Why Oil Refineries Are Hiring Robot Dogs | ANYbotics

Intro You have a YouTube channel and you publish videos of your robot doing cool things. And I was back in the days over 10 years ago when a robot could climb stairs. You're going to immediately 100 thousands of views. We went to events. We'd love to show our robot off, right? That's what happened. And what happened is people would see it and you know, people from the innovation department, it's really forward leaning people. They hit the connection said, huh, interesting. Something that I observe is there's a lot of excitement in what we call building the shovel, right? It feels like, hey, there's a gold diggers time. There's something out there. Everybody knows it's coming and a lot of people are trying to. We don't know how that's going to look like. It's it's too complicated to talk to customers and exactly figure out. But let us build to take layer that we're sure eventually we'll find it's where we'll build the best tech. Introduction to Anybotics and Robotics Journey We are live. Hi, Peter. Welcome big builders. Speaker 1 Hi Gabriela, it's a great pleasure to be here. My name is Peter, the Co founder and CEO of Antibiotics. We're a global company but we started at ETA Zurich. So at in Switzerland we're a spin out and what we do essentially is automating routine inspections in critical infrastructure. We build A4 legged robot that can move by itself, you know, walk anywhere where people can walk. It's fully autonomous. So you install it one that it can do the job itself and ultimately what it does. The robot collect data and the users visual thermocaml acoustics and we'll speak about it much more. But ultimately, it's about, you know, finding problems in critical processes in the industry to avoid downtime, which is very costly, but also keep people out of harm's way. Speaker 2 I want to understand something about you. How did you get into robotics and why quadrupeds? And is it something that you've started at ETH or does your love with robotics? If you're in love with it? I mean, I don't know, right? Ron told me. I think you might you might be does it go back, you know, further than ETH? What's your what's your background story? Where where are you from? And you know, how did your background, you know, lead you to build a small robot dogs? Speaker 1 Sure. Well, big questions here. I'll keep it somewhat concise, but I'm certainly a robot nerd, but not for the robotics sake. I just love building things right already as a kid, like many of us, you know, building the, the, the Lego Meccano building, you know, programming very, very early stage. But I also always loved the creative aspect from design, from from drawing, but photography, video, 3D animations, these kind of things. It's always the interplay between technology and and the creativity. And ultimately I studied mechanical engineering. I already had a sense sense there during one of the visitation days at ETA three, they actually had a little robot that haptic device feedback, It was was amazing. I was like, OK, that's it. Just for the sensation of seeing that machine. I want to understand how to do that. And from there, from my bachelor thesis to the master thesis and eventually the PhD always was focused on robotics and why robotics? And there's many exciting other topics as well, but robotics is so brand new, meaning that you can be really creative. It's a blank slate. It's a white sheet of paper that you can, you know, express yourself, come up with crazy ideas and actually in quite short amount of time do something. Customer Discovery and Use Case Exploration And then, you know, I built a lot of different robots, participated, worked on humanoids, flying robots, robot that was balance on one ball. So all these crazy ideas. And for me it's that, you know, journey over many, many years now over 15 years in robotics, but it's still the same fascination for the technology, Yes, but ultimately also what do you do with it? Try to impact you can have with your customers. Speaker 2 And not want to ask you how did the first contacts with customers happen? So you know, I can imagine at some point you got this built from from scratch robot to walk. Maybe you had some type of navigation systems that you could use and then you had to find some type of application for it. How did that first customer discovery and use case discovery process happened to be? Speaker 1 Yeah, it actually we're still students, right And it wasn't that deliberate. What happened is you have a YouTube channel and you publish videos of your robot doing cool things and I was back in the days over 10 years ago when a robot could climb stairs. You're going to immediately 100 thousands of views. We went to events. We'd love to show our robot off right. That's what happened and what happened is people would see it and you know, people from the innovation departments really forward leaning people. They did the connection said, huh, interesting, you know, and in our case, it was the large operators in oil and gas and other industries that said, hey, we're looking actually for a super mobile device that can move through our facilities. And they asked us, look, can we run a test with your robot? And we're like, why would you want to do that? Why don't you put sensors everywhere? Turns out you can't put that. Every position is sensor, right? Like, OK, why don't you use humans or don't you use drones? And all of these questions, the more questions we asked, they had really good answers for it. And we're like, makes sense. Maybe we can, we can indeed help them. And so this were the first contacts very early phases we participated in challenges that some of these companies organized and then just kept exploring, putting out there what we had and on the other side see the resonance in the market. You won't be able to just invent the market, right? You have to really listen in between and also be careful that one customer who says this is awesome is not a market, right. So also then seeing, hey, if we have this, does it financially make sense? Is the market big enough and to then explore and see where's the strongest connection? There were also other ideas, right? We looked at last mile delivery, we looked at search