In this episode, Audrow Nash speaks to Maria Telleria, who is a co-founder and the CTO of Canvas. Canvas makes a drywall finishing robot and is based in the Bay Area. In this interview, Maria talks about Canvas's drywall finishing robot, how Canvas works with unions, Canvas's business model, and about her career path.

Start

Introducing Maria and Canvas

How the robot works with tapers

Becoming drywall experts

Challenges of spraying

Labor shortages

Challenges of sanding

Mapping the construction site

Picking your hard challenge

How Canvas is funded

Building for IPO

Working with Unions

Addressing misconceptions about robotics

Future of Canvas

Getting the robot to the construction site

Reporting the work done

Maria’s career path

Challenges coming from academia to starting a startup

Leadership experience

Advice to a 20 year old

Future of robotics

Links and contact info

This is a conversation with Maria Telleria, who is a co founder and the CTO of Canvas. Canvas makes a drywall finishing robot for construction sites. One thing that I think is especially interesting about Canvas is that they are working closely with a trade union. In case you don't know, quoting from the AFSCME, a large trade union in the US. A union is an organization formed by workers who joined together and use their strength to have a voice in their workplace. Unions help workers negotiate for better wages, benefits and job safety, among other things. I'm not very knowledgeable about unions. But my impression before this interview, was that unions often slow the adoption of new technology in a trade. And I think most robotics companies I've talked to avoid or try to automate unions out of relevance. So I was surprised to hear how Canvas is working closely with a trade union, and that both canvas and the trade union seem to be benefiting from the collaboration. In the future, I expect to see more robotics companies following Canvas's lead. Since I think working with unions can help robotics companies understand their problem, get into industries much faster, and avoid resistance from people already in the trade. At the same time, I believe unions can benefit from automation in canvases case, their drywall robot helps to reduce the rate of repetitive injuries that end people's careers. And they help a trade that is significantly short on people to meet their demand. I'm Audrow Nash. This is the Sense Think Act Podcast. Thank you to our founding sponsor, open robotics. And now here's my conversation with Maria. Hi, Maria, would you introduce yourself?

Yeah, Maria Telleria. Yeah, I am co founder and CTO of Canvas. And we're a robotics company bringing drywall finishing robots to the construction industry.

Now, tell me about your drywall finishing robot.

Yeah, so when we, you know, we come from a research background, and we've been always looking at why are robots used? In other words, they are used, but not really widely use outside of manufacturing and logistics. And read I think is a challenge that probably many in the audience know that construction environments are much more difficult for robotics. And so my co founder, Kevin, Albert, and myself, were always driven by like, hey, let's bring robots to another place. We got fascinated by construction, right, the size of the market is huge, the problem, but then, you know, you kind of keep digging and digging. And that's where we found drywall finishing. And we pick this test, because it is particularly painful in terms of like a lot of wear and tear on the body. A lot of like repetitive steps, you know, long schedule, even if the work is not consistently being done, because you have drying steps. So really, the goal was like, Okay, this seems to have all the adult dirty dangerous for robotics. Let's go at it. And so yeah, robot is, I keep saying tool because it is really meant to be an augmentation of the worker. So we do have a worker in the loop, they are a key part of the component. They're a key part of how we actually solve this problem, the unstructured environment. But it is, you know, this robotic system, it's a robotic arm with very specialized and the factors that allowed it to spray on the joint compound, which is where you apply on the drywall sheets to cover kind of the seams, and then a sanding and detector that sends it down so that it looks smooth. And that robot arm is on a mobile base that allows the right to reach all the different places in the construction site.

Gotcha. So what does it look like when the robot is actually doing its spraying and sanding? You kind of just position it near a wall and it goes in slowly covers everything with its arm? Or how does it look?

Yeah, I mean, maybe this is where a picture's worth 1000 words. So pull up the video to share it you. screen here, you know this is pretty much a quick peek. But I'll describe it as we go right now really, you know, starting with the idea that there is a lot of construction going on. This is what the job site looks like when we come in. It has been taped. So you have to apply the state between the jump the different drywall sheets that you come in, and you go ahead and install that spray and the factor that you see there. A big part here. I'll pause it to talk a little bit about it. Because you mentioned like, right you position it, like keep our promises we have to build our map to make this happen, right. We don't have a map when we start you know there is things like bid models. So the models the building information model. It's kind of like the had construction sites, right? They do exist, they're pretty detailed, they have everything done that. Rarely, by the time we come in are those that accurate because nobody's updated when like, things had to move a little bit, you cannot make that decision as long as not something structural people can move a wall a little bit because of whatever recent happened on site. So one is not accurate in two, they don't often get shared all the way down the chain of people who are working right there kind of more useable design and modeling stage rather than in the build stage. So most of our customers just have a floorplan, so 2d plan, and they say, right, I gotta finish these walls. So one of the key things that we implemented was said, we can't depend on them having a male have been mobile, excuse me, and we can't depend on the BIM model being accurate. So we're going to have to map ourselves. So basically, the robot comes in recruit scans to space identifies the different areas that it has to treat. And then a key part here is that it splits them up into workspaces, right, this goes along with the fact that the robot arm can only reach so much right just like your arm, you at some point, you have to use your legs to move along. So by splitting up into these workspaces, we really think of the mobile base that you see this pretty big base position said roughly to this spot, and then the robot arm can use all of its precision to go ahead and target it in a blank here, you see our operator select the workspace is the rows, columns that he wants to do. And then you see how it sprays the material work life is very consistent code across and making sure that we target the low spots and and then by the end of it, right, you have a full coverage of your machine. This part we can touch on later. But one of the hardest parts with construction is reporting back on what's happening, right, a lot of the problem is when work is done manually, you have to kind of call people up and get kind of an update on what happened. And really report them back. I think that's one of the secondary effects of robotics and logistics and automate in manufacturing is that you also get that data for free. So we're just showing a little bit of what's possible here. When the robots in the loop, you know, the supervisor can get updates that are live and they're very accurate. And then the next day, same robot arm mobile base, different end effector. So we put the sanding header comes in, it uses the same map that we created the day before, and it starts standing away. And this is, you know, again, a lot of our development has gone around the sanding, because it is a very soft material, right? If you reach out across the wall, you could scratch it with your nail, it's so soft material that you need to make it look really flat. So you have all of these things that you have to work around to make sure that you get the night consistent finish. And this is again, talking back to the data reporting that you can get back. This video is a little, I think, from 2020. So this has 25 projects completed, but we're now in the more like 40 plus range. Oh, that's so exciting. And doing.

