How Mbodi Is Solving Robotics’ Scaling Problem with Xavier Chi

0:00We did a recent test run with our system eight hours straight from beginning to end. 99.6 % success. So that's actually quite good.

0:09Isabelle Johannesen: That's pretty good. Yes. What's the comparison to the human success rate? Human also messed up. There's really no metric for how often human mess up, but human do make mistakes as well. That was Xavier Chi. Last October, he and his team from M-Body took the stage at TechCrunch Disrupt with one of the more memorable demos of the competition, teaching a robotic arm to pour a Coca-Cola live on stage. Which was honestly a risky and potentially messy move on Xavier's part, but to his credit and to the relief of our production team, his robot pulled it off. Since applying for Battlefield, Xavier and his team have continued working with major robotics players like ABB, gone through Y Combinator, raised a seed round, and started putting their technology to work in real-world industrial environments.

0:57Isabelle Johannesen: So, I wanted to bring Xavier back to talk about what happens after the pitch competition ends and what it actually takes to build and fund a robotics startup. My name is Isabel Johannesson, and you're listening to Build Mode. Hey Xavier, how's it going? Hi, nice to see you again. Yeah, welcome to Build Mode. It's very good to see you again. Last time I saw you was on the startup Battlefield stage back in October. That was a lot of fun. Yeah, that was a lot of fun. We were literally like teaching Robo how to pour coke on stage. You know, you guys easily had probably the crowd favorite demo. You nearly killed me in the process.

1:30I remember we were working on your demo and your pitch together

1:33Isabelle Johannesen: for a few weeks and you guys were, you know, working out how you wanted to do this. And I recall that the coke can you were going to pour with the robotic arm was to stay closed during the live demo. Okay, I'm gonna do something risky. I'm gonna open this. It almost messed up, but it worked and it was a nice demo. It did work and it was fantastic. I just remember being backstage and we heard the soda can click open and I was like, oh, these guys. That's actually not originally part of the plan. And then on that same day, we realized, okay, actually, let's just pour it. Why not? Why not? Make Isabel crazy.

2:10Isabelle Johannesen: But it worked done. It was very cool. So congrats on that. Let's back up a few steps for our listeners. So if you can introduce yourself and Embody and what you guys are working on. Absolutely. So I'm Xavier, co-founder of Embody. And what we build is AI software for robots that can plug into any type of robots, mostly focusing on industrial robots and collaborative robots. So anyone can simply go and talk to the robot, teach robot new skills with natural language, and they can run reliably by themselves in production. This solves a lot of fundamental issues in robotics, make these robots a lot more accessible than the past.

2:45Isabelle Johannesen: So let's talk about this past, right? What is sort of the standard today and how are you plugging in to change those processes? So the past, actually, robotics has been the same for the past, I would say, decades, several decades. We've seen like small breakthrough here and there, but it's fundamentally the same thing. every time you need a robot to do something new or just basically put automation in a process, you need to put a lot of integration work in there and put a robot, say a giant industrial robot there, put a lot of fencing around it, and have someone programming the robot to do the task.

3:21Also, it could take a very long time. And after all that, the robot can only do one thing at one line. And that's kind of traditional automation. It's really a headache for a lot of people, and people don't even want to think about it because of that complexity. And that's a common problem for automation robotics. That's why we realize it's not obvious to a lot of people that the whole world, it's way less automated than it's supposed to be. And I was visiting a plant of one of our customers. We saw that there are a lot of ladies. All they do, the whole job, it's actually one of the, I'm not going to say the name, but it's one of the largest pharmaceutical company in the space.

3:59The lady's just picking different cartons and put it in the conveyor all day long for like eight hours. There are two shifts, two ladies, 16 hours total, just doing that same thing over and over again. It should not be done by a human. But traditional automation is so hard. And what we built is enabling anyone to just make this robot a lot more accessible. Basically, pull a human out, put the robot in, and people can talk to the robot, get the robot to learn and operate just like that human. So that's kind of the main thing. we want to make robots or deploy robots everywhere in a very easy way.

4:35Isabelle Johannesen: All right. The robot revolution is coming. Yes, it is. So let's talk about some of these use cases. So, of course, picking and placing products. What else are your customers using Embody for these days? These days are mostly on the pick and place and packing, kitting type of task because those processes are highly manual. Surprisingly, because there are variation in the process itself, the robot might need to pack, say, some type of product this week or some other new type of product next week. It's just impossible to automate in the past because otherwise people will have engineers coming in to program these robots every week, every day to make it work in a very rigid way.