and rescue entertainment, so many things where you could use or for like a robo Ford. But ultimately for us, we settled on industrial inspection because technically it's feasible because these are very controlled environment where you know, you don't have dogs and kids moving around. It's really, you know, you can solve it today already or even you know, a couple of years back plus financially makes a big impact. Now if you do last mite delivery, your unit economics is very, very challenging for robotic system in our facilities, right? Our customers, they produce 100 thousands of dollars of revenue per day. If we can help them mitigate a few hours of downtime, you know, reduce the exposure of people to hazardous situations, that's a big financial impact. So for us, it makes total sense to choose industrial inspection as a beachhead to really focus in the beginning on that topic. Speaker 2 Help me, help me understand what exactly in industrial inspection is missing. Challenges in Industrial Inspection Feel free to go deep. And as I thought at the beginning, lots of the people that listen to this are nerds and engineers, which I love. So what? What's wrong in that market? What's missing and why 4 legged robot is the best form factor to solve for it? Or, you know, maybe in what subset? Speaker 1 It is so over the decades, right, the industry, and that's the industry 4.0 movement started to automate more and more and more. And so you have the systems, the equipment itself often comes with sensors, right? And there's a whole range as a market of by your T sensors, third party equipment you can attach to to to the assets that you have. Ultimately, what is the industry doing right? They're producing, they need to squeeze out the most revenue for their asset. They need to prolong the lifetime of it. And so automation makes complete sense, but you're going to start hitting a limit because not for every problem that you have, you can attach a sensor. Also, these sensors become costly, right? To have a certain lifetime. Then you have communication. So the setup is very, very expensive. And So what happens is for critical processes, safety critical or continuous control of the machines, you use built ancestors and systems and sensors and the DCS systems to control the environment to to control the the process. Now that's not enough. You also need to see in between the equipment. So for example, if you have liquid leakages, coolings or oil tripling out, these are early indicators of a problem. If you have hotspots in the environment, a machine on the outside, if you have vibrations is a big one, right? You have shaft misalignment, you have a ball bearing that's about to fit. These are all things that if you're in a place for every location, right, it's very difficult to add to find that problem. And So what happens is it's a mix. The industry maximized on sensors until the ROI didn't work out. But nevertheless, with every shift you had, you know, people that walked around to collect data. Now collecting data is actually not that easy. Number one, our eyes and ears are not good sensors. So what you end up, you need subject matter experts walking around with thermal cameras and vibration sensors, etcetera, which is a profession by itself. And then it's very hard to find these people #1 they will maybe do it once a month, maybe once a week. So frequency is not high. The consistency is very difficult. And then you also have a delay in what they can deliver. And it's just hard to find the people ultimately. And so it makes sense to replicate that in a part of what you cannot cover with sensors, plus what people do today, and automate that with a robot. And this is also the shift I'm moving from the industry 4.0 to industry 5.0, where it's not about just completely automating everything, but combining what does a human do. And that's understanding contextually, decision making based on data that such robot can bring. So for us, we'd rather have the people, the experts focus on the analysis, contextual understanding, than the repetitive, often distant and dangerous work of the data collection, right? And so it's this collaboration between humans and robots that's the first time possible. And you ask about why 4 legs then? OK, why don't you use drones or something else? The fact is these industries are built for humans, right? It's full of steps and stairs and these industries have grown historically. So it's not that, you know, you can perfectly plan it in the beginning and make it robot friendly, but every week, every month you have a construction site. It's expanded. Many of these facilities have been built 30 plus years ago and still need to live, you know 30 plus years. Even for a Greenfield facilities that are fresh built. It's actually cheaper right to save the space save. You know, space is a luxury. So things tend to be stacked on top of each other connected through stairs confined spaces where you need to crawl through and again, A4 legged robot here is is ideal because you need to be able to climb, including stairs and steps. A drone is difficult to achieve that because of flight time, regulation, payload. You know, you have a lot of sensors on the robot and just the autonomy in environments where you don't have any connectivity to the drone, right or to the robot. So you also bring the compute and autonomy to sense it's the Lidars with you. And here it's quite complementary actually for like robot carries the heavy sensors and does it routinely multiple times a day, whereas a drone is typically human controlled device once a month, once 1/4 and to get a new perspectives into tanks or flying along scaffolding or other, you know, getting a higher perspective than what a human or our robot in touch case could do itself. Speaker 2 You have a robot that you can put on the ground, have it walk around. You can have a record all type of video data. You can have different sensors, multi special if you want. If you want to put infrared, you probably can. Don't know if power requirements allow, but yeah, you're nodding. So I think they do. And it can look at things and it can see things. Technical Requirements for Quadruped Robots But what is the minimum level of technical requirement that you need