Building and Funding the Next Robotics Wave

Chapters

1. Introduction & Welcome — 00:00:00
2. Panel Introduction & The Four Themes — 00:03:19
3. State of Advanced Autonomy: Hype vs. Reality — 00:05:33
4. Scaling Autonomous Systems: Technology, Regulation & Adoption — 00:19:54
5. Financing Robotics: Venture Capital, Family Offices & Commercialisation — 00:36:47
6. Where Will Value Be Created in the Robotics Stack? — 00:45:59
7. China, Europe & the AI Robotics Race — 00:57:16
8. Humanoid vs. Purpose-Built Robots: Who Wins? — 01:06:02
9. Closing Remarks — 01:14:20

‍

[00:00:13] Massimo:

Good afternoon, everybody. Welcome to Rothschild & Co Bank here in Zurich. My name is Massimo, and I'm responsible for venture services here. Together with Ben Robinson, CEO of Aperture Capital, we're really excited to kick off this Venture Insights 4x4 salon.

‍
For over 200 years, Rothschild & Co has been committed to helping clients preserve and grow their wealth. Today, we're part of one of the world's leading financial independent groups, operating in 50 countries with a team of 4,800 professionals worldwide. Our Swiss presence dates back to 1968, when the British and French families began offering a comprehensive range of wealth management services in Switzerland. Our services today cover end-to-end wealth solutions, spanning discretionary mandates, portfolio advisory, wealth planning, to philanthropy, pension and lending solutions, as well as banking and custody services. Our wealth management business is complemented by Five Arrows, our alternative assets arm and global advisory, which provides independent, corporate M&A, strategy and financing advice. As a family-controlled bank, our strength lies in a solid financial foundation and the freedom to act in our clients' best interests.

‍
With that context, the question in the room might still be, what does that have to do with automation and robotics? The link, I think, can be found in Venture Services. It's a dedicated competence center, dedicated to supporting entrepreneurs and investors proactively with our network. We're deeply embedded in the Swiss innovation ecosystem. We maintain an active network and make connections to expertise where it's needed most. For entrepreneurs, those interactions often encompass strategic discussions about pitching a company and finding access to the right investors. For investors, those discussions often encompass complementary deal flow as well as connections to like-minded investors.

‍
[00:02:13] That work's incredibly interesting, but as you can imagine also, rather manual and not so scalable. We decided to launch a small event series called Venture Insights, of which this is the first. Thank you for being our guinea pigs and getting to know us. The goal really is to bring together curated groups of innovators, entrepreneurs, and investors so you can discuss and we can make many more connections than we would bilaterally. We're delighted to be hosting in our halls these discussions and making sure these connections that were created today also go much further and beyond this afternoon, this evening.

‍
While I'm on the topic, thanks to Ben for taking us on the ride for our 4x4 salon. Thanks to all the speakers for joining us. With that, over to you, Ben. That's all I had to say.

‍
[00:03:06] Ben:

Great. Thank you very much indeed. Thank you everybody for joining us on this very warm afternoon in Zurich. This is a special edition of the 4x4 Virtual Salon because as the name suggests, it's normally recorded online. This is the second time we've recorded it live. Thank you very much Massimo for hosting us for this live edition.

‍
The reason it's called a 4x4 salon is because we cover four topics. We have four speakers who I'll introduce in a second. We cover four polls. We also take at least four questions from the audience. On that last point, I would say please don't be shy if you've got a question. We'll gladly take more than four questions. If you'd like to ask a question at any time, please signal and Massimo will bring you over a microphone.

‍
Our four speakers, I'll quickly introduce them. I'll go anticlockwise. Loic is from a company called Coprod, which he describes as being the Android layer for robotics. By which he means that it's a modular programmable interface into robots. Next, we have Amin, who is from a company called LOXO, which is producing autonomous delivery vehicles. I think I'm right in saying you were the first company to have autonomous vehicles on the road in Europe. Now, you operate in Barcelona, is that right? Thank you. Next we have Gianluca, who was one of the founders of Sevensense, which produced 3D navigation software for robots, a business that he scaled and ultimately sold to ABB in 2024. Last but not least, and thank you very much for stepping in at the last moment, we have Michael Fruh, who is the CEO and founder of F&P Robotics, which develops AI-powered collaborative robots for healthcare, hospitality, and industry.

‍
The four topics that we're going to discuss are around the state of autonomous systems, scaling autonomous systems, funding autonomous systems, and then lastly where we think in the value chain the value will accrue in autonomous systems. Michael, we come to you for the first question. Last in, first question. My question to you is just about really where we are in the hype cycle, if you will. Because, I don't who knows maybe on LinkedIn, but every time we're on LinkedIn, I'm seeing images, videos of dancing robots and those sorts of things. What's real and what's hype? Where are we on the autonomy gap today? What has crossed into production? What is now in productive use when it comes to robotics?

‍
[00:06:26] Michael:

Yeah, I agree with you, you see a lot of things on the internet. But I think robots doing back flips, cleaning your kitchen, bringing a beer to the table, I think all of that is possible in a demo setup.
I think if you record it many times, then at once it will work out. We are in an era where you can do a lot, but you're missing the reliability piece of it. To come from demo to 99.99% reliability, which is required if you go to, let's say, reasonable customers in the industry, life science, healthcare, you need reliability. We can do many demos, but to reach the reproductivity of these effects, we have to solve several challenges, including safety, including precision, including autonomy itself, which is limited to several aspects, for example, movements.

‍
We reached a nice level of demonstration on single functions very well. But if you want to integrate that into a very high reliability human environment, which is usually chaotic, there are so many scenarios that can happen. The robots still fail to cope with all those scenarios.

‍
[00:07:40] Ben:

Luca, do you have a view on which of those thorny issues around perception, reliability, is the toughest to tackle?

‍
[00:07:53] Gianluca:

I think that the problem is a bit more hidden. It's the fact that the state-of-the-art research at the moment cannot achieve a level of reliability. The added value of this advanced autonomous system, we talk about the general embodied AI, physically intelligence that they can do indeed, make your bed or serve you dinner. It's not my word, but I was listening to the VP of Google DeepMind, the VP of research, and she was mentioning that robotics is just a very hard problem. In fact, at the moment the promise of robotics is to automate tasks. An automated task is about a physically cost-efficient task execution over what could be the manual work. We are definitely not there because indeed the reliability is not there. It just costs so much money to just train even for a single operation.