5:14It's just not possible.

5:16Isabelle Johannesen: So more than anything, you're replacing those engineers who are programming these bots all the time. Yes, we are also doing that. And there are two aspects there. So because we can do that, then we can actually automate a lot of tasks that's currently done by a human. I believe humans should do more creative work instead of just repeating the same motion. Like singing, huh? Yes. Before we started recording, we both were vibing on the fact that we like to sing. Yes, yes. I mean, I can get behind that. So I think you're right that a lot of people assume there's more automation going on than there actually is.

5:44Isabelle Johannesen: Especially, you know, we see some of the likes of Amazon, right? Why aren't they building this themselves? There are like a different perspective in this space. Historically, large companies like to do more on the fundamental change or the foundation layer of technology, saying training more kind of foundation model to support all kinds of things. But it's sometimes hard for large corporations to think about different end customers' problem, their exact problem, and solve that problem starting from a small scale to a large scale. That's how every generation of new companies show up. and crack different fields.

6:24Yeah, but Amazon is definitely great. They acquired Kiva system quite some time ago. In Kiva system, they make the little AMRs or AGVs taking the whole Amazon shelf to a station so someone can just pick right from that kind of shelf. And it's deployed, I would say, everywhere. So it's quite impressive.

6:43Isabelle Johannesen: Yeah, I mean, I think that's the kind of stuff that the average consumer has probably seen clips of Amazon warehouses. Exactly. But Amazon aside, you guys have some traction and customers with some very large companies. And I know you're always a little secretive to talk about the names of them, which I respect. But tell us a little bit about how you've gone about securing some of those large corporate customers. There are different ways. On the channel side, well, this I can say definitely publicly. One of our partner is ABV Robotics. They're the largest robot manufacturer in the whole world.

7:14And we won their AI Startup Challenge back in 2024, late 2024. out of a couple hundred companies. And we are in a really close relationship with them. And they are amazing at making hardware. And we are really good at making software. So it's kind of a natural fit together. And we can solve a lot of customers' problems, especially some large customers' problems. And apparently, they have a lot of large customers. For example, in the CPG space. We also recently doing a lot of direct ourselves. Try to sell this solution or sell this product. and the most important thing is that it really solved the customer's problem.

7:51That's the fundamental thing because labor is a huge problem. Labor shortage is one thing and labor cost is the other thing. It's a problem for the whole industry for a very, very long time. I had a really funny anecdote with one of our customers is that they have to always overhire by 20 % in their warehouse because sometimes people just stop showing up randomly. It's not great. And then later on, another giant company opened up another warehouse just next to them and hired a lot of people away and they can't find enough worker anymore. So if we can automate those processes in a very easy way with robots, then it solves a lot of problems for them.

8:33Isabelle Johannesen: Yeah, I can imagine. Let's talk a little bit more about this ABB partnership because that clearly was a big unlock for you guys. And you secured that partnership by winning their pitch competition. I'm curious for founders listening, Do you think that that is sort of a repeatable playbook? Or how do you go about identifying which competitions are really worth your time as a busy early stage founder? Yeah, there are actually a lot of, I won't say random competition, but different kinds of competitions out there. But we really need to evaluate what we can get out of it and the quality of that competition.

9:05For example, the one we had with ABB, it's not happening a lot. And it's very, very high quality. There are hundreds of companies in there. They have a very thorough evaluation process. They send one of their engineer, it's actually an ecosystem manager, coming to New York to our office and watch our stuff, make sure everything works. We are not like just pitching a fake demo or something. It's actually working very reliably. It was a really lengthy process, like several months. And having them coming here and several rounds of pitch and a final solution presentation, all of that. And we got this partnership.

9:41and it's really worth it. I think the most important thing is to see what people can get out of those type of competition. And speaking of that, I love the Startup Battlefield competition.