for? If I were to start today, a quadruple company, do you need the tab autonomy to navigate into mapped environments? Can I have planned motion? What do I need to have on the sensor side? Speaker 1 Wonderful question and and that's very very important in the use case that's very very important of will be a longer answer and we'll have a discussion about core to how to bring it you know a robot successful to market. Let me start, I mean, if if you were to build just a quadruped robot by itself and send that into a facility and hope to find something that's not going to work, right? You, you need to be very, very specific on what are you trying to solve. Is there even you know, what is the historic data shown you? What are the problems that this facility has suffered from? How do they show up on what equipment or what sensors do you need to look at how often and what from from what perspective to find to have a chance to find the problem. So at the center actually our, you know, it's important to understand our customers, they're not in the business of buying a four like a robot. They're in the business of optimizing the up time. And So what they're interested in is information insights that helps to take them good decisions, early decisions and precise decisions. And ultimately our model as well, right? We learned that the robot is a core enabler, but at the center of our offering is actually the software, the AI that turns all this data into valuable information. Now imagine and you know, in the beginning, we delivered the, the raw data, the photos, the videos, the thermography, all the sensors you could. And by now we're collecting some hundreds, thousands of autonomous inspection points per month to our customers. Nobody can look at that manually, right? Nobody has to tie me. So what's clear to them? They said, look, all the data is valuable, but it will be archived so we can go back and historic purpose, it's important to look over time. All the orange stuff, warnings we'll ignore. Show me the red stuff, show me the real problems, right? And so there we're building this entire data set and training and AI capabilities such that a reliability engineer has a dashboard, a front end and notification system, user management and root cause analysis tools, multi sensor overlay. So the full contextual understanding to see where the problems are flagged by the robot, but then dive deeper and ultimately support the decision to do something about it or warn somebody about the potential issue or take a decision on triggering a work order. Now working our way back, we have this software. What do you need to collect the data? Certainly you need autonomy. Nobody wants a remote controlled robot. If you're there, then you could also just do it yourself, right? There is moments where the operator would love to can log into the system and remote control it for situational awareness, but 95% of the times the robots are used, this is completely autonomous. And this is where the beauty of robotics and the promise of robotics comes in, right? You have a system that can do the same task over and over by itself, repetitively, very precisely coming back one level even further. You need the quadruped robot. Yes, you need the platform that can actually carry the sensors. And here the minimum. What we found is certainly visual thermography, acoustics, these are the minimums listening touch, right of sense and seeing and then of course smelling gas detection. And then also, and all of these things can also go beyond right. The robot can smell gases that we don't so we can see temperatures that we don't it can heal ultrasonic sounds that we don't. So it's it's going beyond actually what the humans can do. Then on the lowest level you, you need a four legged robot and there what you need is just high performance of robot that can move through the entire facility, that has a small form factor, can carry enough payload, but is also ruggedized, right? You don't want to have a toy that you know, breaks if it drains or something, right? So you have a ruggedized industrial product that can take a beating in very dusty, wet, corrosive environments, reliable, you know, lifetime for multiple years, easy to service as well. And then also very important safety and certification, you have the basics of the people certification such as CE and FCC, but we're entering highly, you know, hazardous environments with potentially explosive gases are so you're not allowed to ignite the spark and that's very difficult to achieve for a robot with batteries and motors etcetera. So that's why we also built a second product line called Animal X, which you can send into explosive environments. Integration and Deployment Challenges But as you can tell already, this is a big stack of topics that you need to solve into robots. The four legged robot is a key enabler, but ultimately the value comes from the higher level software that you provide. Speaker 2 What do you think it's the key challenge for this form factor? Where does it fitness and essentially in these deployments and applications? Speaker 1 The beauty of A4 legged robot is its versatility, right? With four legs, you can literally move anywhere that the person can go, right? We maybe some of the challenges are the viewpoint by sometimes a human is higher than the robot in those moments, actually, the robot can just step back a bit if there's space and use the zoom cameras with high resolution to do it. So we find in many, many cases that that's not a limiting factor. And with that versatility, but you also, you know, it invites that the complexity of the robot is still with 12 motors and the payloads, it was relatively high. So the ROI doesn't always work. If you have a very dense and large facility, it's it's perfectly fine. But if you have a smaller substations, a couple of meters where it's all flat, you don't need a four legged robot, right. So what we're big believers actually in that there's no ultimate one form factor, not for like a robot, not the human and robot that will solve it all. But the beauty of it is actually that you can adapt these robots to the task at hand and optimize it. Instead of having a humanoid, you know, cleaning your dishes, you have a dishwasher, right? So