

13:24 Interviewer: Backflips or box lifting: which video sold more robots?

13:26 Kevin Maximilian Kron: The box lifting.

13:27 Interviewer: Humanoids on a European production line by 2028: myth or reality?

13:29 Kevin Maximilian Kron: That's a myth.

13:30 Interviewer: The first job Atlas does autonomously in a European factory will be…

13:31 Kevin Maximilian Kron: The heavy, repetitive logistics tasks.

13:34 Interviewer: Physical AI in 2026, in one word?

13:35 Kevin Maximilian Kron: Just deployable.

13:37 Interviewer: Where in Europe is Boston Dynamics growing fastest?

13:41 Kevin Maximilian Kron: So we're doing a larger chunk of our business in Europe. We built our service and support center by Frankfurt in, I think, I want to say, '23, because we realized these customers need a different level of support, right? They need to have tickets answered in the right time zone, they need to get the robots repaired on a shorter time frame, and that's why we invested a lot in local staff.

14:09 Kevin Maximilian Kron: And so right now, the three biggest markets in Europe, I would say, without having looked up the number, but I would say it's Germany, the Netherlands and the UK. And across these countries, any manufacturing company has similar issues, right? Everybody has bearings, pumps, motors that are defective, compressed air lines that are failing, some other gases that are leaking. And so all these things are kind of industry-agnostic. But the industries that are currently, I think, booming the most would be semiconductors. I mean, with the entire AI boom anyways, they have super acid-dense sub-fabs and facilities, and these are very well suited for the technology. Then we have, obviously, automotive. If you are able to reduce downtimes there, it's worth a lot of money, and we don't necessarily have all the redundancies that semiconductor facilities have. And then I think my favorite vertical is actually food and beverage, because you have packing lines, you have product producing lines, and every facility is so different, but yet very similar. And yeah, I'm always excited to explore new logos and new facilities.

15:34 Interviewer: How do you compare Boston Dynamics with lower-cost Chinese robots?

15:38 Kevin Maximilian Kron: I mean, I can dive into these aspects of both Spot and Atlas. On the Spot side, I mean, when we started to deploy Spot in 2020, we did not have this ROI-driven mindset, right? We didn't have a customer-centric approach. The robot did not necessarily create value for the customer, and was high-touch, right? So you needed a team of engineers to keep the robot alive, to make sure the robot does what you want, and the person whose task you are augmenting is still needed and is performing the same task, because the technology is not reliable enough.

16:19 Kevin Maximilian Kron: And I think this is something that we invest a lot of time and effort with: to improve our reliability, to improve our service, our software, develop the features that customers kept asking for. And many of the products, maybe it could be from China, from other countries, you often don't see them following end-to-end a certain use case, but trying to bring an R&D platform on the market, which is low-cost. I mean, Chinese, they master supply chains, they know how to iterate fast, but from, like, the go-to-market approach, they have not developed many of the right toolings. And so for many of our customers on the Spot side, they're not really comparing our robots to different manufacturers, because the value proposition we have and the feature density that we have is not really matched by anybody else. So the usual customer on our Spot has an ROI of like 18 months roughly. And even if our products are at a higher cost, if you achieve the ROI within 18 months and you continue having the ROI in the upcoming years, the total savings will be a lot higher than the investment you might do otherwise. Because if I bring a robot in a European facility, it's topics around integration. I have to talk with the data officer, I have to talk with security, safety, I need to do a risk assessment, I need to work on so many topics hand in hand with the customer. And so it is something that we learned throughout the years, but not many have mastered yet in mobile robotic space.

18:15 Kevin Maximilian Kron: And speaking for Atlas, I mean, humanoid robots are not ready to be deployed today. They don't have the reliability like a FANUC arm or KUKA arm, where the robot just works for 10 years. And to get there, it will take a bit more time. And so I think in the past, as Boston Dynamics, we did a few mistakes releasing Spots maybe too early to too many customers, and then suddenly you support maintenance topics and repair topics around the globe, and suddenly customers want the robots working, want to have their problems fixed. And investing the time and resources to get all this fixed, it consumes you. It consumes a lot of time, it makes you slower. And that's why we chose a very deliberate approach with Atlas, to not just sell the robot into the market, but to make sure it's a mature product that's working, and will not have basically these early issues that hardware usually has. And so I'm also very confident that this is a very good approach that will yield trust and good results of the customers.

19:25 Interviewer: New EU rules are coming: how should manufacturers prepare for a humanoid pilot?

19:31 Kevin Maximilian Kron: We are, I would say, lucky to be involved in many of these conversations for our customers, because many of the customers today, many of those manufacturers, did not do their diligence in the last years to get ready for this regulation. So for many, it's the first time, when we are talking with them, that they're encountering the Data and AI Act, the Cyber Resilience Act and all these things that we have. And so for us, it's quite a lot of hand-holding, having these conversations, guiding their teams and coming to joint conclusions, because many things in this space are still opaque. There is not a right and wrong answer. With, for example, the Data and AI Act, we had recently the omnibus released on top of that, and it is a lot of new things coming. Things might still change. And so it's a constant process of re-evaluation and figuring out what's actually needed.

20:30 Kevin Maximilian Kron: And that being said, I gave this timeline of three to five years to actually roll out and deploy a fleet of robots. And so if I'm doing a POC or pilot today, knowing that the same technology will need to comply with all this regulation coming into place in the next years, I have to think about ensuring this regulation is in place today, so that in three or five years, I'm not surprised by regulation not being fulfilled. So I think it needs to be one of the core requirements for any of these initiatives to make sure that any supplier, any manufacturer of humanoid robots that I might choose has done their diligence and will be compliant by the time this regulation comes into place.

21:24 Kevin Maximilian Kron: And so you have a lot of the physical AI cybersecurity topics. How are you treating machinery within your network? Where are models executed and deployed? I mean, we are having a lot of conversations about: where is data stored? Where are AI models allowed to run? Can you share data outside of Germany? Can you share data outside of Europe? Can you share data with the US? If yes, under which conditions? And so it's so many topics, and I think we can spend a significant amount of time to dive into each of these facets.

22:02 Kevin Maximilian Kron: What we are trying to do today is running readiness assessments. So we are basically talking a customer through all these aspects. We are interviewing their leadership teams to make sure that they will be ready for this technology, and where we reflect, basically, what gaps do these companies have today that they will need to fill for this technology to be ready.