9:53Isabelle Johannesen: And why did you decide to do Startup Battlefield? This is a really interesting question because, well, you definitely know Silicon Valley, the show. Of course. Yes. So I watched a show back in college, like first year of college or something. And I watched this TechCrunch Startup Battlefield thing. And at that moment, I decided, OK, at some point, I'm going to do startup. I'm definitely going to come to this show, this startup battlefield stage. I love that. And, you know, we've heard a similar story before. Yes. And so I think it's become in many ways a milestone that some founders feel like is kind of just part of the stamp of approval of being a founder.

10:29Exactly. Also, it's a great stage. Have a lot of exposure, good audience. And we actually we found really important advisors of ours because of the startup battlefield.

10:41Isabelle Johannesen: That's great. Did they approach you after your pitch? Yes. Okay. We love to hear stories like that. Let's talk a little bit about the judges on the battlefield stage. I mean, we pick judges that we know will grill well, that have some really intense questions to ask. What did you think was sort of the hardest question or one of the holes that they poked in your pitch? And did you feel you had a good answer for it? Ah, that's a great question. Actually, a lot of the questions are hard in the start of Battlefield. I remember one of the questions was that how can you scale this product? What is the go-to-market strategy on this through channel direct and anything?

11:21That's definitely a very hard question because in robotics, that's always been one of the hardest problems. Because it's not a, say, a B2B sale where we just have software and we sell to an enterprise and that's it. There are always many different vendors involved because we are making changes in the physical world. There are, for example, software companies like us. There are robot makers like ABB, Fanuc, KUKA, Universal Robots, and there are also system integrators that put the solution together for the customer. And then there's the end customer. It's the complexity of this whole landscape.

11:53And it's also a challenge historically on the go-to-market side.

11:58Isabelle Johannesen: And so, I mean, we chatted about this a little before we started recording, but I mean, a lot of robotics companies, particularly software for robotics, have struggled to scale. And so, I mean, you've outlined a few of those reasons why, but what do you think needs to change in order to get some of these applications into the real world? This is a really good question. And I had a theory around this. It's kind of like a fact. We haven't seen a single successful robotic software company yet. Successful, I mean, like generating really reasonable revenue at scale. At scale is important in robotics.

12:32There are some like decent robotic companies just as a whole, making more vertical integrated hardware and software, just selling those things. But we have yet to see a company like Google or Facebook type of company in robotics. Let me make an example. There's a company, I'm not going to say the name, but they are founded by the top researchers in the world some years ago, along with a professor. They have the best AI vision picking model for robotics. The thing is that how to sell this. Well, the way to scale this, the default way to scale this is to partner with, say, system integrator who put together a solution for the end customer.

13:13So they can just give the software to them and let them deploy this to a different customer. But then the problem is system integrators, they care more about selling, say, a whole warehouse of solution with conveyors and all kinds of systems. And robotics is only a very small piece in there. And after a lot of work, two, three years of work, there are only five robots there. It's really hard to scale that way. The sales cycle is super long, especially that's really bad for a tech company, a VC-backed tech company. And a lot of robotic companies naturally go into the way, okay, let's deploy this ourselves instead of relying on partners.

13:50Then the problem becomes, okay, if we have to deploy this ourselves, then it requires a lot of customization. Every time people put in customization into a solution, it's just not scalable. and that's a fundamental problem. It's not really compatible with the VC-backed tech startup. It just doesn't match that way and the balance sheet looks crazy. So, well, a lot of companies run out of money or just couldn't scale up. But it's going to be different with this generation of technology.

14:25Isabelle Johannesen: Why is that? There are a couple things. Okay, so let's say I will break this into two parts. The first part, the system integration part, that's not going to change. That's just the nature of it. But customization, that's the key. Traditional automation or earlier generation of technology in robotics require a lot of customization for each deployment. But because of this generation of AI, especially generative AI, we see a potential for robots or robot software can kind of figure out in that moment what needs to be done. instead of putting a lot of integration or customization around the work itself.

15:04Because of that, this might scale the robotics solution way more efficiently than previous generation. And that makes sense.

15:14Isabelle Johannesen: But I guess we're still seeing a challenge for startups in hardware and in robotics to be able to raise venture funds, like you mentioned. But you've raised funds. You had a pre-seed round and a seed round. You started with a fairly small pre-seed round. How did you stretch that money while you are building in a capital-intensive space? Even though you're on the software side, you still need to test with bots. You still need help with deployment. So how did you tell us just about the fundraising process in general? Fundraising for us is great, actually. And we raised a very small pre-seed when we started to bootstrap.