that's the the analogy here. And so for us, it's we're also, you know, in not, we're not saying we're for like a robotics company. Our vision is actually creating a workforce of autonomous robots, but that workforce can come in different forms and shapes, right? And optimized to the task. So we can take our robot but keep 90% of the modules, hardware and software, maybe just attach four wheels or tracks, whatever you need. Or, you know, magnetic wheels if you wanted to climb walls, but reusing a lot of the components. And that helps us to build an ecosystem by increasing step by step the value. Right now we're very focused on 4 like robots, you know tonight dilute our resources and efforts, but ultimately I believe in a heterogeneous fleet of different type of robots. Speaker 2 Is it enough to build the three stacks that you explain me or do you have to go further than that and then build also deployment motion that works when you can rely on partners? Speaker 1 Yes, I love that question as well. Touches on very, very important strategic questions that where we took quite deliberately. And you're absolutely right, what I described is essentially the the core product, the solution that we have, but that needs to come with a lot of additional layers and the integration of their and you know cybersecurity, warranty services, all of that topic. But let's focus on the integration part. And this is a key necessity, but also enabling competitive differentiator. Ultimately, the reason I mean, what we found at the beginning is we thought, hey, let's build a robot. And there's the big companies and resellers and others, they will figure out how to, how to deploy these robots. So let's, we don't need to reinvent here, you know, our process. But it turns out because this is such a new market, right? We essentially there's a few companies that do what we do less than a handful, right? That this is pioneering work and it's essentially not a product yet in the sense that people know how to use it. When you buy a car, it's very clear how to use it, right? But if you, you know, just buy SAP, they don't, you know, you're not just download it and start using it. It's much, much more complicated than that. And this is often currently the state with robotics. As well that you need to even in the sales journey, but also then after the sales movement, you start building a deep relationship with a customer and indeed it starts very early on. What is the problem they're actually solving? How is it connected to the top line, to the return on investment? And then you break it down to individual assets, as I mentioned before, what are you inspecting on etcetera, stakeholder management, the cybersecurity, the safety, all of that people is in tech groups and committees that need to decide and you help the customer to navigate data. There's parts that we can do, parts that the customers can do. And so for us, we have, you know, that was very early on. We understood that we cannot just ship a robot and wish the customer good luck, right? That won't work. Also our customers, you speak to them, they love that we're there with the people. There's a consistent contact from A-Z, somebody who takes care, somebody know that takes responsibility, is there to set up the robot, but also understands, hey, you know, where do we integrate it? Where does the data land? Let me talk to the reliability engineers, let me talk to the operations teams. Then it's down to API integration, all of that. So it's a big, big project, but that's ultimately how you need to look at it. It's, it's a solution that consists of a product, but also the entire go to market in, in customer success movement. And it's actually very exciting for us because we get to work hands on with the customers, our customer success teams quite sizable. I believe it's, it's one of our superpowers at the company as well to be out there to understand then the feedback and directly feed it back into the product. And that's a very nice reinforcing cycle. If you let that go in a market that is not fully established yet, right, it's going to be hard for you to keep competitive. So we see it as a necessity, but also as a beautiful opportunity for us to build that understanding. Speaker 2 You guys thought you needed X, but let's figure out together what you actually need. And you guys, you know, thought you could achieve Y, but let's figure out what you can achieve and and a timeline for that. Is that more or less the? Speaker 1 That's right. It's a very consultative sales process, but a customer knows that the problems that that's what they're really good at, right. I mean they know their facilities inside that, but they don't know what the robot can do. So for us consulting going deep into the problems and we have by now we're hiring, you know, we have expert reliability engineers on our team that understand also the customers reliability strategy and these kind of topics, but also understand the robot. So we're bridging that gap, agree the sales, but then also after we have a defined project where we agree on implementation milestones and KPI's we want to see from the robot. And that's a project we're going through together. And ultimately, this is also the trick from, instead of just deploying one or two robots to the customers for trial purposes, you have an entire scaling plan, two robots here, 2 robots over there. Once we have reached these milestones and KPIs, we're expanding, right? Market Dynamics and Future Outlook And that's the beauty. Our customers, they all have the potential for dozens, if not hundreds of robots. So for us it means instead of diluting efforts across many, many companies with one robot, you rather focus on some of the strong ones and the believers and almost Co develop some of the solutions through that very close contact. Speaker 2 So what does a company like Antibiotics that makes its own hardware and integrates it? What should you guys be aware of as you look at what's happening in this phenomenon? Where where like your strengths and where like your weaknesses, essentially. Speaker 1 Yeah, that's a wonderful question. Certainly in the beginning, if you, you know, if I take you 10 years back, there was no ecosystem of robotic