transcribe

Construction Robots on Real Job Sites | Yogesh Ghaturle, Co-founder & CEO, Origin

Core Matter: Physical AI, Robotics, Supply Chain · 41m · transcribed 3h ago
More from Core Matter: Physical AI, Robotics, Supply Chain Business
𝕏 Share ▶ YouTube 📥 PDF 🤖 .md

Section Insights

# 0:00

Misunderstandings in Robotics

What is the biggest misunderstanding people have about building a robotics company?

People often focus solely on the model layer of robotics without understanding the broader context of the business and technology involved. A successful robotics company must consider the full stack approach and market opportunities.

  • Customers prioritize work delivery over the technology itself.
  • A full stack approach is crucial when opportunities arise.
  • Understanding the market and customer needs is essential for success.
# 8:16

Opportunities in Construction

Why did you choose to focus on the construction industry for robotics?

The construction industry has seen minimal technological adoption and faces significant labor shortages, making it ripe for innovation through robotics. The increasing demand for efficiency and productivity further supports this choice.

  • Construction is an under-automated industry with high potential for robotics.
  • Labor shortages and rising costs create a strong demand for innovative solutions.
  • The right technological advancements can address existing inefficiencies.
# 16:33

Customer Expectations in Robotics

What are the key customer expectations for robotics in construction?

Customers expect high completion rates, human-quality work, and throughput that aligns with their ROI needs. If these criteria are not met, large-scale adoption of robotics will be hindered.

  • Robotics must deliver quality and efficiency to gain customer trust.
  • Meeting customer expectations is critical for adoption.
  • Robotics should be adaptable to various environments and conditions.
# 24:49

Economic Considerations for Robotics Deployment

How do subcontractors evaluate the cost of robotic solutions?

Subcontractors assess the value created by robots compared to human labor costs. They expect robotic solutions to match or exceed the value of traditional labor, factoring in any risks associated with adopting new technology.

  • Cost comparison with human labor is a primary factor for customers.
  • Robotic solutions must demonstrate clear value to be accepted.
  • Discounts may be necessary to encourage initial adoption.
# 33:06

Team Building in Frontier Tech

What qualities do you look for when building your team?

In a frontier tech company, it's essential to recruit individuals who are not only highly skilled but also passionate about the mission. The team should consist of superstars who can contribute to building advanced technology.

  • Hiring top talent is crucial for success in innovative fields.
  • Team members should align with the company's mission and vision.
  • A rigorous selection process is necessary to find the right candidates.

Transcript

0:00 And a customer does not care about a robot. The customer cares about their work getting delivered. What is the biggest misunderstanding you think people have today about building a robotics company? You can clearly identify a person who doesn't know anything about robotics when they only talk about the model layer. So let's go back to before Origin, you actually sold your first company to Amazon at 27 and came out with three rules for your next company. Tell me more about that.

0:37 Yeah, so, Michelle, when I actually sold the company, I was all of 25, 27 is when I actually ended up leaving Amazon, but, from my journey at Perpule, we had three core learnings, first of all, bit of background, like, we're building what Square builds for US, which is a cloud-based POS, but on top of it, they also build the hardware and payments, at Perpule, we were only focused on building the software, the cloud POS itself, so, effectively, we did not take a full stack approach, we shied away from doing hardware, we shied away from doing payments, right, and we only wish if we would have taken that full stack approach.

1:18 So, that was my first core learning. Always go all in on full stack approach when there is an opportunity to do so, right? Obviously, you can't force the full stack approach when there's no opportunity, but when there is an opportunity, take a full stack approach. That was the first learning. Second, for majority of our journey at Perpule, we were focused pretty much on the Indian market. We ran the company for four years, and for three odd years, we were only working in India, servicing retailers here in India.

1:47 Only towards the tail end of our journey, we actually started doing international sales, and within three months, we were able to acquire some very large customers in Southeast Asia, and we had around 20 enterprise customers here in India, one customer in Southeast Asia was equivalent to the revenue that we got from 20 customers by being in India, right? So, so the learning again, that was always be global right from day one. And then the third one which is the most important one, is that we were playing in a very competitive market, which was, like, building POS, there are, like, thousand companies which do that, so, there's no technology edge when you are building something of those sorts, right, you are playing on your execution capability, on your pricing, your business model, and things like that.