Awesome. The video was very informative. Oh, yeah. Okay. So let's see, what what exactly does the Operator do?

Yeah, so I think the operator really does all the things around it that allow the robot to be successful in a structured and dynamic environment, right. So they do things like selecting which areas you want to skip, right, there might be a reason that you don't finish that right then and there. They can provide feedback to the machine as well on the quality so that the machine can tune in and get that feedback process from it. You're also doing a lot around coordinating around with other trades, right? So the reality is the electrician is also in there, you know, the floor guys coming in the painters right behind you. So that coordination with all the different people, but in reality, we they really are the ones that at the end of the day, they're the quality assurance, right? They make sure your feedback or even sometimes a little bit of manual work to ensure that the final bullet so that the machine really takes the 90% plus of the brunt of the work. But they're the ones who are skilled who have this. So our operators are tapers, they're journeyman takers who have been trained and the skilled and so they can pick up a tool if they need to, like a manual tool and fix up anything. But more importantly, they have that eye in that training to say, Okay, this needs a little bit more here. A little bit less there. Okay, that's how you get a wall to look.

Yeah, so you said they're called tapers. So that's the construction role, like job role for someone who would do this. It's really interesting to me that you are having the person who would do this normally be the robot operator. Can you tell me a bit more about that?

Sure. Yeah. And I like spend a little bit on the call papers because you string this paper tape across it so right that's where the name comes from. There you know it's more like drywall finishers. It's kind of the more formal term but everybody in the construction side be like oh the tape you know, construction they ended up getting to that to the point do you string tape your tape? Up to your point the plug why? Right because we can have gone a lot of different ways. But as we really thought about it, again, to ensure that we get that quality finish, we needed that expertise that skill level, there's going to be things that may never make sense for the robot to automate, right? Imagine somebody decided to make like a gothic column out of drywall, you could do it, I don't think I'd want to spend the time programming it because you'll never see it again, right? But a worker has the years of experience that delicate touch that adaptability that a robot doesn't do. And it's okay, I'm going to start the robot here, finish the brunt of my work. And I'm going to use my skills where it's really needed, right to provide that feedback or to finish some delicate part of it. And for us, really to be successful there. You needed somebody who could work with a robot, right? So it's like, oh, we have this great, skilled, talented person, we augment them. That's the way to go. Right? As opposed to maybe bringing in somebody who's like a robotics expert. Okay, maybe the robot kept running, but not sure that well, we?

Yeah, for sure. Okay, that's so interesting. And so what are some of the challenges? So if you brought in a robotics person, the robotic person may not know anything about the wall, but can keep the robot running, as you're saying. But so if you work with a taper, then, like, what would have been some of the challenges of working with the taper kind of bringing the robot to the craft instead of the craft to the robot? is what it seems?

Yeah, yeah, that's a really good way to put it, it really is making it super easy to use, right. And so because we want their time that they have, right, where the robots running autonomous to be doing other things, like we said, coordinating, tackling tough spots. So it's really about like, making like a tool, right, they learn it, they do come to a training and Canvas, but after that, they should just keep on running. Right? They it's sort of be asking them things in a robotics manner, right? Like, would you like to increase your workspace? Right? That might be a little bit or like, Hey, your inverse kinematics failed to solve like that, that would have mean something to them, but something like, Hey, what's the body look like to you? Right? And that's more your, I guess, specific example, for example, we wouldn't be like, is the thickness, you know, 1.2 millimeters as specified? They don't, that's not what it works. But we could say like, Hey, do you think this needs more material added? Right? Yeah. Are there most faults? Are they have spots, that's language that they know, that then we can translate into? Okay, they said there was low spots, we gotta build up more material here, you know, go from 1.2 millimeters to 1.4. So I think it's a lot about think about translating things to the, you know, the words around them that the craft words, if you will that make sense. And then really making sure that your robot is a tool, right? Not something that has to require programming. We also haven't really kind of on site support within the user interface. So when they do run into these problems, which they might, there's somebody just on the other end, to keep them running and to do that kind of debugging that you'd expect a roboticist to do.

Gotcha. So what does the whole step setup look like in terms of even the infrastructure behind the robot? Like, I guess it gets to the business model? Is it like you lease the robot to a company and then you surprise the service? Or how does it work?