‍
Traditionally, robots have been working in environments where they can produce or work at scale. We're talking about manufacturing, let's say in automotive, most traditionally. There is where the cost of installation of robots is. A dedicated, very precise bespoke programming of robots was then worth the investment, because then it would be spread over a huge batch of production. Now with the general intelligence, this most advanced autonomy, the promise is that we would reduce that cost of transferring skills basically for screwing, using a screwdriver to using a hammer. This is just very difficult at the moment and from a research perspective. It reminds me a bit of the parallel with autonomous driving 10 years ago. It was a hype, and it was always next year, “We're going to see the cars on the street.” Maybe we won't take a decade or more to actually see something real, but probably you're going to go through a similar journey.

‍
[00:10:28] Ben:

That's a brilliant segue to you, Amin, because, why did it take so long to crack autonomous vehicles? Especially when it came to things like trucking, which theoretically has very predictable routes and so on. Do you think there's things that you can learn from the EV experience that can maybe save time and allow robots to skip, build on that learning and maybe get to autonomy faster?

‍
[00:10:57] Amin:

Perfect transition. In Italy, it takes always longer with the experiment. This is a normal cycle, I would say. But you're right, you are there with autonomous driving. We have this decade of research and why it took longer, because if you look to the research, we had in our industry a complete reset in 2020 because we had this something AI that's called transformers Whatever you have been before was reset. You put it in trash, you start from scratch.

‍
Now with this level of intelligence that we can get in AI to transformers, then you can see very good results here coming by the tens of thousands of the cars. In different regions of the world, like China and US and Europe it is waking up and it's coming. What we see is that shift of attention from technology and value to the business case. When it comes to the business case, and I put to the review this case, they are making a lot of sense of generally when we are talking about physical AI. The best thing to solve, if doesn't matter if it's on the road, a robot, the best starting point is repetitive and predictable tasks. Because cooking is not predictable; every time it is something else. In the same with autonomous driving, if you want to do robotaxis, you need to go everywhere. This is the hardest thing. If you want to solve the hardest things first, this is taking time.

‍
But now with the trucking, these are repetitive roads and predictable, but why did it take so long? This has nothing to do with automation technology itself. This is a pure automotive mechanical and production challenge. We need to have a car that is autonomous-ready so that you can put your intelligence apart. When you look what technology ready is available, Class S Mercedes, when this technology comes to a Mercedes truck, we are talking about twenty years.
That's the reason it takes longer when you go to the lower level of technology from the car. Otherwise, the technology is available.

‍
[00:13:11] Ben:

Loic, do you agree with all this? I guess your thesis is a bit that the hard part is just the integration of all these technologies rather than the reliability, the datasets. Is that fair to say or not?

‍
[00:13:26] Loic: Half.

‍
[00:13:27] Ben:

You half agree or that's half? What half is true?

‍
[00:13:29] Loic:

It's both half, yeah. I agree with most points said, but my take is a bit different. My own view is that we lack infrastructure. For any tech, even the roads that we drive on, it's infrastructure. For the phone, Android, the Linux kernel was also this. For the internet, we had for a while cool people in labs in Stanford, very advanced, researchers building mainframes, which is really cool. Doesn't bring much value to you. It doesn't ping your phone and say, "Hi, my friend,” because it's working in a lab environment.

‍
For the tech to diffuse to actual people and to bring value to them, different layers were built up on top of the tech, which was proven to bring value, store data, share bits, whatever you want. With the HTTP protocol, the World Wide Web, which by the way was also launched at CERN, so it's a local thing. You're in the right place to invest. Then this brought more and more value as people could build the compatible bits of technology, which all works together. Then 10 years later, the whole phone infrastructure layers were built up to bring it to the people through a cell phone network.

‍
In my view, what we need is more infrastructure for robots. People are focusing on building one humanoid, the most complex robot in the world that can do everything. But it's just not ready yet. If we work well on what shape of robot has the most value to bring, and how do you build it up, I think that the people who focus on the application layer, so the Ugur or we have YouTube, will be able to focus on the integration of a robot with blocks like Legos.

‍
The development of a simple app that takes you a Roomba, that delivers value. It's in your house. It's not 10 years away, 10 years ago or 20. It's very simple, but it brings value. If you build the right robot for each application that ends up, with good infrastructure you can bring value today. It doesn't need to wait 10 years.

‍
[00:15:30] Ben:

Does everybody share that point of view? I guess what you're saying is that the robots become almost like thin cloud on top. They have a lot of infrastructure. Do you agree with that?

‍
[00:15:43] Michael:

I think it's part of the problem. To find the solution to the problem that you can actually solve. There is the traditional industrial robots behind fences, and then there is the fancy humanoid that can do everything. But in the middle, there is a lot of sweet spots. This is where we're tackling, the problems. We're really defining a still complex but doable problem, and we're solving it using AI and using all the robotics technology that is out there. But the customer is not looking for a robot, they're looking for a solution. How can you solve my problem?

‍
I think to really focus on one, and we're focusing on life science to solve this missing link between different machinery and life science, then you can solve it effectively and reliably and cost-effectively. Because a humanoid that can do just everything is not the best cost-benefit ratio in any case that I've seen so far. Finding the right product for the right problem, I think that's the challenge we have as well in robotics.

‍
[00:16:43] Loic:

Agreed. Maybe people to accept it as well. In Europe, we don't really want robots in our house, in our world. In Asia, they're much more open to this.

‍
[00:16:53] Michael:

Yeah, maybe one sentence to that, if you allow. We started elderly care homes with our mobile robots, so we have a lot of experience using robots with elderly people and nurses. You would expect the elderly people are very afraid of robots, but they were extremely open. They said, "Hey, very cool. I have a boring day, and now comes a fancy, cool robot that can help me." So they were really into it. But their nurses have been much more skeptical. "Okay, what is the robot doing to my people?" And so on. "What is it telling the other person?"

‍
There you have to try out a lot of things to exactly find out where are the problems of the users that you might need to address with communication or with the product itself. Test early to find out these points, really important.