2:35 And that does not give you any unfair advantage compared to your, competitors, right? So, the learning from that was probably built something on frontier, right? Because frontier is difficult, so not a lot of people, do that, and then if you can sort of nail, frontier, you can be a very large company, right? Like all the companies which have been generational companies, all of them have always played on frontier, and that's why they are so big, right?

3:02 So those were my top three learnings, build a global, full stack, frontier company, right? And that is what exactly I did, when I started Origin. So global technical really makes sense. Like just from your experience how did you decide that, hey, it's full stack is something that an opportunity that you'd like to take whenever it's available. Yeah. So when we were building cloud POS, right, we obviously had to integrate with the hardware that particular software will get deployed on, and then the payment hardware that the POS would have to work with, also the payment APIs that the POS would have to work with, right? And in effect, we were dependent on, speed of execution and innovation that those hardware players and payments player that we were working with like our sort of future kind of were tied up with their execution capabilities or their speed or them being receptive to what we want them to innovate on, right?

4:07 So, in a way, you are giving the keys to your future to some other folks, right, and that's why I say that if you have an opportunity to own the full stack, do that, because then you control your own destiny, and that's why I sort of decided that it should always be a full stack company whenever there's an opportunity. How did you decide to build a company in construction? I think, when I had this whole thesis, that I want to build a frontier tech company, like, which is relevant for global markets, the sort of, figuring out on what are the right set of ideas took me some time, I spent, like, good, six months figuring out or betting on, like, what is the future going to look like, right?

4:56 Because when you're building a frontier tech again, you have to be slightly ahead of the curve and figure out, like, what is going to happen in the next five, ten years. And there were a few signals that I could really catch on to. The first one was when I'd left Amazon, the LLM era had just started, right? As you remember, like, ChatGPT launched in November 2022. I left Amazon in March 23. And within that period, I could really sort of imagine what kind of difference LLM is going to build, is to bring in the knowledge workspace. Obviously, the capabilities back then were not what it is right now, but you could sort of look at the journey and say that, hey, then the all of knowledge workspace is going to get impacted with LLMs, right? There would be agents which will do work for you, which will do knowledge work for you, right?

5:52 On internet, in your laptops, right? So I think that was pretty evident. I was super confident that is going to happen. But I, again, sort of went back and said, okay, what do I do with regards to full stack, right? And that is when I started looking at things at an intersection of hardware and AI, right? And I kind of came up with this hypothesis that hardware powered by multimodal AI will be able to automate physical work across different industries. Right, and the most relevant embodiment of this thesis is that you're probably able to sort of build and deploy robots, which will do physical work for you across different segments, right, so that was the thesis that I came up with, and then the other signal that I could really see was that the way LLM is creating difference in the text or, sort of that kind of, data.

6:45 It will also be relevant for images and videos and things like that. So, in a way, video and image models will become better, at a much faster pace compared to what has happened in the last few years, right? So when I combined both of these, it was very evident to me that robotics, powered by multimodal AI is, like, a bet to take. So that's where I kind of zeroed down on robotics to begin with. Construction as a segment kind of came out of my own personal problem. It was a lived experience.

7:18 So, after my Amazon journey, I, ended up buying a house. House, the house that I bought, unfortunately was in a pretty bad state, because the owner that owned the house never really cared about it, there was all kind of seepages across the house, and it was unoccupied for 10 years, so, like, they kind of never took care of it. There was all kind of seepages, bunch of windows were broken. There was a lot of, sort of structural work that had to be done.

7:47 But I thought, no big deal, it's a great house, let us get in, renovate, and we'll be able to move in. But unfortunately, in the process, we ended up making a couple of mistakes. The first mistake was, we believed the timelines that our contractors gave us. Contractors told us that, "Hey, this doesn't look like a big deal. You'll probably be able to move in in four months." That was first mistake. Second mistake was that I tied a lot of my life around that house.