Yeah, we are doing a leasing model for the customer. So they come to get trained, they trained their operators, and then they lose the machine, you know, for year long leases. And then they have different projects that they would move in, we help them with the logistics of moving the machine since it's something new, but from the other part, right, they pay per day, and then they have access to all the support around so they have remote support that they can call that's usually the what they would normally use. But if you know something really big happened on site, there's also people from campus who can go and support them on site. And it was a problem if needed. It's very rare. But it is something right now, we're only in the Bay Area, because one of the things we wanted to make sure is like, be really successful in a small radius, such that you can ensure that your customers get their work out the kinks, keep them running, for sure. Actually, when we first started, we did the service ourselves. So we became drywall subcontractors, and hire tapers onto our payroll, and have our own people running it right. And that was really important. Because for the customers, I'm excited about the idea of a robot. I'm not really sure how it's gonna work for me. So we said, Fine, you buy drywall at $1 for you know, certain dollars per square foot, a canvas the same amount. And we'll get it done, right. We just got into AP robots that do it with our operators. So that was the first two, three years, once they started working and really like saw how it works, and they're like, oh, okay, this works into my workflow. They started saying, Okay, now we want to run it right? We want to scale up with it. So that's really the transition point we've entered around the middle of last year is really saying, okay, the robots that a state that you guys can run it will rent it to you and we'll help you make sure it's successful.

Gotcha. That's really Cool. So it sounds like it was extremely iterative to get to where you are now, would you? Would you tell me a bit about that like, so now, I assume you guys are somewhat of drywall experts. But like, you have this problem, and you think it's going to be a valuable good big value add kind of thing. And then you start in this way that you were talking about where you work with the contractors, and you basically become the contractors, but hire people to amend your robot. And then growing that to the point where it's like useful enough that you can just give it to that contractors to do with themselves. Like, just tell me a bit about that whole process?

Sure, one of the big things from the beginning has been testing on site as quickly as possible, right. So the very early tests, like, I mean, probably now I'd be embarrassed to call that a robotic system. But you know, it has to be moved manually

of our team members is much stronger than being so he had to come to every site because exactly how to position the robot. But it was a way to do point break on paper. And in the lab, there's a lot of value. But we all know that the real issue is this. So we took out that, you know, cart with a robot arm demonstrated that they kept iterating, right? We're like, Okay, we can't hire, you know, just a person to move the robot. So we ended the mobile things, and all of that. In with that we did learn a lot about drywall finishing itself. But even before very early, when we're going to do drywall finishing. Yeah, let's hire somebody who knows what the drywall finishing is. So we brought on team members who, you know, years of expertise, you know, journeyman unique people themselves who just been trained in that skill. And when I sat next to them, I really said, Okay, how do we get to that level? The really interesting part, their radius, and we were both learning from each other, right? Because people have optimized the street incredibly right standings that were sparked. So they bought to my stick to reduce the amount of Sandy. You know, so when we've reached brought them in, and what I think a lot of people think with robotics is like, Okay, well, let's do what the person does. But we have to take a step back and be like, your constraints are different, right? Yep. Standing robot doesn't care, you know, I'll say, we'll spray whatever you tell it to do, it'll do it. And so if you remove that constraint, maybe you approach it differently. And so you sold the video, we applied all of the material in one go, and then Stanmore versus a person applies that day over day slip, slip on, sorry, slowly building up the profile that they want to get, and then just stance a little bit at the end. So

is it so it dries like you, you keep letting it dry, so it drives very evenly, because you do like thin layer after thin layer, where each layer can be quite even. And therefore the whole thing appears even?

Yeah, it's really minimal. That minimal sanding and you need to use kind of the data breaks of the sheets gonna look like this, there's a gap between them and use the edges to apply the material, but then that shrinks. And as it dries, so then the next day, you come in with like, a little bit bigger, and you use the next data. And this is how people do it, right, because they need a reference point, they need to touch the wall. And it means to your point, they have to be thin layers, they have to wait till it dries, even though maybe they're done for the day. And they're like, Oh, I wish I could apply the next coat. Because it's right, you can't do anything. As we looked at me, like, oh, robot doesn't need that data, right? It has this really good map around it. It can spray material very consistently versus a person, right? Imagine trying to move a point one meters a second, for eight hours a day, might be a little hard. It was sitting with each other and be like, Okay, you guys do it this way. But you're also really good at adaptability, you're also getting the world's best feedback control system through vision to touch right, the amount that they gather through their touch is incredible. And then we said, well, robots don't have that. But they have really high repeatability. Right. And that's it, they don't get tired of sanding. So is it Oh, and then let's make the process a little bit differently. Let's make a process that the robot can be successful at, apply all the material ones, and then sand a lot. But you're really using the things that the robot is good at, and not depending on what people are good at.

Gotcha. So I imagined during this, there's a lot of like, picking the brain of the tapers, and your roll up your roboticists to really try to understand kind of like, what's the whole process what's actually being done, not just the way that they do it, so you're not trying to recreate what they're doing. And then from there, you try to then optimize for what's best with the robot, given the way that the robot behaves and what it's good at and what it's not.

Exactly, yeah, I think that's the right way to think about it is like what are you trying to accomplish? Not necessarily how but let's start with what are you trying to accomplish? And then we can work out the how and the actually are the people we had to reassess our superintendent. You know, she came with all this experience. So one of the first things she was like, You guys have to take care So you don't have all the engineers out there we take. I don't think we've passed her quality standard yet, but she lets us run the robot. But it was a little bit of like, yeah, I understand why you're doing it. And it was seeing those things and understanding that we're like, Okay, well, that would be difficult for a robot. What were you trying to do is create this profile? Okay. Well, there's other ways to create a profile, right? If the same constraints,

can you tell me some of the difficulties about spraying and sanding that you guys came across?