‍
[00:17:36] Gianluca:

My take on the infrastructure is, as a matter of fact I agree that the infrastructure is important. But also I think it's important to invest in infrastructure and to build infrastructure on the pattern of where the infrastructure is needed. If we do the pattern now, what happens with the OEMs? Now you just build a lot of infrastructure, for instance, specifically data centers? It's quite impressive. They saw that there was a clear pattern of what was needed and how to solve it. Clearly, data centers are a huge piece of Lego of the entire AI economy.

‍
If we look at robotics, I think the situation is a bit more fragmented. You can build components that's going to make it easier. You can add services that's going to make it easier. But the fundamental limit of how to address this general intelligence, tell it to write a prompt for a manipulator and tie my shoes, it requires specific research. It's not about engineering, it's about actually figuring out how to do it.

‍
[00:19:04] Loic:

I completely agree. It's the interface layer. What is the interface layer? How do the developers want to use it? What's the most cost-efficient way to do that? The interface is not clear. We're hedging against this by building different layers of interface, very tightly degrees or a bigger subassembly, and trying to sell it and see who wants to buy what. But ideally, we need to figure out what makes sense and what can we build at scale in different tiers: tier three, tier two, tier one suppliers, that the auto industry has done and in very resilience thanks to this. Then downstream, we'll figure out by testing end-to-end layers.

‍
[00:19:41] Ben:

I'm going to put this onto our second topic. I feel like my role is mostly just to time manage. We're going to do these different sections. Just moving on to the second topic, which is around scaling autonomous systems. Amin, I'm going to come to you first. When you started this residence, you're already in Germany, you've been scaling quite successfully. What scales linearly and what scales exponentially? What do you see in terms of economies of scale, unit costs, as you expand?

‍
[00:20:18] Amin:

Indeed, through scale, these few components in our domains. One, the first component is the regulation, because if you are blocked by regulation you cannot do it. But this is the on the box. We have the regulation in Europe for autonomous driving at the European level since 2022. We can put your vehicles everywhere in there. This is given.

‍
The second piece is you have the right car. Who has the capacity to produce what I call these autonomous-ready cars? There are many OEMs. For us, this is a tick on the box. We are now collaborating with one of the largest European OEMs and we are producing the cars according to our instructions. We tell them how to produce cars for our customers.

‍
Then comes to the technology. The scaling is happening here because you have a technology that you are going to put driver safety. When we look to the business case, you are saying, when you deliver, because your main duty is delivery, and then you are releasing safer. The most important thing is the OpEx, not the CapEx. You buy a car, then you do tens of thousands or millions of the parcels. Your main challenge is the OpEx, and the OpEx is mainly in the rich countries like Switzerland and Germany or Scandinavian countries. You are talking about driver safety. There, if you can show that you are better, you can sell virtual rides. We don't change the business model of our customer. They say, “Give me the car." The difference is that there are just few sensors on top, and you can get this still from your favorite OEMs. Tick on the box.

‍
Then it comes about, okay, I need more drivers, and I want to reduce my costs. Because this is a penny competition between the logistic companies. Who is likely to go on as a customer? When we come with a cost cutter of the OpEx, which is like more than 50% of the total cost, then you can reimagine the old system, and then you can even invent the use cases. This is what we are seeing happening because we are giving virtual drivers, so they can imagine as much driving as they want. This is where the savings are the quickest. We give them licenses that they pay for a few months.

‍
Since we started the company, we have been seeing our solution as a replacement of the human. But now we learn that the very large customers are seeing it as a perfect help for the existing drivers. Why? Because they don't find enough drivers. The drivers have more and more pressure. They have to deliver more, and they cannot. Now they are saying, "Okay, this boring job of repetitive routes can be given to these semi-autonomous drivers." And then they keep the cool part of the job, which is mainly the last mile and going to the customer and say hello and so on, and spend time, is kept for the human.

‍
This collaboration, this split between what makes sense for automation, repetitive jobs, and keeping still the human in the loop for the last part or the human interaction, this is something that is working well and for us is realistic.

‍
[00:23:15] Ben:

I want to double-click on that because one of polls we asked in advance on social media was about what's the biggest misconception about autonomy today? People said it was robots replacing humans. What do you project then in terms of going forward in terms of the mix of autonomous vehicles and humans still being involved in driving vehicles or supervising vehicles?

‍
[00:23:45] Amin:

If you look at the commercial part of deliveries, let's take a very large transportation company, out of four drivers that they have hired newly, one leaves the company and the other one is getting retired. This is happening when volume of ecommerce and delivery is continuously increasing because all of us are ordering more and more things online.

‍
For them, it's not about replacing, it's the prosperity. How can I keep my biggest customer, like the Amazon of the world, happy even in the next 10 years? I see I'm losing or I'm not finding new drivers, and they are going to retire and so on. This is totally a misperception of the automation. Automation is not to replace. It's there to help the customer, and the customers keep the growth.

‍
The other thing is that also at beginning when we started the company, we said, "Okay, it's only about reducing the cost, and this is the main selling market." Honestly, now it is the second selling market because they are saying, "Okay, with that we can reimagine completely this whole ecosystem." For example, in our approach we learned that they have, we could show we are driving significantly less with the concept of delivery that we have compared to this human we found. Then you get immediately less pressure from the cities. There are immediately more benefits, more space, less traffic for the city. We can reimagine the industry that hasn’t changed for the last some years, and it has sometimes honestly nothing to do with the cost. Of course, it's an important thing, but it's coming as a second, sometimes even third on prosperity, but efficiency of the assets.

‍
[00:25:34] Ben:

That's it. I also want your commercial vehicles market, but maybe come back to that.

‍
[00:25:44] Gianluca:

I have a slightly different perspective. I have an augmented perspective of this. My premise about automation is basically beating the baseline of human costs. Which means that eventually, maybe not now because now we have limitations in the technology, there is only that much can do, there is only a portion of tasks we can automate and we can still increase productivity, that's how it is now. But if actually the level of autonomy increase, the level of agency also of a physical machine will increase and tasks will be completed 100% by machines, I think the story changes a bit.

‍
It's not also too much of a future perspective. In China there are already the dark factories. Dark factories are factories where you can turn off the lights because there are only robots. There are no people inside. They are truly built specifically for that. It's a very structured environment where all the tasks are executed autonomously. You might say, “We are building a new factory. Nobody's losing a job." But as the economy progresses, I struggle to believe that in the future there will be just a net increase of productivity and there won't be any victim in the market.