15:47The interesting thing is that back then, we just start with an idea. We don't have a lot of knowledge in the industry. And later on, like I mentioned, we won this ABB Startup Challenge. we got really deep into the industrial space. We see all the problems people have. And actually, a lot of problems can be solved with this generation of technology. And later on, we went through Y Combinator in their spring bash, spring 2025. YC is great. It's amazing. We can also talk a little bit about that. But we closed our seed round after YC. And I mean, the fundamental thing is how to successfully raise capital.

16:23It's to make people believe or make VC believe that the problem is real and it's going to scale and you're the right team to scale it.

16:37Isabelle Johannesen: Well, that part I want to touch on. So you and your co-founder Sebastian have impressive backgrounds from Google, but you were both on the software engineering side, not robotics experts, right? So how did you sort of, first of all, discover that this was even a problem you wanted to solve and then convince VCs that you were in fact the team to solve this? Yeah, so Sebastian is super smart. He triple major from UPenn and then studied robotics in the Grasp Lab. So he's been a robotist for some time. And then I've been building robots since college. Okay, okay. But it's a different time. We built a little drawing robot, but there's no AI.

17:13It's a different time. Yeah, still pretty cool. Yeah, it's cool and fun. I love robots. I always love robots. And we met at Google. We worked as teammates for a little bit over two years. and we always talk about robots from time to time. Yes, at Google. And around 2024, we see the time is right. We see a trend or a path of the digital intelligence going into the physical space and many industries going to look different because of that. That's how we kind of started. And we just decided, okay, it's time. Let's quit and start Imani.

17:43Isabelle Johannesen: So for any potential VCs listening, I'm curious how you think of sort of the future of robotics, right? I get that with the age of AI, it's shifting a bit. But what do you actually think people should be looking closer at these days? Is it humanoids? Is it the software behind it? Is it flying cars? I don't know. What do you think is kind of where we should be? Humanoid is definitely a very interesting topic. I think humanoids are fascinating. They look like us. But I always feel like they don't have to look like a human. I've visited enough warehouses and factories to know that we all can work perfectly there.

18:21or even why not four legs? Why two legs? Right? Anyway, I think in this robotic landscape, I would say there are a lot of excitement around foundation model, saying the foundation model, if we collect enough data, the hope is that we can have a model that's being able to do every single possible thing in the world. It's going to take some time. It's hard. It requires a lot of capital. So that's the foundation model route. And also, the route I like, personally, I like better is the more realistic route, actually deploying solution out there using AI, especially what we have, generative AI, or even classical robot.

19:02I use a combination of both and deploy solution out there because there's always a huge gap between what's deployed in production and research lab demos. In Research Lab, for example, if we can publish a really nice paper with, say, 80 % success rate or 90 % success rate, but if we imagine deploying that to a warehouse and the robot is doing it, say, just 10 times in a row, it's going to fail 80 % of the time because, well, every time it's a 10-20 % failure rate. So that's a very important thing in robotics. Just how can we kind of fill the gap between the frontier research and the deployment?

19:46And I see a lot of like VCs interest or excitement around deploying real robots out there these days compared to maybe several years ago, people are more excited because of ChatGPT. People are more excited around. Can we have a model that's capable of doing everything in the world? And maybe we can start some research lab. But there's definitely interest there as well.

20:05Isabelle Johannesen: But how much does the industry need to come together to really get to this next generation of these kind of robots? I mean, you guys, of course, are the AI software layer with natural language. But I mean, there must be some element of computer vision technology that needs to improve and the actual robotic arms that need to improve. How do you sort of collaborate with other players in the space? There are different things. So software, it's very important. And hardware, traditionally hardware is very costly. They are not cheap, especially the industrial arms. And that's actually a prohibiting factor in robotics because while people are optimizing cost, especially end customer, people care a lot about these three aspects, speed, cost and flexibility slash capability.

20:48We had to fulfill all of these three, not like fulfill immediately, but maximize this kind of triangle. And it was kind of an impossible triangle in the past. And that's also why a lot of companies failed. And this requires improvement in every part of the whole robotics stack, including software, to make the robot more flexible, more capable. And also hardware, make it fast. And also on the manufacturing side, we want the robot to cost a lot less than they were. And because of that, and all those improvements and progress can really push this out there to the world to automate a lot of things.