Definitely in the spring side, right, that that consistency, you are the one deliver delivers about three and a half gallons per minute. So it's a lot of material coming pretty fast. And you have to apply it fairly consistently. So that all sounds for you don't robot can do that. The problem is more. And I guess, upstream, if you will, the materials that have been developed so far have been for people. And so they're not very tightly controlled materials, right? The viscosity varies widely, which when you're assuming that you have the same viscosity, and you get the same result, that assumption, yeah, exactly the problem versus but from the manufacturer side, they've always had the person in the loop. People like their own, you know, viscosity, so they add a little bit more water a little bit less. So it's, for them, it's also not worth it to optimize to a certain thing, because people are so varied. Right? So you, that was one of our biggest challenges. We're working with a material that is not designed for robotic, right? Imagine like paints that go into painting your car, I'm sure there's very little variation when they go into that paint shop. Right? And then the robot sure does its thing, and it's going to be great. We were working with materials that weren't designed for that. So we had to think a lot about like, how do we close the loop on that? And again, hey, how do we use the operator who knows that to help us close the loop? So the the helped us a little bit with preparing the material to the right one, but also providing that feedback of it's still in? Well, I would say that's the biggest challenge with the spraying is the material itself and modeling that.

So how does it work? Now the person gives you feedback on how it is, but they do some prep work. So make it so it's like a fairly consistent viscosity, or

we'll kind of give them some guidelines, but they do have to the material comes in, it's a little too thick for spring. And so they have to thin it down with some water. But this is where we've seen right at the beginning, we were like, only at this amount of water, you know, and that's where they provided the feedback and a lot. So that we started being like, okay, let's be more descriptive, get into this type of consistency, right makes me think of the baking shows, to the peaks. So we started being more rounded, like, Okay, what we really need to do is to get it to this consistency. And then they use their own feedback system to be like, Oh, well, this batch was a little bit more liquid, I'm only gonna add half the water that I normally do our dispatch in this actually temperature dependent to when it's really cold and material. So yeah, everything that could be different about it are processing. So as a couple, everything is variable. And we need to declare a process that achieves the right result, even with all that variability.

Have you ever I'm sure you consider this, but I'd like to hear your thoughts on it anyways, did you ever think of working with a manufacturer of these materials to try to like get them to control it pretty consistently. So you could, I don't know, work with it more easily.

Like where your head is that we have, and they've been really great. I think they also see a place where introducer robotics, robotics also allows you to maybe improved on the materials, because again, your constraints have changed. So they've been great. We've been working with them for since the beginning, really. But you also have to manage that when, like, what's already acceptable on site. So, you know, we kind of did it in parallel said, like, let's work with, again, as few barriers as possible for our customers. So you know, no map, no, this. Okay, let's also develop the site because there is a future work, you know, don't have a better result. So that's happening. Yeah, they've been great. I think they also see the vision and especially their scientists get excited, right? Because, like giving them a new design space, the design space for people, is this, the design space for robotics, like, Oh, you wanted 1.3 millimeters thick? Done? We'll do that for you. And then what? Well,

that's so funny. Yeah, it is interesting. It's so it's so interesting to consider when you change the constrained space, like you were working with people, people are really good at certain things, not good at other things. And now you take a robot totally good at different things and bad at other things. And so it's like, okay, you can use the best of both. And this is very interesting. It's an interesting spot to be.

Yeah, they're just like, Oh, these ideas like constrained space of cost, because it is a very labor intensive task. The materials budget, if you will, is very small, right. It's all when the cost has gone to labor. especially as things get harder and harder, it's just so with the materials manufacturers will also have that conversation of like, hey, maybe you can have a little bit more expensive material, if it gives you more productivity. That works, right. And so they also start seeing the idea of maybe there's greener materials, you know, or materials would better, and properties, right? What if we could have better sound absorption in your drywall walls, or you get into really cool spaces once you start opening up that space? And so I think we're really excited for that future where you a robotics coming in also change some other of the upstream

assumptions. Yeah, it's so interesting. So if you can almost, I mean, like, if you take this to the limit, where you have like, infinite robots flying this and all the dry walls are, so then you have the robots already. And then it's basically free to have the robots do it other than like, the cost of energy and transporting them and this kind of thing. And then maybe, like, really nice, highly optimized materials, because and then the materials cost rises. But the total cost even in like, would be less because the robot would be affected, effectively free labor. That would be very cool. I like where that the like, new potential optimisation space? Or I don't know if space you can optimize? That seems awesome.

Oh, yeah. For us the you know, it really, we will always have that operator in the loop. So it's about their costs, because there'll be time spreading it over more square footage, right? If Yeah, I produce 1000 square feet in one day. That's something if I could do 2000 square feet. Now there's that ROOM FOR THEM materials, there's that room for that optimization? So no, I think that that gets pretty exciting for us. And it also protects their themselves where they can have a longer career while we're working with better materials and better things. So it's kind of this win win by really thinking about that augmentation, because it's about the how much can that person produce? Rather than replacing them? Not necessary? It's better to augment what they can do. Right? Yeah, not necessary,

and probably not even possible to because you need their expertise. I was thinking, it's a little bit like the grocery store model, like the clerk where you have one person Manning 10, registers, but this kind of thing, but it's also different in that the person is highly, highly skilled, which may puts it into an interesting space.

Yeah, that's a really good analogy. I think this idea of like, I can supervise more, but you know, my job does change. Right now they're much more technical it is, they're much more customer success, right? Making sure that they're making these things. It's kind of how things evolve. But if you completely tried to remove the person, I think once your body senses to that gets really hard, why not utilize them with their best, like, utilize their, you know, their adaptability, their smarts? We don't have to wear down their shoulder to get this word. Yes.