8:16 So I was dating my now wife for eight long years. I told her, let's finish this house. Let's renovate this house and then we'll fix our wedding date. And no big deal. It's just four months. But we were in for a surprise. Those four months, turned into frustrating 14 months. So we were, like, behind the schedule by 14 months. And during that time, I spent a lot of time on the job site, right? At the construction job site, learning what does not work, in a construction setup, right?

8:46 And, I think a couple of things become very clear to me that obviously construction has not adopted any tech of any sort. Automation is, like, not even in picture, right, compared to any other industry. So, that was my lived experience that I had. Earlier, and when we came up with this thesis, we said construction sounds like, one of those biggest industries where there's been hardly any robotic adoption. There is a very large shortage of people across the world on top of it, with AI data centers getting built, manufacturing coming to home a lot of basis, there's just so much of demand at this moment, and the cost has been going up everywhere, productivity has been going down every year, this seems like a right set of sort of signals coming together to solve for it, right, and then we said, let's go all in, right, like, the technology is there, the demand is there, why not build a company to solve for that problem?

9:42 Did you look at any other industries? And then you decide that, hey, this is not as good as construction to go after. I think the other use case that we were very excited about was robots at homes. I think that was our second, sort of, best, favorite idea, like, in future, you can imagine that everyone will probably have a robot companion, whether that be humanoid, in fact, not just humanoid, we believe that there would be 10 different form factors that would come up in future, all of them different kinds of robots, in fact, like, you can actually put in multimodal AI, in all sorts of non-living things or daily objects that we use and everything is going to be smart, in some or other way.

10:25 So I think the second best idea that we, that we really loved was the home robotics space, but Yeah, we did not take that back. What does full stack mean for a robot company like Origin? Yeah, I think, there's no end to how full stack you can think as a company. But right now the way we define it is that you should be building, hardware, yourself, while you're also building the AI models, and the platform layer.

10:55 Now, when it comes down to building the hardware, obviously, most of the form factors, which are getting built in different spaces, they don't exist yet, and we believe strongly, that some of these form factors will be innovated for different segments. So owning up the hardware design, is surely part of being full stack, owning up the entire platform or software layer is being part of full stack, owning up the entire model layer is being part of full stack. And then owning up the manufacturing, again, I think is a big part of being full stack company.

11:33 Again, like the robots that are getting built right now are some of the most complicated, machines, that the humans have ever built, and there are not a lot of companies which are very proficient with manufacturing, things like these, so, we believe that, like, being a full stack company in robotics also means, like, owning up the manufacturing, so, right now, we own all of these four, you can keep going backwards, you can probably own up the actuators, like, own up the chip layer, but I think, that's for future, we don't own up those parts, but yeah, everything that I defined we still take care of.

12:09 That's true. You can really go down to, like, even the materials and all that, too, when you think about the full stack. I think one of the things that people struggle with when they think about full stack is how do you prioritize which layer of the stack to start first and how do you optimize for the go-to-market and deployment timeline? So talk to me more about how you guys think about it internally. Yeah. Yeah, I think, those are the right, trade-offs to make, Michelle.

12:39 So, you have to be focused on, how much time are you ready to sacrifice to own up a particular layer. We have always been very deployment focused, and, we always wanted to ensure that we can go back, or we are in the field as much as possible, as soon as possible. That meant that we had to make some trade-offs, during different times of our journey. And right now, in a period of last 18 months, we have built three different generations of right now, the third generation is already in our office in the validation phase.

13:11 And in each generation, we made different trade-offs. The first generation was obviously about, okay, how soon can we get on in the field, right? That meant that, we said no to not building our own controllers. We said no to not, fabricating some things in-house, and, like, in generation two, and in generation one, again, there were a few capabilities that we said no to, right, that, hey, there would be certain part of the workflow that would be done by the operator, right, and those were the trade-offs that we made for generation one, right, and across all the other three generations, like, we have been making different sorts of trades.

13:47 And every generation that you sort of build, you keep getting more and more full stack as a company. Now in generation three, we have built our own manufacturing capability. Right, that was the trade-off that we made for generation one and generation two, but generation three, we did not want to make that trade-off. So, again, it all comes down to what are your priorities as a company, how soon do you want to be on, in the field, and things like that, and then you pick up, what is possible as a company, for you, and then, sort of, move forward from there, yeah, I think that is how you should think about it, how soon you want to be in the market, how much capability you have, what kind of talent you can hire, how soon you can hire them, and, like, get stuff done.