suppliers and parts. So we have to go really deep in and say we need to build our own actuator because the ones from the industry are just not designed for lightweight robot. We had to build our own batteries, inspection pedals, all these things that were just not around for a form factory. Now this has evolved and absolutely there starts to be a commoditization of robotics hardware, but you can see this as a threat and say, hey, OK, you know, we cannot do this business anymore. I see it actually has an opportunity, meaning that we can now stop doing everything ourselves that we needed originally, build some strong partnerships and work with suppliers globally that have high quality components, including from Asia and build on those and then truly focus on what you're good at. Partnerships and System Integration So we're not a mass producer, low cost manufacturer, right? For this, clearly we'll need partners. We're neither a motor supply or no battery. So things that are out there, if somebody else can do it, we should buy it, right? So that's the idea. What we're then really good at is still the robotics, the AI, the autonomy, those pieces, and also safety relevant, right? Because that flows through the entire user experience up to the AI inspection capabilities. Cuz ultimately a system that you use needs to be easy to use. It shouldn't feel like multiple systems clued on top of each other, but it's one cohesive, nice to use, nice, but also highly reliable and performance system. So yes, heavy focus on the software, but also on the system architecture. Ultimately. Now, yes, you could buy A4 legged robot, but that's just, you know, as we mentioned before, that's just a carrier. We have great ideas already for the next generations of robots how they will need to evolve to serve some of the topics you mentioned here before to build an ever more specialized and performance inspect, you know, data inspection vehicle we call it. Then at that point it's not maybe not even a robot anymore. It's just we're then used to that name. It's a system that can significantly collect high amounts of data. So yes, for us it means use the opportunity partner much more including some of the commoditized equipment and with that stay cost competitive. Yeah, and, and, and our power here is that I mean, number one, we have a few 100 robots deployed already and and growing right every week, meaning that we're very close to customers. We have the relationship, we start having the brand, we start having the date and we have the critical insights what is needed for the full solution. But then we need to get, so we're taking ownership and we design the full stack, but it doesn't need to mean that we built to full stack. And wherever you can find the right partners with, you know, the specialized solutions. The Future of Robotics and OEMs So you can you become yourself a little bit of a system integrator of, of different technologies to do that. But this is quite normal in all the tech industries, right? If you from mobile phones to other devices, it's it's a play where you say what you want and you find the best and cheapest way to bring that together with global partners. Speaker 2 So what is the case for building an OEM in 2025 essentially where you're still building your own stuff rather than buying, fitting, coding on it, and then going and integrating it? Is it that the technology is still not there, so you need to build your own? Is it that you believe there's a longer lasting note? They're building your own platforms. You know, what is the mental model there? Speaker 1 Yeah. So we have to answer that the market is still in the early phases that in for many, many years now we'll see rapid fast iterations. Right now we're talking about four legged robots and a couple of years back there was nothing around. Now we have a bunch of two legged and four legged systems that can walk. But that that's just early phase and we'll look back in a couple of years and we'll be smiling how rudimentary that was. And right now I think it's specialization starts to happen where depending on the customers in use case, you'll have your your specific models. So for us is not even an option to, to buy now off the shelf, you know, robots because we need to provide a customer as a service up time. We need to show them that we're in control of the data, etcetera. So just putting things together is enough for a demo, but something that has performance guarantees, services, spare parts, apologies, security guarantees, all of that. And certification a big one, right? When you need to prove part of the software and hardware together that today you cannot put together from individual components. And for us to serve the market, we need to do it ourselves. But you're right, what is the long term play is for us to make sure that we're in end contact with the customers, but we're also in control of the edge. Now the edge, I mean the robot itself, we need to determine what software runs in it and what third party and the communications because this will then create this stickiness and to trust we have something on the edge. It's like owning the phone, right Apple and you know the other vendors, they take the profit here because they're in control of what happens on the device itself. Might be the possible in the future that you're not making the biggest margin on the hardware itself and that's OK because then you have your value, the software and the capabilities that come on top of the of the system. And again, it comes to the very beginning when I said the customers are not here to buy a robot, that's just the entry ticket. Humanoids vs. Specialized Robots Where does the value come from? And we're very clearly focused also on on the value aspect, the the growing use cases and inspection capabilities, but then also the software in the cloud and how you access an API. Yeah, I hope that I guess it's, it's, it gives you an answer how we think about that the the robot itself, what we're talking about commoditization is a part of the solution. But ultimately it's not going to be the cheapest or best robot or the cheapest or best AI model. It's who can package that in a solution that consistently it reliably delivers the value. This is how we look