21:25Isabelle Johannesen: And so what do you do when the hardware stalls? And I don't mean like the literal arm is stalling in the moment, but like say the progress of that industry, you know, has a stall. I mean, that would impact your business, right? That's not really a concern at this moment because we are seeing really good progress in the hardware space. They are definitely getting cheaper and cheaper. And it's funny, I was back in China like a couple months back and I was so amazed by, on the manufacturing side, the speed of iteration and the cost, how they manage the cost for that hardware. It will only get cheaper and cheaper.

22:02And hardware is becoming more like commodities. And also, that's why we think software is going to generate the most value in the future.

22:09Isabelle Johannesen: So we were talking about how for hardware and robotics companies, it can be difficult to raise VC funds because it's not always the most rapidly scaling. What would you advise startups building in this space and how did you approach your fundraising journey? Well, yeah, we definitely know fundraising in robotics in hardware is hard, especially because, like we mentioned before, we haven't seen a lot of success in the robotics space in the past. But it's going to be different for sure for this generation. Every VC is kind of different and they always prefer different type of things. Some prefer more like humanoid, some prefer more foundation models, some prefer more vertical integrated solution.

22:48And I think the thing, one thing to keep in mind is to know what they want. The more important thing too is to like stick to what you believe in and show this thing is actually going to work and convince people that this is going to work. One, we know this customer or like these people have this type of problem. And actually we've proved that our solution or our technology can work to solve that exact type of problem. And starting from that, how this technology can scale even within that use case vertical or like expand more broadly. And if there's a good story, not just a story, it has to be true like around that, then I think that would be a convincing factor.

23:36Also, like I said, every VC is kind of different. For a startup, people might have to angle it a little bit, but I think do whatever you believe in.

23:47Isabelle Johannesen: And I mean, I think you guys are in a particularly sweet spot because your software is enabling some of these robotics that otherwise have a challenge scaling to scale faster and to be more useful. And so I think what we see is sometimes the really early stage hardware companies, I mean, they need to raise massive amounts of money even just to get off the ground with that R &D. How did you guys start testing? Was it because of the ABB partnership or did you have other opportunities to test your software? ABV partnership was a major one. And also a lot comes down to us validating what's needed in the field.

24:22So we were out there like visiting factories, warehouse and see, okay, these processes are all manual. Like they actually, our technology can solve that. And we formed a kind of a theory or assumption around how we can scale this and how our technology can solve this. So I think like all of that combined. In the end of the day, it's about de-risking the investment. And it can be basically deployment proven out there or have the assumption validated or have a piece of technology that's kind of validated by the market.

24:57Isabelle Johannesen: And so you raised a chunk of money for your seed round. And what, I mean, that was what back in June? June, July. Yeah, back last year. Yeah. What did you use that money for? And how did you, you know, how are you making it stretch? I would say for any startup, talent is the most important thing. The team. Because without the team, we can't build great product. And yeah, that's how we allocate our resource. We have a really capable team. One of our engineers is from the Bell Lab, and he's amazing. Very experienced in the AI research field. And while we're still expanding the team, speaking of that, let me know if you're interested in joining Embody and if you're interested in robotics and AI agents.

25:47Isabelle Johannesen: Careful what you wish for. You might get some inbound here. But what kind of roles are you hiring for now? Robotic engineer. Okay. Robotic engineer, that's the main role. And also AI engineer, familiar with agents, how to orchestrate different agents. So I take it you guys are still in the founder-led sales era of Embody? Yes. Yes. That's right. Yeah. Yeah. So let's talk about sort of the reliability. So if you have a human who is currently doing this job, obviously humans make mistakes, but they also, you know, these people have been doing this job for probably many years and know what they're doing.

26:17Isabelle Johannesen: How do you approach the conversation of trust and reliability with customers to make sure that, you know, the solution you're bringing in is actually better and convincing them that it's better than their humans? Reliability, I would say that's the most important thing in robotics. Because, say, you have a manufacturing line. If the robot messed up or stopped, then the whole line is going to stop. It's catastrophic. People are losing money every second, every minute when the line stops. So it cannot fail in production. That's the thing. And for us, how we do this is we have to prove it out that this thing can work really reliably.