14:36 There are dozens of interior trades when it comes to construction. Why start with finishing walls? It's a great question, Michelle. So I think there were top three reasons. One, wall finishing is an extremely repetitive, labor intensive trade, right? Second, when a professional does wall finishing, they are taking a lot of safety risks. For example, they are up on the ladder holding, like, big equipments or drywalls, second, when they are doing sanding, they are effectively inhaling a lot of cancer-causing sort of air particles when you sand the wall.

15:10 So it is a health and a safety risk. So, from, for a customer, right, it's a big pain to hire for, right, so, that's why there's a lot of shortage in this particular trade, but on the other hand, when you look at, from an autonomy perspective, majority of the time, robot is dealing with 2D surfaces. And the robot is doing work which does not require a lot of dexterity. Right, so, it is slightly more easier to automate that kind of work, right? So, these were the top three reasons, right?

15:45 On one end, it's a labor intensive repetitive job. Second, it's a health and safety risk, while it being conducive for the autonomy, right? So all of these combined, we felt like that this is the right trade to go after. So when first generation is an OG-1, right? They go on to the sites to do their first job. What was the most surprising thing for you? I think there were just so many different learnings.

16:17 I think the biggest learning, that most of the team, came back with was that a customer, does not care about a robot. Customer cares about their work getting delivered, right? Because they are functioning in a live setting, there is a job to be delivered to their own customers, right? So, we work with subcontractors which are delivering work for general contractors, right, so, there's no room for error, right, it's real work that is happening wherein our customers are answerable to someone else, right, so, they could care less about the robots, right, all they want is the job done, at the quality of a human job done completely and with a throughput that is conducive to their timelines, right?

17:08 So what we realized by doing our deployments was that you have to, as a robotics company in a live setting, you have to deliver on all of these three fronts. You have to have very high completion rates, you need to deliver human quality, and you need to be able to sort of do work at a particular throughput that it is conducive from a ROI perspective for your customer. And if you don't solve for all of it, there would not be large-scale adoption of your robots.

17:47 So this was the most important learning from a customer POV. The second learning was that how should we think about our architectures, right? And over there, the revelation was that whenever you are designing anything, whether that be from hardware perspective or autonomy perspective, you have to think as broad as possible. Robot, both from a hardware and autonomy perspective should generalize across all kinds of environments. For example, the robot will probably work with 10 different kinds of layouts, the robot will work with 10 different kinds of lighting conditions, the robot will work in day, night, a commercial space, a residential space, so the robot should work across all of these different kinds of setups, right?

18:29 There's no room for saying that the robot will only work in this kind of setting, because, again, customer couldn't care less about that being not the case. Same thing was for hardware. Robot should be able to enter all door sizes. Robot should be able to do, to work on all kinds of surfaces, whether there is a bump, whether there is any other, sort of, different site condition, right? So, build a technical architecture which is generalized across all kinds of environments that the robot would have to deal with.

19:01 So, I think these were the top two core learnings that we got during our deployment. So speaking of throughputs, you mentioned somewhere that your robots are able to achieve three times the speed across the whole shift, setup, maintenance included. How did you measure that? So, effectively what we do is, we compare it against the human productivity. So, for the same amount of time the human professional does the job for what are they able to achieve versus when a robot, does that job. Including the time the robots spends on, any maintenance work, any breakdown, any reliability issues, all things put together, how does it compare against the human, professionals shift, right? That's how we do the math.

19:54 And who were the first customers for Origin and how did you get them to say yes? So, the first customers for us were subcontractors, which do wall finishing in New York. We got them, because one of the first person that we hired in our team was someone with construction background, who had spent 10 years of his life, being on the other side of the table. He was, project head for a general contracting firm, for about seven to eight years, and, he brought in his network of customers, along with him, and since he knew them personally, we were able to sort of convince them to be our design partners, and sort of allow us to like, put our prototype on the job site.