‍
I say this as a roboticist and I say this because I think it's for sure it's going to impact. I believe that in the few years from now it's going to impact society potentially heavily, but then this becomes a matter of regulators and policymakers to address this.

‍
[00:27:52] Loic:

I agree. If I can defend my generation as I'm probably the youngest one here, we see it today. A lot of my friends have a hard time finding jobs, finishing up EPFL. I think people as time goes on can see by the education level in Switzerland, tertiary education increases every year again and again. People don't want the driving jobs. They don't want the picking jobs at factories. But the reality is people want all these advanced jobs, there's not enough of them.

‍
I think that there is some value to be brought by bringing robots. Not the general purpose one that does everything for you, does your code, does everything, no. That does one job specifically, maybe delivery packages, and then use people's time better and free the job not as a delivery driver but, I don't know, an operator of the fleets or a garage robotics. I'm going to the garage robotics fixing scheme to have a really strong resilient system, where you have garages that you fix any robots with standardized parts that come on the shelf, a bit like the car industry, and there'll be a lot of jobs that people fix the robots. But I agree, I think jobs will be impacted.

‍
[00:29:02] Amin:

I would like to comment. I think if you look at the story of technology and society, I don’t have an example but maybe you can find one today, where we say not in a short period of time, maybe over a long period of time, I don't know one single new technology that killed more jobs instead of create. Don't forget that they had people in all lifts, and they were saying, "Oh, now lifts are autonomous, so all the people in every single building are beyond one people, and before they are going.” Now, the lift industry hires significantly more people for service and so on compared to the number of people we had in the lifts just waiting. The job for every senior is not really needed now for not only could talk about the delivery, the added value of a human is bringing the package, going to the room, knock on the door, hand it over. There's no amount of driving in here.

‍
When you get your Amazon package, you don't care about the drive. You just want to make sure that the guy is handing over to you in the right moment, in the safe way. Or you get up to the telecom industry, we have a lot of people that were exchanging the cables when you call. At that time, we had thousands of people. Okay, now this new routing routers from Cisco is going to keep jobs. How about Cisco? They have ten times more employees. Over time, I don't see generally technology kill jobs. It just transforms them.

‍
[00:30:34] Ben:

If I may, I'm just going to move on from this because I think we could get down a rabbit hole, like do we need a universal basic income and so on.

‍
If we get back to the scaling autonomous systems is a second, you were so successful in Sevensense at leveraging other people's distribution power by embedding your software into other people's robots. Do you think that's the best distribution strategy in this work?

‍
[00:31:08] Gianluca:

Just to clarify, at Sevensense we build the eyes and brains of the applied robots. Basically we give autonomous navigation skills to robots, whether they are for professional cleaning, like cleaning facilities and warehouses and hospitals and schools, or for material handling, could be autonomous forklifts and moving around pallets in factories and warehouses.

‍
We were building the intelligence layer. To some extent, we are component suppliers. Component suppliers don't sell the directly to the end customers. Or at least we decided not to do it because we're retrofitting the intelligence of the robot at the end. The guy that is managing a factory, it's not feasible. It's not possible, as there are many technical complexities. We were selling through the OEMs, the robot manufacturers. This came as natural. Whenever you use that, we could have decided to build in full robots. But then clearly this brings up a different level of challenges also from a capital perspective.

‍
We founded Sevensense and we started to look for capital when autonomous driving was actually getting out of the hype and where hardware was basically the enemy for Venture capital. That's why we decided, let's focus on a software-only company. Once you get one customer, then you get the full distribution. If you are embedded in one of these robot manufacturers, you can leverage their entire customer base, and this brings a very nice multiplier in sales.

‍
[00:33:13] Ben:

I'm going through that. We have, I think a largely investor audience here, so I'm going to ask you an investment question. First of all, tell us in your business, are you full stack? Do you manufacture as well as produce software? How much of the infrastructure is yours?

‍
The second question is, based on the short conversation we had before we started, you said that most of your funding or maybe all of your funding had come from family offices. Is that deliberate because you're trying to find investors, like patient investors, whose time horizon matches with the time it takes to scale a hard asset company?

‍
[00:33:57] Michael:

Our background is at the AI lab of the University of Zurich and from the ETH. We have also a deep tech platform, a robotics platform consisting of a software which controls the whole robot. We have a software stack that really controls everything in the robot, including the autonomy part of it. We have mechatronics, gear concepts.

‍
Then we also had the challenge or the decision, do we build the robots ourselves based on our tech stack, or do we go into OEMs with tech itself? We went the first way because we wanted to test our tech first. During that journey, we also had to figure out how to scale the whole thing, and we did a lot of proof of concepts and lighthouse project with famous companies and so on, and this was good. But we realized we are in a global competitive environment. We are very good in tech, in Zurich tech ecosystem we are really top globally, we could say that. But in terms of scaling, you need muscle, you need power.

‍
Then we teamed up with one of the top 10 robotics companies globally. They have over three hundred thousand of their own robots in the field. They are massive. They sell tens of thousands every year. They now sell our product side by side their product with their name, their brand. Because no one knows F&P Robotics, obviously. We also have zero marketing cost more or less, because we sell through that other partner which has a huge marketing division. They also produce everything themselves; they have sales and service.  That's our way to scale. We earn license fees and so on, on that business. That's our approach to scale.

‍
The second thing about funding, I think, yes, deep tech takes long. We're already twelve years in the game. We are in the beginning a little bit, the testing phase. We tried out, as I said, in many different verticals. Until you really come from prototype to industrialized product, there's quite a way to go and quite capital required. We were lucky enough to find entrepreneurs and family offices that believed our vision, that believed in us as people and in the global market to invest in us, even though they knew that it's not going to be a quick buck, so to say.

‍
This really helped us through difficult times. COVID, for example, where you couldn't go to customers or when the hype cycle somewhat came down or the high valuations came down. They also helped us to survive. This entrepreneurial spirit we are very happy to have. We have a really great shareholder base which is highly supportive of us, and we are also planning to extend the shareholder base in a capital round that we're doing right now.