at it. Speaker 2 My think is what we're showing off today is what raises money. These very complex as machines with, you know, tendons to walk like mechanical tendons to walk and this very complicated and delicate bipedal motion, etcetera. All of this is what raises money. I think what's going to make money is this sort of like generalist 2 armed torso on some mobile base that you can just drive around, have it do things without having to have security concerns of safety concerns of this very unstable and we can dive into a white unstable robot that can fall on people. It's like 80K of metal that, you know, squishes you down. So I think in general, we'll have them in the future. It'll be cool to have them around. But most of the applications actually come from manipulation robots. It doesn't matter if they look like humanoids or not, they just manipulate things, right? I think that's at least that's my take. And so when I see a lot of the big razors, right, So the companies that are raising multi billion dollar rounds to build them, it scares me because it looks like capital location is going to a place. We're probably not all of the revenue will lie. And probably I'm wrong because, you know, probably I'm going because that's going to trickle down and some form factor that's going to help these companies also make money. But I want to understand, you know, what's your take here, right? And and yeah, because that's mine. Speaker 1 I'm afraid we're a little bit too aligned. So for me, it's also the humanoid form factory. It's a manipulation device. And the biggest challenge today is not to build actually the humanoid itself, right. You see all these wonderful videos of backflip model, all the crazy stuff that's fine. But that what has not been solved yet is the fine manipulation, the end effect or so there's still ways to go right to to to reach that level of manipulation control. And they're absolutely right. These arms might be stationary or on a mobile platform. We have three arms, one arm. So it's, it's that has been the need is unlocked. So it's clear that we need that the flexibility thereof. But I also agree that they might not look like, you know, the human shaped with a head and two arms and two legs, because often it's not needed. So there will be a specialization nevertheless. I mean, I mean, I'm excited about humanoids and everything that's happening. Just imagine, you know, five years back, we needed to explain to investors why, why robotics, right? Why is that a thing? Why is that even needed in the industry? And so that opened up a a race, but also a general understanding that robotics is here to stay. It has a big impact the the race slash, the form factor and markets will will evolve very quickly. And so I have nothing against human rights. I I also see there might be applications where it's completely feasible to say I need 2 legs and two arms fine, but I do not think that human rights will be the one robot that will, you know, do it all. And this is also when you speak to humanoid companies, they see it the same way. There's all these still the questions about define manipulation, the safety you talked about. Then the cost will start to play a role, the exact use case. Also, many of these humanoid companies will go through the same journey as many other robotics companies did. My concern is a bit there's too many humanoids robotics companies out there right now. Many of them will not work out. So I don't want investors in general to be scared of robotics. It is a difficult topic. It takes more money, it takes more time than it usual SAS company, a software company, but the opportunity is very exciting. AI and Robotics: A Perfect Match So I'm thankful of the the hype a bit, right. So it's good, but we need to be mindful to quickly deploy and show results and customer adoption. And this is what Antibiotics has been doing for a couple of years now. And so we can position ourselves quite complimentary to what's happening in the humanoid market. Speaker 2 I have another topic on the hype front, which I'm actually more excited about than humanoids, at least in the short term. I have just bought from from an online store a hugging face letter. But, and just because I, I was looking wonderful, too many people in my YouTube feed building it and and and teaching it to do things and I wanted to try it right. I, I, I left my automation kind of effort, a bit of university and automating some things in in my parents house where I'm actually calling you right now from. And I want to get back into it. And the reason why I can do it without losing my entire, losing my job, for example, getting fired by spending too much time on it is that it has evolved. The entire ecosystem has evolved where essentially I can take frontier end to end models, which I hope you will define for us. But you know, that'll be awesome. And I can just push them on a common hardware platform that it's somewhat well understood and well mapped. And then that hardware platform can do things for me on command. I can tell it to do things. And if the controls and navigation permit it, it will try to do these things. And it gets trained on pre trained on an immense amount of data, which is not, you know, as big as we need it, but at least it's somewhat big. And it doesn't need perfect modeling of everything that's happening around it, which I want to put in our asterisk in because I think it's a limitation. And then, yeah. And then I can just do things with it, right? So I can tell the robot, hey, move my glass from the kitchen sink to the kitchen table and if it has the capability to do it, if the glass isn't too heavy for it, it could do it. Like there's models they to do it. So I wanted to understand if you're excited about this AB, what the limitations you think are and C doesn't have anything useful for antibiotics this entire end to end models wave or it's something for a different use case, a different customer segment and and just people like me buying a layer robot online. Speaker 1 So we, I mean AI and robotics is a match made in heaven very clearly. And we saw this actually in 2019 where we already pioneered to get away the ETA Zurich. And just a small anecdote here for the technical people, it was, you