20:39 So yeah, that's how we got about it. And as I said, like we worked in New York, in fact, within New York also, we're very focused on the Brooklyn. We're working out of Brooklyn for most of our sort of time, during those first few pilots. Wow, and how many robots working today and how many square foot have they finished? So, right now, Michelle, we have around six robots, over a period of last few months, we have done about 100,000 square feet of work, at different job sites, and now we are scaling up our deployments from here on I think every month, we plan to add, like, anywhere between 50,000 to 100,000 square feet of work done, on a regular basis, and then I think now that we have manufacturing capacity solved for, we also will be able to manufacture five to 15 robots a month, right?

21:34 So we'll increase our deployment scale and velocity as well. Well, congratulations. That's a really big ramp up. And you mentioned that the deals average around 2 million. How many robots? How many months is that? So, the few contracts that we have signed generally have been with the SMB customers that I was speaking about, we charge on a rental basis. Every robot is like around $20,000 a month. And these contracts of about two and a half million dollars, they last for a year, right?

22:10 So $20,000 per month for 10 robots. And you multiplied by 12, you get two and a half, odd million dollars, so that's how it works. Got it. So, it doesn't matter how much the customers using the robots every day. It's just a flat monthly fee. So, see, if the customer does not use the robot, they will give us back, in the robot, right? That means the robot is not useful. But, generally, we are working with customers who have worked throughout the year, and, it makes sense for them to sort of rent it, on a monthly basis.

22:44 And if all the months they end up using the robots, the overall contract value would be two and a half million dollars. How long do you expect these robots to last in terms of on the field? Let's say they're doing their job on an expected basis. And, you know, is it a year, like two years? I think, we predict that our robots will probably last between four to five years, and we are doing a lot of work on the reliability, on the serviceability, a bunch of things like that, to ensure that we, the robots have a life of five years.

23:21 But, and we have obviously done a lot of HALT testing, to sort of be very confident about it, but I think, like, as in, when we do more and more deployments, we will, come across more challenges, and issues to make that happen, but we are very confident that we'll be able to sort of resolve them, build systems in place so that our robots can be there in field for around four to five years. Got it. And in terms of the 20K per month, is that a global pricing?

23:50 Because you did mention in the beginning of your call that, hey, we're first company, you sold it to a different locale, and the pricing is so different. Is New York your biggest market right now? Like, what's the split when it comes to when you ramp up the deployment? Yeah. So, Michelle, I think, the pricing on an average in U.S. And Europe would probably remain in that range. And again, as I said, like, we are averaging it across, like, the capacity that people may want, so in some cases, we'll probably give some more discounts to customers.

24:22 When they are getting more robots on the sites, and then in some cases, because we would also be delivering, a higher throughput, you'll be probably charging more, when the throughput increases, but I think around about 20 to $30,000 a month is what, we'll be able to charge to our customers. And as long as we kind of, continue improving our throughput, we can probably in future think about doing more, charging more. But yeah, right now, 20 to 30 seems fair.

24:52 Look at this pricing from the subcontractor contractors perspective. So how do they arrive at that, like comfort to pay that 20, 30 K a month? Is that from their perspective, are they replacing like one or two of their, human labor or how do they do the math? Yeah, absolutely. I think that is how they look at this Michelle. Again, as I said, like, customer would care, they don't care about robots, right? They care about what we're getting done and what price ideally they would have paid otherwise.

25:25 And, that's their comparison factor, so, if a customer is paying us $20,000, they expect that we would do the work that they would have ideally paid $20,000 for, right? And, in some cases, because, they are also, taking a chance on a new company, a new platform, they certainly want certain discounts, right, while they do that. But yeah, that's the simplest math in their head they are doing, that what is the value getting created by the robot getting delivered, and, would they have paid, lesser, if they had a human crew. If that is the case, obviously, that pricing will not fly to at least have to match what they would have paid otherwise.