‍
[00:36:34] Ben:

Excellent. This is a good segue into the third topic, which is around financing robotics. The general premise of this section is almost, and maybe you can all take this as a question, to open up, do you think in general hard tech companies, robotics companies get the capital stack wrong? As in they tend to take too much venture capital or not enough debts, whether it's asset-based lending or government grants, whatever form that might take, project finance, whatever it could be. Is that your sense that in general, that the capital equation is non-optimal? I don't know if anyone wants to start.

‍
[00:37:31] Loic:

Sure. I think we've got quite a particular take on this. Initially I was quite against investors. I wanted to bootstrap through my EQO. It turns out if you do hardware, it's really hard. I don't recommend it. It also turns out that investors with the right one can be a great help to the team, the mission, because they have their own network, which myself at twenty-six I don't have. Then they've seen things in the past and they can pattern match on other industries, which I can't. I can look at it in the past, but it lives through. It’s different.

‍
If you find the right people who can actually help you, in my experience they can really be a great help. It's almost like hiring a team member and not just taking money. If you find the right people, I grew my team to ten people now for free. Actually, they paid me. It's amazing. I think that getting the right investor is, it's true, they pay me.

‍
The right investors I think can really help. I'm convinced of this, and that's where all the LP days in real too. I was in Munich last week attending the LP days of one of my first early investors and they all came afterwards. It was very wholesome asking how they could help as the LPs, so they can see me trying to change the world. They're like, “How can we help you to do that?" And there's many different things that they can do. We here went through other bigger robotics companies to scale the business. That's something which I have no insurance, no contact, no network, no insurance at all. I become quite pro investor for the right ones.

‍
Now debts, I was at the US this Monday, and they will not give me any debts. In your business you don't have analysis. We do all the hardware. Just we were pretty young, not making much revenue, and they will consider thinking eventually for very nice and come with a smile, but debt is at a million of earning. Yeah. I'm taking what I can take, and this is usually equity-based financing.

‍
[00:39:41] Ben:

Okay. Have you taken debt? Because I guess the big difference is you've got assets that are productive and therefore there's an income stream attached to those assets, so you could theoretically leverage those assets. Michael, have you done so?

‍
[00:40:00] Michael:

The best money is customer money. We always try to sell as much as we can as early as we can, because the money is good and you also learn a lot. That's what we try to make revenue from the very beginning. Then after that we increased to equity about 30 million. We haven’t refinanced since foundation. We have also selectively taken debt, but only at a very low portion, because at some point you have to pay debt back, and that can come at a time which maybe is not the best time and not very liquid. I think having a good shareholder cap table with supportive investors for us next to the customer money, which is most important, has been a good strategy, I think.

‍
There are also many opportunities for leasing or venture debt and all of that, but it's very expensive and also a little bit tricky to get. All this financial due diligence, you have to do a lot of paperwork before getting it, and then it's also very costly. For us, it hasn't been the best fit at the moment.

‍
[00:41:05] Ben:

Yeah, do you have a view on this? It sounded like when you were speaking earlier on, it sounded a bit like you almost balanced it. Away from being very hard, having too capital-intensive business to being more of a software business because it was hard to get the right capital set.

‍
[00:41:20] Gianluca:

It was part of the process at the beginning. Eventually, we built our own sensor. We had a multi-camera sensor because we couldn't find it in the market. We had the expertise to make it, and that was the key ingredient to our technology that was making our navigation superior to everything else around. We really didn't want because we were a bit scared about it, and indeed VC didn't like hardware. Now it's the new moat since software is commodity.

‍
I think robotic startups have a long path towards profitability, and banks like profits clearly, because you have to pay back. If you add an additional way of burning money by paying interest, that doesn't fit within the equation.

‍
At Sevensense, the year before the acquisition, we were in the finalist of the Swiss Economic Forum Innovation Award. Because we were selected among the finalists, we got a special stand, and we were able then to get some debt. This is due diligence somehow from experts. But we never really did. Yeah, also the acquisition came short after.

‍
[00:42:47] Ben:

One question. Amin, I see you've leaned a lot into partnerships as a way to help with funding. You've taken equity funding. Any debt funding?

‍
[00:43:00] Amin:

I think generally it’s VC, family office, debts and so on. For me, it's most important to find the right mindset rather than tag the right financial instrument. Because, we do hardware; and if you do hardware and you want to convince VCs that have made money in 2000s with the software, they will never give you money and you're losing time. For us, it's just also how we can be more efficient. We found our sweet spot, a combination of more or less everything, that we have VC investors, but we have also very large family offices that are coming from our domain. Not from the technology cycle, from the use case. They understand why this is needed. For them is they are building their future generations.

‍
Also I think it's depending where you are in your company journey. At the beginning, yes, family office is great. Perfect. But after a certain point of time, you need to know distribution channels. As you said, you need muscles, and you need to go to the larger inclusive companies. For us now, we are choosing that route. We found we were in a much better situation with very large American Fortune 500 companies’ investors rather than a big VC, because with them you can really scale worldwide.

‍
I would say for me, it's just a matter of finding the right mindset. It doesn't matter if it's a VC or it's a family office. You just need a mindset during the ups and downs, and to make sure that the ultimate goal of the company is understood. But also motivation. If you are a type of founders that you are absolutely looking for an exit as soon as possible, then most likely you go for the VC.

‍
But I'm a second-time founder. I sold my previous company to a Fortune 100 company. The exit that I want I'm having the reality plan. It's also a personal question. But I think when it comes to the hardware, they always forget that the still largest companies are first on the hardware. In Switzerland, and in Europe I would say, and socially in that region, we are excellent at hardware. I don't understand why everyone's not investing personally in hardware, because this is the future. The software with all the AI and so on becomes a bit more and more commodity, and at the end of the day you need very high reliability hardware.

‍
I think we are best placed on Earth maybe. I don't know why we consider a lot of VCs, family offices, whatever debts that they are heavily investing hard. I hope I will see.

‍
[00:45:46] Ben:

Okay. We're going to move on to the fourth section. On the surface, the thing that we want to dig into is where in the value chain would this emerge? Loic, make the case for the Android layer. What you were suggesting earlier on is everybody would build on that layer, and therefore it becomes an aggregation point and it creates a lot of value in that layer. But do you think there's a risk that you get squeezed between the component makers on the one side and the application providers on the other?