It sounds awful. For sure, for so with this, I How, how has the building climate, like economic climate? And in that, like I've had several interviews, where there's shortages and people to do different work, is there a shortage of tapers? Or is it like, does this let you stretch it taper to do what 20 tapers could do? Or like, is there a big shortage there? Also,

construction is facing an incredible shortage, you know, across all traits. vaping is definitely one of them. It's you know, one in 40% stand shoulder surgery. Oh, that's so bad. Yeah, yeah. And you know those things, right. They maybe fix it, but it's never the same. And so they often either have to retire early, or you know, work live with that pain, which

is that's because of the sanding, the like,

sanding, that repetitive motion on your rotator cuff. Right? Oh, god. Yeah, ceilings, also really hard. There's also a dust, which the sandy part? Exactly. They optimized to reduce it, but it's still there. And even if you're not standing that much use the tools that they pick up, you know, 10 to 20 pounds and their hours. Yeah, that's crazy. Somebody's just wears down on you. It's icy really? Yeah, think about that. The it's been really hard for like the union where we work very closely with to recruit new people. You can see that right. It's like, such a really good paying job. Like I said, it's amazing the skill these people have, but when you put them in front of like, maybe he'll return 50 Your shoulder will be blown out because their shoulder. Exactly. And you're definitely be hurting. You know, at the end of the day, you're gonna have a long day and you're just Yeah, so I think it's hard for them to bring in thankful dust themselves, right? They're having a really hard time bringing people in, even with the people we have. There's a lot of work right, we have to double our building stock and there's just not enough people. The shortage in construction is just outstanding. Yeah. So how do you bring new people in and I think that's giving them to so then you are excited to do that work like Oh, for sure. You know, I could protect myself and then also for those people that You do bring in getting more done.

You're letting them be, like craftsmen with underlings kind of thing. It's like the apprentice model. But the robots are the apprentice in a sense, that does some of the work. That's really cool. What do you so just Just curious on your perspective on this? So do you think it's the more skilled jobs that are harder to fill? Or the less skilled jobs in the construction industry? Like, is it? Is it the tapers or I don't know, the general construction worker? That I assume the shortage is in both? But are they kind of is it more in the tapering area or more in the general worker? area?

I don't know exactly. But my sense is, in general, everybody's having a shortage, but you see it in those harder on your body. So I think they all have like our journey men that we work with is a believe a four year process to get trained. So it's a long apprenticeship. Similar electricians have to have their apprenticeship. But the wear and tear is not equivalent across trades. So I think trade like drywall finishing, in particular suffer from that, like, well, if I'm gonna pick a trade, I might pick one that is not as hard on my body or before, they were in a different time. And people see you like our controller, and they get excited, right? They this is things that grown up with, it looks like an Xbox controller. It's a robot. So there's also that like, engagement of like, I'm working with a really cool tool. So I think between that in making their day to day just be a lot less painful. Yeah, it's the right way to bring people in.

I imagine it's like the you keep that because I imagine that they got into it, because they enjoy it somewhat like the meticulous attention required for this kind of task. I imagine like I mean, I love programming. And it's like the same kind of thing, I imagine where it's, I enjoy the details. And so I can imagine them enjoying the details. And in this case, they get to they love the inspection, they love going this needs a little more of this. And then the problem solving that probably goes along with it that I don't understand what it would be. But so you let them do that. But you let them delegate all of that repetitive work to the robot. That seems really great to me.

Like you said, Yeah, that's right that the people have on their workers you know, they love Yeah, created that your wall looks amazing, right? This those walls, they call them like walls where you have like, the skylight that shines on it is glossy, when you know, the architect really went out laughing to get something like that they look beautiful. So there's a ton of pride and a lot of our feedback with them working especially on the user interfaces, how do I give you that ability to use your crap to use that and get that engagement, right? Because they do want to keep that in theirs? You know, they want to keep that pride that's building up for years of training, you know, that you're, you're that person, not everybody can do it. Right. So even if I gave you the robot, we could easily get you trained, probably faster. But I don't

click a giant hole. And it's like, no, no, it's good. Other than that, yeah, just put that picture as the plant over there. Yeah, that sounds very funny. Okay, so what what other challenges did you have with so sit spraying the big challenge was the viscosity? Were there big challenges with sanding as well? Or was that kind of straightforward?

No 10 Things really where we started with a, like a research phase, because sanding requires contacting the surface, right? A lot of the, you know, robots started a lot in welding. That was nice, because you have that little bit like, I don't have to be perfect as long as I can get the arc across it good enough, right. And then of course, you say, well control everything around it, and then we're gonna get there. But the moment you needed to stand, you needed to have that contract. And it's not just, you know, the, I guess, this sense of the wall, but like, your angle of approach, and that soft materials I was mentioning, so that was a lot of our work was like, how do you make this thing happen without a perfect map? Right? We do map our own space, but you're gonna get there's still

accuracy, right? Yeah. So that's three centimeters. Okay, that's pretty big. For a precise job.

Yeah, exactly. And so part of that is just your project is

probably not more than not much more than three centimeters.

Yeah, it's about Yeah, I'm gonna go two inches, but like half inch drywall is very common. I mean, we put up

yeah, that's three centimeters or something. 2.4 or whatever it is. Or so you.