26:09 Awesome. And one of my favorite things that you shared about how you think about the economics of deployment is the distinction between task and workflow. And you wrote that tasks are model evaluations and workflows are customer evaluations. What do you mean by that? So, Michelle, I think if you look at any industry, the crews that do work in the real world setting, they do, multiple things, right? They are not just doing one simple task, at the end, right, for example, a wall finishing professional, he can obviously move up and down across the job site, can take an elevator, he can enter a particular apartment or a commercial space.

26:52 He can move from one room to another room, he can, like, move from one wall to another wall, he can identify, a window, a door, he can identify, like, where are the seams located, he can figure out how to use the tool, he can climb up the ladder, he can do the ceiling, he can do all the edge cases, so, there are so many different subtasks, right, that the human professional does. So, when a customer looks at a robot, their expectation also is that the robot will be able to do all that a human professional would have done, right? But if you look at most of the robotics companies, they always are very focused on what are their evals looking like on a particular task, right?

27:42 They never look at a workflow. Just folding a tablecloth is a task, right? But getting that tablecloth out of a washing machine, then folding it, then putting it into a basket, then moving it to an area where it is going to be useful. That is a complete workflow, right? Picking up a tool, moving from one room to another, applying a compound, sanding on top of it, moving up and down across the wall, working around the humans, doing up the ceiling, going and figuring out, how it should move from one room to another, that is a workflow, right? So when a customer looks at it, that is their evaluation, right? That, can I do everything?

28:23 So that's why I say that the tasks are, like, robotic companies' evals and then, workflows are customer evals. Walk us through one example when a robot hits a problem on sites and how did you turn that into a workable workflow for the whole fleet? That's a great question, Michelle. So, I think we've seen a bunch of different, problems that we sort of have encountered for the first time, at a job site. I think something that we literally faced on day one was a very funny scenario. We were working, in New York.

28:59 We were on the 19th floor. The robot was expected to finish a particular room in an apartment. There was a window, on the right. And the building actually was west facing and towards the end of the day, obviously, when sun is setting, and a robot was right in front of that window, like, doing the walls around that window, there was a lot of glare that was coming from the sun itself, right? And when the robot was evaluating the wall, and figuring out which part of the wall it should work on, the models just stopped working, right? Now, and this was a scenario that we literally came across the first day of our deployment, like a west facing building, in the evening, the robot is just right in front of the window, and the model kind of gave up, right?

29:50 When we came back, we figured out, okay, then what, like, should work in this scenario, right, even if there is a glare, there's a window, wherever the sun is whatever may be the reflection, it should continue to function, so, obviously, I did all of that back into training data, and then figured our way around it, but, yeah, I think that we've encountered a bunch of such scenarios wherein, robot couldn't do anything and then we had to build up a solution and put that back into the model there.

30:18 That's awesome. And that's also to illustrate how important going to the field early is, right? To understand these edge cases. So after, the wall finishing task, what trade comes next for Origin and why do you pick the next one? Michelle, so, in wall finishing, So, you start by putting up a frame, you put up a drywall, then you put up a tape, then you apply compound, sand it, prime it, paint it and then inspect it, right?

30:48 So it's a eight step process. We only do compound application sanding and inspection right now, along with priming and painting capability, right? So we have unlocked five of the eight step process at this moment. So right away, we are building capabilities to do the taping. Drywall installation and framing, so, these are the top three priorities, because we want to do everything in wall finishing, because we would be selling to the same customer, right?

31:20 Would love for us to do everything, right, all the trades, in that entire, wall finishing gambit. So that is our immediate priority. On top of it, we are also unlocking a lot of MEP trades. We are planning on doing things like, drilling, across different sort of environments. So MEP comes next, now on MEP front, we've been kind of keeping it close to our heart, like, we are working with a few design partners, and we are working, unlocking, those workflows, so I think I'll have more to share when we kind of launch, or unlock that capability on the platform, but the ambition, Michelle, in the long term is that we will do all things interior construction. That's how we build up the platform, yeah, but in next year, finishing the framing, drywall install, taping, and then a few of the MEP trades would be the priority.

32:12 You guys also have this construction action model. Let's talk a bit about that and how much of that model carries over from one trade to the next. Yeah, that's a great question, Michelle. So, first of all, the, talking about construction action model, I think we have, we have a state-of-the-art architecture or frontier architecture on these models. A model can be divided into three parts. There is a reasoning layer on top, which also acts as an orchestrator. The reasoning layer is deciding, what, the robot is seeing and what decisions, it should take.