‍
[00:46:24] Loic:

I agree. The case is pretty easy to be made. Basically, let's keep our focus on what actually brings value to their fans. If you work on the windows, you focus on painting windows. Or painting the walls with a great paint gun that's very straight, that mixes the paint, the paint as well that cleans itself, doesn't ruin the floor. That's the actual value of the company making the paint for us. Making a motor controller, an actuator, a power system is not exactly your best. If you spend so much time working on that layer, it's time and energy and cost not spent working on the painting robot.

‍
We will use our layer, and I'll pray that everyone should use this layer. Why? Because the competitor who's also painting, who is using our layer, maybe has 15% if he's selling a quarter of a robot, and it's a bit more offensive on the pure end of the chain component cost. They will spend a year, two years, three years reinventing the stack, like every company does today when the competitor is beating us. For the robot, for a fraction increase in price, in six months it can be on the market creating money for the customers.

‍
I think at some point, if this does work out, they'll end up coming back and going "This is not fair. They're cheating. They're not building a robot. They're buying some assemblies and just making it." Iterating and again, back to the, interface layer and easily adapting the shape of the robot to find the optimal painting robot. Maybe it's two paint guns, maybe it's a symmetric one that doesn't fall over. Maybe it's not even into the elevator and have the arm stowaway. I don't know. But as long as you're iterating on the hardware, it takes a lot of cost and time to reinvent the whole thing. Either you own it in-house and you're fully integrated and you can do this, which means you build value to have anything else, or you buy some assemblies, and it's never reliable. It's always a jumbled mess of stuff which works and then breaks, and it breaks from time to time. You don't know why because you can't debug this thing. This voltage you can't measure, because you need your own bus and your own sensor. Long story.

‍
We are pretty simple. We all focus on one part on the stack, and it enables companies to focus on their vertical, on their value add, and just use bits that are commoditized. I can keep this work. China will be making the hardware as well. We need to make this open. People who make hardware that's compatible with the stack, great. Maybe it sells, more availability, and if you have some great features, and that's where the second part of the answer of your question, comes in, where lies the value? How do you finance this? I think there's many things that companies need, not to make the robot work. That's going to be in all the products. But you need the robots reliable, certified on the markets, and avoiding all this time of SIL three to four certification, which are hard to go through.

‍
You can pre-certify component assemblies by making some cool Swiss advanced components, which today I think it will take a while for China to make as well. These bits can be sold as a freemium software with a button that you have to see if it works. Plug it in, just works free. We work for the communal good, it's communism. We make the whole robotic stack compatible.

‍
I think the value will come mostly on these advanced features that companies need. That's where you can stay defensible while building this horizontal layer.

‍
[00:49:51] Ben: Oh, good. That's a really charitable answer too.

‍
[00:49:54] Michael:

I just would say that in Switzerland we are very good in managing complexity. We are really good in making a complex world. Sometimes, we squeeze it and then in the end we create fantastic products. When we asked ourselves where are we really good, it's about the technology layer of those autonomous robots. This is our core competence. But then, we add on top of that our engineering skills when it comes to physical AI. We combine our tech stack with the expertise of actually building a robot, and this together really solves a lot of the complexity issues our partners and customers have.

‍
[00:50:30] Ben: You're arguing that things need to be vertically integrated?

‍
[00:50:34] Michael:

Not completely, because it's also very use case specific. I think you have really to pick those elements that then are really needed for the solution. I'm not sure if that's the case that this is always the same things that you need. But I just wanted to add, we outsource production, for example. We are not good at production, so we outsource the whole thing. We are not good in sales and service, so we found a partner; really to focus on your core competence and then really manage the complexity to create a product. That's our take on it.

‍
[00:51:01] Ben:

In your business, the distribution partners you have, what is their relative take compared to yours? The reason I ask that, you don't have to go into specifics, but in the poll that we ran in advance of this meeting people are buying the companies’ value we accrue to those that have the enduring customer relationships. Do your distribution partners who have the ultimate customer relationships, are they generating more value? Is more value accruing to those guys than you as the IDN? Who's, making the most margin? Your distribution partners or you?

‍
[00:51:44] Michael:

Good question. In the end, it's about the absolute margin. I get this question often, what do you do? We could either grow linearly then have a bigger margin per robot. But when you look at the absolute margin, so number of robots sold times the margin, we are much better off and much shorter time going through a distributor partner because you need to reach scale in terms to be cost efficient, and in the end, compete globally and be innovative.
To me, it's a very clear choice. The absolute margin, we are able to increase. I would say margin per robot is smaller.

‍
[00:52:26] Ben: Similar question to you, Gianluca. You sold your business to ABB, and I believe it's now been sold again. Is that because ultimately who wins is the company that has distribution? Why did you sell?

‍
[00:52:41] Gianluca:

We sold a fundamental technology to enable a key line proposition of ABB's offering. Their strategy, given that we had a broad customer portfolio, and they all like our technology and decided to go through with the risking strategy. When it comes to retaining the value, when you then sell and you send it to a distributor, I think that in the end there are two aspects of it. One is what is the market pool, and the other is your suitability. On our case, I could say that our technology was all far to that. It's been always evaluated by multiple robot companies, and they all say "This is the best thing we ever tried."

‍
If I had to say what was one of the challenges we had, it is that the end customers were actually conservative. If you think where mobile robots are, used in automotive, automotive factories that use technology that is twenty years old, when it's automated, if you blow up the line it takes millions an hour. It costs million an hour of liability. They're very careful when they choose new technology. Being the new entrant, there is a bit of a gap. We had a bit of this resistance. But I think that there are cases where even when you are at the bottom of the supply chain, look at NVIDIA, you can definitely make good margins there.

‍
[00:54:28] Ben:

Great. Amin, maybe fairly significantly, if you think about the existing stack, if we call it that, where do you think you might see more or less convergence essentially in that stack?

‍
[00:54:42] Amin:

I think now we are talking about the hardware challenge that can compute your team. If you go to the automotive industry, you have only one car, which we call it YYY. The term was really by production line and it's the Cybertruck. But this technology will go over the production lines of all the car manufacturers.

‍
I think the prediction is that still the biggest margin will remain with the technology because if you are capturing the OpEx and you can only relate, I cut your OpEx by half. I want a big chunk. That's just normal. You have to update system cubic. Where we see a shift right now in this autonomous system is the operational layer. The everyday job, how do people interact, how I charge it, how I tell it to go, and so on. The basic things are underestimated. Now there are companies using billions and billions in operational layer, which is technically not very hard, but it's a smart move. From the investment perspective I think that's a nice opportunity as well.