So you might not be right, you know, know exactly, plus or minus. So really about thinking about how do you approach that Walbert? I have a good guess where it is. I have a good guess of the angle that it is. But I want to make that really soft contact, right because I can scratch it. I'm going to create more damage. So this is where other research came back to compliance. We really thought that one of the ways to get robots out into the world is make them compliant, right? They don't have To be, have a perfect map as long as you can tolerate that. And by doing compliance, so we build a ton of compliance into our sending end effectors such that you do your best to get the best map, but then you say, at the end of the day, I'm going to have to, you know, start this, the reality to like, the joints will be like hump or zip. And you don't want to like fault or cut through that hump, you actually want to write it out, that's what people do is that reflect is just the illusion of flat. So you also need that your Sandy had to kind of write that hump and spread it out. So it's, that was a lot of work. And, you know, we have like, four years of research with the government before even doing this that leverage.

So what do you mean compliance in this? Is it like, springs? Or do you mean like, what what do you mean by compliance? I guess?

Yeah, in general, we're looking at it, achieving it whichever way you want, right? Because spring is a perfect example where it allows you to right absorbed that positioning. Exactly. So we look, you know, their spring scampers, but also under your control side, how do you detect that compliance and react to it? So you don't have to have all of your compliance be mechanical, some of it can be coming in from your control system?

Gotcha. So do you do a mix, where you might have some, you might have something so you have the sanding head, and then you might have something where it can kind of move a little bit given, like with a spring? I don't know, keeping it kind of facing forward. But like giving up giving a little bit if the wall is a slightly different shape than you expect? And then you do it in hardware? Or in software to where, hmm, and that's, you kind of just do emulate additional springs in the, in the motors or what do you? What kind of approach is that?

Yeah, there's a few ways. But one could just be, you know, enabling, or sensing that mechanical displacement, if you will, that you're having and then utilizing your software to give you a larger range, right? So you're coming in like this, and you're already detecting but you're going to hit your heart stop, you go ahead and like tilt your arm, which some of it is more like a feedback. But there's other ways, right? How you run your control system around it, allowing me to, you know, be tolerant, right? My ultimate goal is not perfect position. My ultimate goal is perfect force, right? constant force it elsewhere control system.

Your goal is constant force is that is that that's really useful for sanding. Interesting,

okay, that this goes back to that hump where you want to write those out and like, apply the right force to level it out. But I don't necessarily want to grind down. Yeah. Because that's even going to get through the other side of drywall. One of the big challenge rates is it's not just the drywall, but it's how it was formed. So if somebody didn't hanger very well, you have that? And then how it was framed? Yeah. So all sudden, you start, like, there's a lot of personal for sure. Yeah, they're the ones who fix everybody's mistakes. Everybody, it's true, right? It's like accumulation of the error that then you have to then they refined back. So you need that. That approach of, you know, let's move that out, not necessarily flattening it out.

So if you have a hump in your drywall, so it's hanging, and there's a hump, and you're trying to apply a constant force throughout it. Now what I'm thinking, if you had a point sander, like, then it would just make everything deeper, like shorter into the material by the same amount, but because you have a, I don't know, a sander that has an area, and you move it around, it kind of averages everything out. And that's why constant pressure is important. Is it correct or?

Yep, that's correct. Yeah. And then you also have to look at, you know, having your compliance also in your tip tilt, so that you can can also write out that thing. Good. both axes, right. You want to be able to

adjust different ways. Yeah, I was. My explanation was 2d, but yeah. Yeah, I imagine the same thing holds true. 2d and 3d for that. Okay, and how do you get your initial map? For the robot?

Yeah, we use a lidar system to create our map, you know, and then from the point cloud, because this also has to be fairly fast, right? Anytime that you're not running the tool, it's not productive time. So from the point cloud, find the planes and then work off of those, subdivide them into spaces. So it's a little bit it's, it's a different application, right? So much work around LiDAR, some point clouds, but it's what are you trying to do with it? For us? It's finding surfaces, right, as opposed to maybe identifying objects like an autonomous car has to do.

So do you? Is it like a person carries around something or do you drive your robot by it or how does it work?

We drive the robot by it. So the operator they have a little like, do a little loop and they get some feedback from the user interface if we're getting a good little good map, and then they also get to do once it pops up, they can use again their smarts to say like, I'm missing a wall or something like that.

Gotcha. Was that part pretty straightforward? Or was that there? Was there challenges there too? Yeah,

I would say, definitely feels more straightforward because it is mostly a technical challenges. And you know, sometimes I get excited when we find just technical.

Because it'll be one thing that's relatively easy and fairly established to Yeah.

Don't get me wrong, right. It took a lot of work and optimizing and getting it right. And we're all still tuning and stuff. But it is something that like, it's well within our control. And probably one of the, it's a little bit of a challenge. But also exciting is the technology advancements on the lighter side has been incredible for starting. So a lot of it has actually been like, Okay, should we upgrade to the next LIDAR? And that means retuning our mapping algorithm and so play that trade off. Yep. But yeah, I like those challenges that are mostly technical, not that the other ones are interesting. But definitely, when you're trying to change people's perceptions, or construction business alignments, you're just like, well, that's a lot harder.

Oh, totally, I think. To me, it seems like successful companies are founded on just a few very hard challenges like one or like two or something like this, this is your sanding and spraying. But if it was like every single thing is an incredible, like moonshot, then it probably wouldn't be a good long term bet for the company, because one of those moonshots will be a giant bottleneck, or those kinds of thing. So it is nice when you have just an engineering problem.

definitely agree that we think about it, it's like, let's make sure you're successful with like, just just two or three challenges they're picking up. Yep. And then we talked about material development, that's great, but don't depend on that. It's a longer timeline. But so we work on some of this research on the side. And as those kind of get proven out, we end up but you cannot keep this team running on more researchy stuff.