know, the time where it was all model based control, the very, you know, classical approaches and that worked OK. It just needed a tremendous amount of of engineering and debugging and you always ended up at the limitation of human creativity and what you could note in mathematics and program. So it was a steady slow progress. You would get somewhere. Especially botanics has done a wonderful job over the decades that they have worked on this. But then I saw a demo of one of our students, a deep reinforcement learning algorithm. The robot was shaky and works very poorly. But I just never played around and it was never trained to do that. But I ran it into a couch on purpose, you know, just pushing the couch a bit. And in a traditional sense for model predictive control, but say this mention, etcetera, the robot would have freaked out. That would have been pretty short. But that robot suddenly stopped with the front legs kneeled down a bit in front, like pushing, and the hind legs kind of like scratch off the surface and just recover. It was like like an animal or if you're human, pushes against a heavy object. Wow, it's exactly, you know something has not been programmed to do, but it's performed so gracefully. So that was for us a moment I said, OK guys, we have to fully embrace this, right. We switched the entire team from the traditional approach at the company to do deep reinforcement learning within six months. We have model stack worked really, really well for the last four plus years we have been shipping all our robots with AI based locomotion. And what I mean by that and I'm sure there's a different definition of end to end, but we use raw input from the cameras, from the encoders from the IMU, we spit out joint trajectory. So it's, it's quite of a broad net frost autonomy is still a different layer and we can talk about how to do the architecture, but that's so it's, as I mentioned, AI and robotics is a match made in heaven. And for us, we're using that across the board. However, we're, I'm not a believer that will be that one end to end model from raw data to the entire, you know, everything that you do, including inspections. We're more, you know, believers in a smart architecture also for control purposes, for iteration speeds, etcetera. And what we have seen, the locomotion is pretty much solved by now. You can always optimize, but it's such a high performance and reliability that that's great. So we can build the next level autonomy on. So we do obstacle detection, navigation also now AI based. And of course everything you do in inspection is already AI based, but different models for different use cases. Again, in between there might be some traditional approaches, but if you have a flexible smart architecture, this also allows you to do third party models, right? If somebody has an amazing, I don't know, detection of certain components or a new autonomy stack, we can very quickly integrate that in into our architecture. And so for us, it just opens, It's a new tool that we're completely embracing, have been pioneering for many, many years. And I think the journey is on to do even more. You asked about where is this relevant for antibiotics. What I'd love to do is what you mentioned in the beginning, right now is it's still a robot that the human sets up. So we need to figure out to get with the customer where the robot should go, where should not go, what to look at. But imagine now with hundreds of thousands of hours these robots have, you know, spent at facilities, Think of them not as a, you know, intern in the first day that you need to onboard, but think of it as an expert from the industry, having spent hundreds of years. So this person, this robot could come in, walk one through your facility, and instead of you telling him what to do, he will tell you, these are the things I've found. These are the equipment. These are typically prone to this error. And it will start having a dialogue with you, right? And so it's like Chachi PT Co solving the problem together. You're not bogged down by the nitty gritty details on how to set it up, but rather on a very high level, determining together the the strategy on how you're going to look at things, the whole circle data, what you already do, what you don't do, what the robot can do for you. Building Trust and Reliability in Robotics I think there's a vast opportunity to build that higher level, not just autonomy, but true industrial intelligence engine for the robot to be an ever more performant Co worker. That comes down to data. How much data can you actually collect real world data from the facilities And they're, I believe we're very well equipped because you need to have the robots out there, right? So you won't be able to skip that by being a pure software company and and simulating it. So I think that it's an exciting field we're heavily investing in. Speaker 2 What you mentioned, it's more of a traditional deep learning. It's fun that we call traditional land right now, traditional deep learning approach and a lot of new transformer based approaches with Vlas and kind of like these big visual language action models that are taking data that's coming from the Internet to understand what's what from text on the Internet to understand what's this, what's my what's my handler saying? As well as joint controls. So data is coming actually from robots or from haptic sensors, wearables that people are are using to collect data, all of this to spit out joint, joint positions and torques. That's sketchier to me because like you don't know what's going on and interpretability there. It's a big issue. So that's also what I want to try to talk to you about. Is that the step to get the generalized robot autonomy or is more building it up by block from a specific use case growing into a horizontal variety of use cases? Speaker 1 So for us, I mean, history will show, right? For us, it's the block by block approach. Yes, VLAS and other models on a high level for the highest level intelligence, very valuable. But at the same times, I'm very happy if we don't need to update our locomotion controller constantly. It's also just builds the confidence to know if you commanded this, this will happen, right? The same thing for