‍
But this type of technology after maybe five years will become a commodity. What remains, basically at the end of the day back to the basic. Who has the volumes to deliver? It's a volume business. That's also why you see the partnership, sometimes the partnership is always in the intention of bringing the volumes because this is a volume business.

‍
In the last years, you're arguing, "Okay, who has more data?" No one asked the question like, "Which is the company that would make it happen with less data?" But now we start to ask those types of questions. Over the time, when you have optimized everything and there is not a lot of R&D, either you have the volume and can you keep layer worldwide, or you are not. This is where I think we need to go back to this partnership to bring good muscles, but always having this volume mindset, not the functionality.

‍
[00:56:55] Ben: Okay, we're almost out of time. I do have a follow-up question for you, but if there's any question from...? Okay, wonderful. Last question.

‍
[00:57:03] Audience member:

Thank you so much, sir. I'm wondering about, I can do a backflip robot, which we call IDemos, which China does a lot. Where is, in your eyes, China versus the US versus Europe/Switzerland? Because we see a lot more of demos from China, but I'm pretty sure they're not that advanced as they look. Where are we?

‍
[00:57:36] Ben: Who will tackle that?

‍
[00:57:39] Gianluca:

I can give at least one part of the interpretation. I think what China has been showing is truly impressive. I think overall the Chinese New Year festival where all these robots made the headlines, it was truly impressive.
Now, the main point is that's not autonomy. It really depends on what you want. That's great for entertainment. But I think the biggest gap and the biggest misconception for the public opinion is that when they see this humanoid making a backflip and they actually think that, "Oh, if it can do a backflip that I cannot do, then this AI must cook wonderful dishes." No, that's not the case, because it's been trained on an extremely repeatable task to do always the same thing. The intelligence that, at least the one I was referring to, comes from the variety of the environment, comes from the multiplicity of circumstances and disturbances that might affect one's work, which cannot be accounted at the moment in an efficient way.

‍
China is investing tremendously into this, and they have a national program that really pushes into this direction. It's a very brute force approach, and definitely they're doing some progress. But, again, Atlas from Boston Dynamics was doing also backflips a few years back, without using much of deep learning. It was actually entire MPC, a traditional control. Advanced, but still a relatively traditional control technique.

‍
I think it's important. I can also understand it's difficult for the general audience to disambiguate between what is what. But there has been a lot of hype that has been driven by these entertainment robotics. Also an insane amount of money goes into a lab of humanoids and so on in the US, and now they're probably, I would say from a technological perspective they are leading. It's very difficult to say.

‍
I think that the most remarkable example is from Fear AI company in the US, making humanoids, and they have shown that for a week a robot could do the work of parcel handling, which is a fairly simple but is a very long horizon task. That was a new thing. As a response, it was shown that in China they already do it at scale.

‍
But what does this mean? It's also hard. There is a lot of, technological warfare in the sense that it is difficult to really see. But what is clear at least for the time being, there is no robot that if you ask to cook a roti can do it properly.

‍
[01:01:16] Audience member: But you think we can catch up in Europe with states where you're with Silicon Valley or however you want to call it? Like on the autonomous side, not on the demo show?

‍
[01:01:30] Amin:

The autonomous side, the requirements are different in different markets. Because if you say, the production line or reliability is not high, it's accepted culturally. Second, it doesn't cost me to correct errors manually. Then you can do it with a certain quality, and then you impress Europeans and Americans by volume.

‍
But if you dig deeper and you know what are the market requirements for Switzerland or Germany, there they expect you or your system to work in a very highly reliable manner. There, this is a layer that we barely talk in general audience, but we are doing generally in Europe very good. I think there, there's not a lot to catch up, but over the time we are going to also recognize that.

‍
At the end of the day for certainly new states, because of the cost of structure which are very high in Europe, we need a very reliable robot. Therefore, we are going to count more on the reliability than this wow effect doing backflips or cost or low. I think in the different markets we will look differently.

‍
[01:02:53] Loic:

Just if anything, on the hardware, if I may just rephrase the question, from my understanding in the whole autonomous stack and robotics stack, we're more advanced literally here in Zurich. I'm going to sell the RSL and Mark Woodrow's lab and the RI recently. We're more advanced. We're going to, as you said, with Baltics, and the robotics are trained here in Zurich. They're not just doing backflips, they're doing random backflips, and more smoothly too.

‍
What China does, and by the way they're using our stack, the Western stack training the robots, which is mostly NVIDIA at XM, which we also use in-house because we also make our own robotics. It's a Western stack. What they're better at that we cannot compete on at all is the cost of the hardware robots. A humanoid for six grand, no one in the Western world could do that. No one. But we are really good at some things. Typically, I've got to sell some more Swiss companies; Maxim Lawyers is very good. They ship actuators completely. The hexagon humanoid is built with their actuators. The cost is much higher. It's also better from my understanding and what I've been able to see too, but still it's a factor. It's an order magnitude difference, at least.

‍
This I think we will not be able to compete with, of the whole hardware full cycle of bots. If you get to a layer where autonomy is being commoditized and it just works, then it becomes an issue of how much can you scale down the price of the hardware. But today, I think we're not there just yet. We are absolutely ahead on the training side. On the hardware, we're keeping up quite well also because European industries are powering a lot the robots around the world.
Typically the motor controllers, the bits that actually spin the windings, there's a lot of companies in Germany, one big one in Spain, who are selling to companies all over the world who themselves integrated intact inside of humanoids, and these are Western stacks. We're competing hard and working harder to not become irrelevant. Let's not lose hope.

‍
[01:05:01] Michael:

Maybe just one small thing to add. I think I'm not worried about technology. We are fantastic in research and startups here and in the US, and I'm not worried about that. But I'm worried about the supply chain and the supply chain power China has, against all of us. Maintenance is one of the famous examples of rare earth in this kind of thing. China is extremely low cost, yes, that's very true. But also how fast they come from product idea to industrialized scale is incredible, and only China can do that. They have the best production engineers. They have the complete supply chain in their country. I would worry here, that we in Europe and in the US pick up on really reshoring a part of the supply chain and getting this dependence on China down.

‍
[01:05:47] Ben: I think we have another question.