And it's almost like a parallel thread to the main value add for the company. So it's like, we're going to be around in 10 years and like, then maybe that'll kick in. And we can have these conversations now. But they're not like, you're not waiting to be a profitable company before. For that to work. It's like, we got it we can we can just make do we have to mix it to approximately the same consistency. And it's all fine.

Exactly. Yeah, just sure you provide value at that point to the customer, right? Because they're like, that's great that you're gonna do that in seven years. But I have a project tomorrow, what can you do?

And venture capitalists are like, Oh, seven years, I think you'll start to be it's like, oh, it's a robotics company with all got all the infrastructure and funding that that requires? Yeah,

exactly. It is definitely yeah, that's where you I think you really want to show this inability to get out on site sooner. And so having those things and I act, do you know, where we began, where it's like, keep the person allowed them to do that helped you get to that place, right? And that's why it's both, I think, the right way to do it for Mark values, but also the right way to do it. When you look at the problem, you're like, Okay, let's do that. Let's get the person in there and help us get out there.

And so I'm, I'm torn between two segues for this the first one. So how are you? So we talked, I mentioned just a little bit with like venture capital in this kind of thing. How are you guys funded?

Yet, we're a venture backed, we raised our series B, about a year, a little bit over a year ago. So it's been great. We previous to that, like I mentioned a little bit was we were doing research for the government. So we did have some NRA dollars. So it was a great way to develop the A get back to your point, to many problems. So research was a great place. So we worked with NASA and Office of Naval Research, they were really looking at this, you know, compliance in the spaces. You know, they're the most unstructured of spaces. So then we we sold some of them. Yeah, in construction, much better. And yeah, once we really found that application, that was another decision for us is like the, you know, the Swiss Army knife of robotics. That's pretty hard to do, right? Trying to solve all those problems. So we knew we couldn't just be the construction robot that did everything we've said, let's find one trade, one task that we do really well. Yeah. And then from there, we can expand. And so once we found the task, and we had demonstrated that there was value with the current technical challenges as opposed to future, that's when we went and fundraise around and did a follow up round in 2020. Gotcha.

That's exciting. Oh, yeah. Can you talk a little bit about kind of bringing on investors and the process of that because getting really savvy investors can be a huge benefit, but getting ones that have like incredibly draconian clauses and flood their funding agreements can I've seen it kill several companies? So I Assuming you guys are doing fantastic, like, it sounds like things are really good. But would you tell me a little bit about kind of picking the right investors?

Yeah, I think a big challenge is when your hardware, you're already probably looking at a different space of investors, right? At least here in the Valley. It's like, there's so so many investing firms, but you should research what they invest in, what are the pieces, right? If they've never done a hardware company, let alone a robotics company. They might not be the, the right fit. And early on, we talked to everybody. And then very quickly, you realize, Oh, you guys are used to software, and all the software? And maybe that's, you know, you guys are great. But do we really see like, are you set up in a way where your fund can support something like where you wanted them to have secondaries because we know this is a longer process that you kind of like they can come in and reinvest in the future around so like, have you researched, you know, they're putting x amount into your first round? Do the reserve some amount? They don't have to use it. But do they reserve some amounts so that when you're raising your next one, they can reinvest it? It's just a nice way to support that it I think, it just shows that they thought through like, look, this is not going to be one round hit, right, we are going to have to develop it just the nature of Harvard, it's a huge, awesome win rate, the defensibility is much higher, but it is a longer path. So a lot for us was really a search problem, right? Find those people who do, who understood us. And then to your point of like kind of that relationship, who supported one that company we wanted to build, but also brought in maybe the areas that we weren't so good at right and provided that guidance and really served as advisors at the center.

Now, so you mentioned you have like 40 jobs or so done so far. So I assume this means you don't have that maybe you probably have less than 40 robots with this, which makes me think less than, like, it's not a manufacturing problem, really for you guys, because it's kind of so this is interesting to me, because it changes possibly the type of investor that you would want to get. Would you tell me a bit about this?

Yeah, that's definitely a big challenge, right? It's We're just two things, the way that we're looking at it is breeding a lot of value with one machine. So we're not going to be a consumer electronics, Molly, right, where it's like literal value, but you still didn't really know that you're doing great, right? It's, it's a much different approach is much smaller volumes, but create a lot of value for them. So that already changes your perspective on how you're going to manufacture. Right, you're not talking 1000s of units to your point, right? We're talking 10s of units now and then scaling up from there. It's your design space changes from when building it right, there's maybe manufacturing methods that you don't have as high X as you might, you know, you can't injection mold every piece, you have to come up with another way because you know, every move is so expensive. They're like how many un 20,000,020 20

So it just makes it a little bit but it so it just comes again to pay you create enough value and then your product of the lifetime to support it, right? Because it can be an expensive product. But if it runs over the lifetime, and brings in the revenue, you can actually have very good payback periods. And that's been a metric for us, like, Hey, make sure this robot pays itself back in within a year. And then you have a five, five year lifetime. Yeah, it's a good business model, right, as opposed to maybe it could be something cheap. But if it only brings so little value, so I think you're really like have to look at payback periods, to see if your hardware has the right thing. And then work with your design team to say, Alright, how do you get a good cost? What with without access to maybe the cheapest manufacturing methods that are out there?