26:54in our lab or in production that, okay, we run this for eight hours or even 24 hours. And here's the success rate. Here is the metric that's required for this type of job. And, well, we have to fulfill that. And, well, actually, we did a recent test run with our system, eight hours straight from beginning to end, 99.6 % success. So that's actually quite good.

27:20Isabelle Johannesen: That's pretty good. Yes. What's the comparison to the human success rate? Human also messed up. There's really no metric for how often humans mess up, but humans do make mistakes as well. Yeah. But I mean, your pitch to the customers is that Embody is making these bots more reliable than a human would be. Yes. So there are two angles here. First, it costs less. And second, it works nonstop and reliably compared to a human at relatively same reliability. So what is next for M-Body? What are your plans in the next couple of years? Where do you see this going? For the robotic field as a whole, what we believe is that there won't be a chat GPT moment for robotics.

28:01Some people might hope that there's a robot just show up and can do everything in the world. It's not going to happen in robotics because there's no internet of data in robotics in comparison to GPT. But we will see a gradual process of robot solutions being deployed so we can solve a certain amount of use cases or problems and expand, have the robot or the software to be more and more capable and go like that. And for us, I would say the most important thing for the next few years is to build a system, polish it up, deploy out there, and solve real problem out there in warehouse factories and collect data, feed that data back into the system and improve the system over time, kind of create a cycle like that because production data is always the best data possible compared to anything that's done in the lab.

28:52In one sentence, it's that get our stuff out there. That's the most important thing.

28:57Isabelle Johannesen: Yes, that sounds like a good goal to have. Do you think you'll be raising more funds in the future? Yes, we believe so. I'm curious about your take on sort of, obviously you guys are very much in the B2B space and the manufacturing space and industrial. What do you think about some of these consumer applications for robots? I mean, I have a Roomba, you know, I mean, everyone these days has some kind of robot in their house, even though, you know, they may not, not the humanoid doing their dishes that maybe we have imagined. But to some extent, we have robotics at home. Where do you think the next wave of consumer robotics are going?

29:27I actually, I like this idea. I'd love to have a robot in my home. I don't have a lot of, I don't have a lot of safety concerns. Well, me personally, but.

29:36Isabelle Johannesen: I want a little robot so I can take it down if I need to. but there's a really interesting robot by Hugging Face. I forgot the name, but it's like this big and people can, it's kind of, it's running a MCP server in there. It can do like some fun things around it. I see a trend of some kind of toy robots like that being deployed in a consumer space because it's still kind of unrealistic to say a robot shows up in my home and do all kinds of like chores and immediately. But I think we can see some toy-like robots. I know Amazon just acquired Fauna Robotics. Fauna Robotics made that also a little cute robot.

30:18That's very interactive. I feel like it's kind of a next generation of Alexa kind of robot at home and interact with people. It's like a really nice toy.

30:32Isabelle Johannesen: Yeah, I can imagine. There's a lot of applications that I think are more for fun and entertainment. There's also some that are sort of leaning towards curing this loneliness epidemic where you have a JAT-TBT like therapist, you know, running on a little bot that feels like you're talking to a friend. But then, of course, you also have these, you know, talks about the actual bots that would unload your dishwasher, the vacuum robots and whatever else, laundry folding, ones that might be more useful than just an entertainment. Well, fun is important for a lot of people. Yeah. I mean, and I remember there's a, I can't remember the company name.

31:04They make the Transformer type of robot, and it's just for fun, and it's very expensive. But people love it.

31:13Isabelle Johannesen: Yeah, and I'm sure we'll get to a place where everyone has some kind of bot toy running around their house. If only Grandma could see us now, she'd be horrified. But the rest of us are excited. I'm very excited personally to see where Embody goes in the next few years. I think you guys are going to be at the epicenter and forefront of the robotics transformation. So thank you so much for joining us and we'll see you again soon, hopefully. Thanks a lot for having me. This is a lot of fun. Build Mode is a TechCrunch podcast. Each episode is produced and edited by Maggie Nye and hosted by me, Isabel Johanesen.

31:47Isabelle Johannesen: Our art and design is also by Maggie Nye. A big thanks to Morgan Little, who leads our audience development, the Foundry and Cheddar video teams, and most of all to you, the builders and everyone else in the wider startup community. We'll see you back here next time. This episode takes time!