‍
[01:05:49] Audience member:

Hey, if I may, also a VC here. I worked a few years ago with a robotics company, Picking Place, very standardized with post office and so on. Today we had an example of a humanoid that can do the task almost as fast as a human being. Do you think there's still a case by case for robotics and all the money flows in there, or is it only going to be humanoids anymore? Because we don't need the cheaper, somewhat. Not really. Unitree is 8K, you said 6K. Tesla wants 48K, something like this. As an investor, would you go case by case or generalist robotics?

‍
[01:06:38] Gianluca:

Hi, I'm a fan. I believe the following. There is a big difference between software and robotics. It is the real world, and the real world is fragmented. I think that just sometimes it's possibly how overengineered we have to make some solutions that make absolutely no sense. If you look at a specific demo I was talking about, you have a humanoid with hands. It's not using hands. Could have had just the wrists. It was completely 100K appendices that were completely useless for taking or and probably they use it for a few hours. I don't know. There's going to be a process where these robots, the hardware is going to get mature, and then there's going to be a way to scale general purpose robotics, AKA humanoids or human-like robots to an extent that they can cover a very broad set of tasks.

‍
But at the same time, I think there is still a place for very specialized niche applications where you need a very specific shape, where you need a very efficient way of dealing with a task that simply is not necessarily what a human would do. Especially if you have cases where you have fairly large batches of operations, where you can still afford to build a hard bot solution.

‍
[01:08:14] Audience member:

Which is better or cheaper? Because I would argue with the amount of humanoids that Elon Musk at least says are going to be built, the cost will come down. Then the case-by-case robotics where you need, a thousand, but you never get there. Even though he doesn't need legs, but he will have legs also just for taking place.

‍
[01:08:35] Gianluca:

Cars have come down to be very cheap if you think how expensive they were when they were invented. It's very hard to go against this claim. I still believe that the complexity and the over-engineering might at some point need a trade-off, so that they need to actually be better to have something simple, and it's going to be cheaper, it's going to have less maintenance, it's going to be faster at doing the task. That’s what robotics design is about.

‍
[01:09:16] Audience member: I agree. I think in those spaces it'll be a specific, I would agree for the next three to five years. But as an investor, you want to think long term. If the humanoid robot costs 10K…

‍
[01:09:34] Michael:

Maybe I can just add that the technology overlap between humanoids and humanoid-like, is very big. The sensorics, the models, the arms, we use the same things, the same gears, the same that are used in humanoids, but we give it a more functional form factor which makes it much more competitive for a given use case. There are really these two hypotheses. Generalist AI models that can run every humanoid, I don't believe in it. Also a humanoid form factor which solves every problem in the best way because there is such a high technology overlap which can be used similarly efficient or more efficient in a cheaper, easier, safer form factor. That's why I believe these form factors will win.

‍
Read the interviews of, for example, Rodney Brooks. He's a professor from North America, very experienced. He also says the humanoid concept will probably go into more practical form factors using the same technology and AI that is also applied in humanoids. It will take some time. Maybe in some point in time, no one can say when a humanoid will be better, but there are many problems like fingers and tactile and safety, battery, all that stuff, practical real-world problems that can be solved quicker, cheaper, faster based on the same tech in other form factors.

‍
[01:10:54] Audience member:

So your prediction is in seven years, we will have a specialized robot at the post office, just changing the things that Ronald didn't test, and not in humanoid?

‍
[01:11:07] Loic:

I think manipulating test will be much more efficiently done.
For example building or even sweep the floor, you can hire a fleet of humanoids from China, hundreds of them, and give them brooms, and they can all sweep the floor. AI, it's great. Sixty joints, very complex hands sweeping the floor. Does it make sense? Is building it worth it because you won't sell a billion, you'll sell maybe ten thousand? Sure. Or even clean the windows, same thing. You want to put a harness on a humanoid and give him a spritz bottle and have him clean the window. Done. Maybe they'll kick the window and break it, and then you’ve got to fix a window or two. How about building a robot for that application?

‍
This mindset can be applied so many times that even if you have way less volumes, I think using the hardware for the actual purpose brings way more value because it's a whole humanoid or general-purpose robot point of view. It's a bit like the phone ecosystem. Phones have to be in your pockets, so they're limited in size. They have to connect you with the internet, so they have big screens so you can see many pixels. Cool. Robots, people think about it like phones. They're saying we'll build one, have one proper platform, and sell software on top of the hardware. But robots can do many things that phones can't do. They can be any shape and sizes.

‍
Limiting it to one form factor, because you'll be able to make more and benefit from end of the scale faster, makes no sense and I don't think will win. I'm betting my life and my work on the other option. I think if I can debate against myself, where it will win or it could win is the household. Because people can't buy fifty robots, one to iron and one to clean, and they don't have the amount of windows or brick walls to be built, which we can spend all day. But an industry would have five shirts to iron. If it's waste order, who cares? You have five shirts to iron. In this case, having just a simple mobile manipulator makes more sense. I'm against humanoids because in the lab they're dangerous. If you have a child next to the robot and it falls, it kills the child. This is game over. No more humanoids. I think a mobile base with battery as heavy at the bottom of it makes more sense.

‍
In the household, I see more meaning for general purpose because you have only a few things to do for each task, and many tasks. But in the industry, I won't do it.

‍
[01:13:22] Audience member: Agreed as well. Cool.

‍
[01:13:24] Michael:

I just wanted to say, because at the moment humanoids are built on metal and hard gear, so they're very heavy. I believe before humanoids get into every household, we need to have easier hardware, lightweight, let's say with soft robotics things that solves the safety problem. There we have also a long road to go because these humanoids at the moment are very heavy, very energy inefficient. Even for standing they need energy. If they don't have energy, they fall off. These practical things with these electro motors and stuff makes it difficult to bring them into every household, is my belief. I think we need softer, lightweight, different technologies on hardware.

‍
[01:14:07] Ben:

I think all that remains to say is to thank you guys. That was a very interesting panel. A very wide-ranging panel as well. We went way beyond just talking about scaling, autonomy, and so on. We got into geopolitics. We got into the future of humankind.
Thank you very much for your insights, and thank you all for being here in person for this live version of the 4x4 virtual salon. Lastly, thank you, the Rothschild team for hosting us. Thank you very much.

‍

‍