32:48 Now those decisions are effectively taken by the action expert layer. The action expert layer, is built with a bunch of small, experts. Each expert is efficiently capable of doing a subtask in a entire workflow, right, so, as I said, like, there are 15 different subtasks in a workflow, so, action expert, like, we train different action expert to different part of the workflows, and then we have controller layer, when the action expert decides on the action plan, the controller executes that action plan.

33:21 So that's the model architecture that we have at our end. This is the construction action model that we're building. Now, the best thing about this architecture is that while, you look at two different trades, you want to unlock workflows across both of those trades, right? And 80% of the subtasks in both of those workflows in two different trades remain exactly the same. Right, the workflow may be navigation across the job site, moving up and down, on a wall, working on the ceiling, figuring out what is what at the job site, right?

33:58 All of those subtasks remain exactly same. So around 80% of the model carries over, to the next task. And, in a lot of cases, because the model has built intuition around how to come up with action plans, It is much easier to train for the next task. Right, so, for example, in the first year, we were only able to do two tasks. Now the robot can do seven tasks, right? So within a year, we went from two to seven, and I think next year we believe that we'll be probably able to do 30 to 40 tasks, right? So that's how sort of the architecture looks like, and then that's how much of it transfers to the next set of tasks.

34:43 I love that. So let's go to the team. You mentioned that you went through 5,000 interviews to build the current team that it is today. What were you looking for in the people that you bring on to the team? I think, Michelle, we are a frontier tech company. You want only superstars in your team, right? You want people who, first of all, believe in the mission, to begin with. Second, people who are extremely skilled, and the sort of really build the frontier technology that means that they need to be extremely smart and then while you are doing this, you need to be also extremely, passionate about what you're doing, right, and need to really work very hard at it.

35:31 And on top of it, you have to really care, right? Like you have to care about the customer, you have to care about the product that you're building. So I think some of these were our criteria we have been interviewing people. We want people who are in for the mission, who are long-term players, who are extremely skilled and extremely talented to build on frontier, right? And that's why we've gone through, like, more than 5,000 interviews now, to build a team that we have.

36:00 Tell me more about that story. And how much does the ceiling of the people around you matter to you? I think that has been defining part of my journey. Every time, I've been able to surround with myself with more ambitious people, it has only helped me as an individual to aim for larger things in life. The teacher that I had, in the childhood, obviously, was someone who really made a difference, who really, sort of, give me that belief that I should aspire for more.

36:32 And then I think that has stayed true consistently throughout the journey, right? I've had a few other people who played that role. Whole ambition now is to sort of keep surrounding yourself with those people, right, who push you more, who help you, sort of aspire for more, right, in life, right, that's what I keep looking forward to, yeah. What role do you see India playing in robotics over the next five years? I think India is going to play a pretty central role.

37:06 One of the statistics that not a lot of people are maybe aware about is that if you look at the robotics engineering community, There are a lot of Indian engineers, who have been excited about building robots. Some of the very best robotics companies are, founded by Indian founders, Skild AI, it's a completely Indian team, a bunch of other robotics companies which have done very well are driven by Indian founders and Indian teams.

37:36 I think India as a country, the engineering talent in India, is extremely excited about robotics. So, I think the biggest contribution is surely going to be from a talent perspective. Second, I think India is a good training ground, for, sort of, collecting more data, third, I think India has a very unique, ecosystem in manufacturing. Obviously, it is way behind what exists in China. But I think given the overall, geopolitical sort of situation, maybe India will be slightly favored when it comes down to manufacturing robots in some case in future, right?

38:14 So I think, India has a big role to play, on all of these fronts. What is the biggest misunderstanding you think people have today about building a robotics company? Oh, no, I think that's a tough one. There are just so many things that people get wrong. You can clearly identify a person who doesn't know anything about robotics when they only talk about the model layer, right? You would hear so many people just saying, that, hey, hardware exists. You just have to solve for the autonomy and you're done, right? Once the autonomy is solved, you'll only have robots around you. And I think that's a giveaway for people who have not really spent time in robotics, right?