us, it's still, as I mentioned, the new markets. So consistency, reliability builds trust and this is needed for adoption. So just going out and constantly with the craziest new AI models that you know what's happening, that that's a risky business. So for us, block by block has been the successful trajectory directory because quite early on you can start just deploying robots successfully. If you're waiting to collect all that data to build this vast end to end model that you believe, I'm not sure if you know if you ever, you don't know how long it's going to take. So that's a bit of a dangerous trajectory. Meanwhile though, yes, collecting data and systematic learning is key. And what you need to do that you need a trust from the customers, You need all the cyber security in places, you need anonymization of data. You need to be in the front end of the customers. If you just have A tag player, you will never get the feedback. If this data was good or bad for us, it means we have the front end with the reliability engineers. They give us feedback. Was this correct? Was this wrong? And so with that, we're getting expert training over time. But you need to build it in your architecture together with the customer trust that you need to build that you actually get hands on this very sensitive data. Speaker 2 You're saying I'd rather know what's happening inside my robot essentially at all times. And then as I as the stack evolves, you can probably add more and more features that can have that more of a one shot, you know, a one shot thing trying to understand what to do next. But I want to be able to have the control layer at the bottom that I really trust. And I know I have, you know, rugged eyes and and made reliable for myself. Is that is that pretty much the idea? Speaker 1 That's the right philosophy because the actions of your robots, especially the physical ones, they have safety implications and for our robots are even part of our certification. So there's very, there's no room for error in that, right? Well, on the higher level for inspection capabilities, I'm happy to iterate quite quickly. And if maybe one of the interpretations is not perfect, there's still a human in the loop and mistakes are OK for the sake of speed and iterations. Whereas in locomotion, as I mentioned, I'm happy if we don't need to touch that ever again, right? We'll, we'll still upgrade it, but the steps have become smaller because you also have a bunch of hundreds of robots out there. So whenever you touch it, you got to make sure that it's still working perfectly for all the existing customers, right? So the more physical, the more safety implications there are, the more testing you will have to do. And so iterations those down as well, right? And performance expectations grow. So separating these two has allowed us to do still both be conservative safe, but still it tried very quickly with the latest technologies, inspection capabilities. Speaker 2 One last question that I ask every guest at the end of every episode, which is what is your advice the day the people that are building hard things? Advice for Aspiring Robotics Builders And I think in this case, I think that the market is big enough and, and, and the interest is big enough that we can narrow it down to building robots and building robotics. What's your advice to these young people that are struggling to understand what to do next, what to focus on, what type of principles and guiding values they should have, especially with the entire, you know, the monetization fear that I think it's, it's coming at least in, in the niche environments and, and discussions. What should they, what do you think they should do? And they should keep in mind as they move their first steps towards building robots for the, you know, the next generation of hardware that's coming online. Speaker 1 Yeah. I mean, something that I observe is there's a lot of excitement in what we call building the shovel, right? It feels like, hey, there's a gold diggers time. There's something out there. Everybody knows it's coming and a lot of people are trying to we don't know how that's going to look like. It's it's too complicated to talk to customers in exactly figure out, let us build to take layer that we're sure eventually we'll find it's where we'll build the best tech. That's very, very dangerous, right? There's more robotics tech companies, there's more fleet management software out there than they're actually robotics companies actually do it right. That's crazy. And so here I would embrace the actual state of robotics and take one customer problem that is big enough, but walk in the shoes of the customers truly understand it. Do it yourself, right? And then reverse engineers the problem and build your robot to do that. It's we also came out with a technology push, but I think what we did well is from the beginning say, hey, we need to be out there. We cannot just do research. So I myself, I spent, you know, weeks and months with customers. I stopped programming the moment I graduated and built up the business development, marketing, the sales, etcetera. You need to have that deep understanding and build that market. I think that's very exciting. Not only fall in love with your partner, that's OK, but also fall in love with the customer's perspective. And marrying these two things together is, is very, very rewarding. And it's one of the unique opportunities, right? If you're working on automotive these days, I believe it's pretty clear what a car does. But for robots, you can be very, very creative and still build the fascination, the positioning of how people will think and use, you know, and, and yeah, interact with robots and somebody's creative and has the energy for it. I would clearly advise in that direction. Speaker 2 I love it Peter. I think this will be very valuable and I thank you for all the time you put into this. I hope we can do a part too soon. And actually I've already made you promise that if I, you know, if I come by around, around your HQ, I can get a tour. That's still valid. Speaker 1 100% You're very welcome here and very happy to do a Part 2. This has been a fascinating discussion. Thanks a lot, comedian. Speaker 2 I love it. Thanks, Peter. Thanks.