39:00 People don't understand how full stack you have to be a robotics company, you have to build on model layer, you have to build on hardware, you have to be really good in manufacturing, you have to be really good in supply chain, you have to be really good in deployments, right, you have to be really good in selling the aspiration, right, to your customer about why they should adopt robots. So I think that is what really people get wrong.

39:27 People believe once you solve the model layer, everything would be solved, but I think that couldn't be further away from the truth. Yeah, definitely. Think, the people that think about building a robotics company without touching robots will not last very long. Everyone needs to get their hands dirty. So, what would you tell an engineer in a small town in India who wants to build a robotics company down the road? This is a great question, I think you just have to believe, someone told me this very simple line, he was, like, when I started the robotics company, it always felt like that building a robot is going to be as difficult as rocket science.

40:09 But if you are at it, right, you would be able to figure your way around it, right, as long as you are really in it for solving the customer's problem, really in it for the mission, you would be able to figure out a way. It is difficult, but it is not impossible. And I think that is what everyone probably needs to keep in their mind, right? It may feel like a very very complicated frontier, but then as long as you are at it, you can do magical things.

40:39 I think we are an embodiment of that, what we've been able to sort of do in the last 18 months with with very limited capital, it's been, sort of really great to see how team have come together and delivered, and yeah, and that is what everyone should think, right, as long as you have that, customer love and passion, you would be able to solve for the customer problem, and don't, like, think too much about it, just go for it.

41:07 I love that. I think this is a great place to close that. And thank you so much for joining you guys and sharing your story, both about Origin, but also like all the lessons beforehand as well. So thank you. Thanks, Michelle. It was great being here on a pod with you.

Summary

The discussion centers on the journey of building a robotics company, particularly in the construction sector, highlighting the importance of a full-stack approach, understanding customer needs, and the challenges of deploying robotic solutions in real-world settings. The speaker shares insights from their experience, emphasizing the need for robotics companies to integrate hardware, software, and manufacturing capabilities while focusing on delivering value to customers.

- Customers prioritize work delivery over the technology behind it; robots must meet high standards of quality and efficiency.
- Key learnings from previous ventures include adopting a full-stack approach, targeting global markets from the start, and innovating in less competitive sectors.
- The speaker's personal experience in construction highlighted the industry's lack of automation and the potential for robotics to fill labor shortages.
- The robotics company, Origin, focuses on wall finishing as a starting trade due to its repetitive nature and safety risks for workers.
- The company employs a construction action model that allows for the transfer of knowledge and capabilities across different trades, enhancing efficiency.
- Building a successful robotics company requires expertise in hardware, software, manufacturing, and deployment, contrary to the misconception that solving the model layer is sufficient.
- India is positioned to play a significant role in robotics due to its engineering talent, manufacturing ecosystem, and potential for data collection.
- The speaker encourages aspiring engineers to pursue robotics, emphasizing that while challenging, it is achievable with dedication and a focus on solving customer problems.

Questions Answered

What is the biggest misunderstanding people have about building a robotics company?

People often focus solely on the model layer of robotics without understanding the broader context of the business and technology involved. A successful robotics company must consider the full stack approach and market opportunities.

Why did you choose to focus on the construction industry for robotics?

The construction industry has seen minimal technological adoption and faces significant labor shortages, making it ripe for innovation through robotics. The increasing demand for efficiency and productivity further supports this choice.

What are the key customer expectations for robotics in construction?

Customers expect high completion rates, human-quality work, and throughput that aligns with their ROI needs. If these criteria are not met, large-scale adoption of robotics will be hindered.

How do subcontractors evaluate the cost of robotic solutions?

Subcontractors assess the value created by robots compared to human labor costs. They expect robotic solutions to match or exceed the value of traditional labor, factoring in any risks associated with adopting new technology.

What qualities do you look for when building your team?

In a frontier tech company, it's essential to recruit individuals who are not only highly skilled but also passionate about the mission. The team should consist of superstars who can contribute to building advanced technology.

© transcribe · For agents Built with care and craft by Gokul Rajaram