Definitely, yeah, it's a totally different problem for this. And you can also kind of embrace that it's not going to be like the one off thing and try to get it so you can make 10 of them at like, not very well or not, like not as optimized as possible to be the lowest costs. But then you're focusing on kind of the longevity of it, and then it kind of pays itself back so you can invest a little more. How do you so I'm just wondering, and maybe I'm a bit naive with this how do you with robotics being fairly new and things coming out all the time? Like you're getting a robotic arm maybe it's fairly well established? Maybe it's not and then some of the light are how do you predict how long something will last is there? Like like this kind of thing is an interesting problem to me because I don't know that every part would have a track record for many years.

Now you're 100% corrected. Not every part has said some parts do or mobile basis actually we retrofit a construction based on these things actually fought for like, I feel like you find everybody years later be like, maybe change the batteries. So it's kind of going through your components, and then also build your model to be realistic, those things that you're not super sure about, or maybe sensors that maybe they do have the lifetime. But do you actually want to keep it for five years and be using that old or the company could

die, and then it's like, you have to support and they had a closed, like a closed, proprietary sensor driver that you can access for this kind of thing, no longer works on a bun to 2026, or whatever it will be in a couple of years.

That's right. It's a what we found is like, I build that into the model. So is that either we're not so sure that the lifetime will be the five years that we target? Okay, just built in, that there's going to be replacement parts, or the ones that we're like, we're not sure we're gonna want to use this either company's not stable, or, you know, there's a much better sensor, why would I still be using, you know, five year old LIDAR technology when I can get this one? That you build that into your models? Like, hey, there's gonna be so many 1000 flowers, so replacement parts that I do over the lifetime? Do I still get a good payback period? Right? Does it still pay itself off if I make those changes, but of course, you also don't want to be changing them every minute, just because it's a better sensor, you gotta make that ROI analysis for, for component, but do your best and then keep really good track, right, as we're out there. We've been running now for almost four years. So we know that those machines are still going. So that's good. But you also have to keep track, what things take your replays? Was there any leading indicators that could have told you that you needed to replace that before? A true failure? So you know, predictive maintenance is like this awesome future that, you know, manufacturing and logistics already has so many cycles, but we're making sure to, like collect the data. Now, even if you're not sure what you're looking for later, if you have a failure, you can go back increasing? A little bit more? Yeah, what was my leading indicator that would have told me to replace it?

That's so interesting, I've never really thought about that. I really, I like that perspective, you kind of gather the data and then after, and you just, like, preemptively are gathering it, because perhaps you'll be able to see trends in the future. And it's probably fairly cheap to grab now. That's really cool. Yeah, and I guess manufacturing and logistics probably has that down. But that's very cool for you to apply to your systems.

Yeah, I think some years ago, maybe the data would have gone expensive. But now it's so cheap that sometimes when we do we need all this data, that it's worth it to keep it and then look back. And there, there has been some things where we've been able to kind of look back and go, Oh, okay, we saw that here. We saw that there. Right? This is starting to be at least something to look out for, right. And we put maintenance in the week to service our customers right after every job that comes in and gets inspected. And so we can at least check for those things, you know, and identify those areas that okay, maybe built into the model, that being a batteries will be your place. And let's go ahead and adjust for that in the cost.

Gotcha. Now, this question, I don't know if it's too early. But I wonder what you're thinking around these lines. So in my understanding, when you take venture capital money, they're typically expecting an exit I hear like, the timeline they say is 10 years, but it's often a lot closer to five. I'm wondering, and so there's several different types of exit events. You can be bought by someone you can IPO. Or you can wait, I don't know if there's other things too. Are you where do you think do you think you'll have what kind of liquidation event Do you think you guys will have at some point? And what are thoughts around that? I suppose.

Yeah, I plan around and we've always thought is like, built a successful company. So build with the goal of IPO or something like that, in your mind. If you're doing that successfully, the other liquidation events? Yeah, exactly. It's your think when you build too, like acquisition, you get into a really tricky spot, right? Where like, you're now trying to guess what they want, or you build so much into them. And then they're like, actually, we decided we don't like B vs. sustainable company, right? Build a company. That's, you know, at some point, you stop taking venture capital, you can produce your own things, right? We want to expand drywall finishing is our starting point, we want to expand and really be the tool that you just know that Oh, but structure you use Canvas, great tool that you use to make everything. And so make sure that your models and your plants are built to a point where you do have that feature where you can support that r&d And the new product lines off of what you're

growing and keep building. I see. Yeah. And then that

way, it's under your control. You're building what you want to build. And it should mean that all your liquidation takes care of itself possible, right? And

you're basically aiming at the one that gives you the most option, in a sense by being such a compelling business as opposed to overfitting to one customer buying you or I don't know, or I've seen people be like companies be bought when they like Miss A funding round or something like this, this kind of thing. But if you stay competitive and keep, keep growing what you can do and stay profitable? I mean, that sounds fantastic. And then you'll be in a wonderful situation to like, get the best investor and all this other stuff. That sounds wonderful. Oh, yeah, shoot for that. For sure. It's just an interesting thing, because it's a trade off in a sense with venture capital, as opposed to so you get to grow quicker, but you are on their timeline to some level, because you they are expecting some sort of event where they'll be able to get their money out for more. And so just working with that.

That's right. And that continued progress towards that event. Right. So like, increasing validation with ROS No, it's It's definitely, you know, a decision that you have to make and look at and for sure, yeah, like you said, speed, they definitely have the what you need. If you want to go fast, you just gotta make sure that you can deliver and, and you look each other in the eye, and you're very honest, and like, this is what we'll deliver. We're looking, you know, we talked about building that successful company, what's your timeframe? What's the timeline? Do we all agree that that's a reasonable timeline, right? Have they been like, in a year you have?

A five years, but we meant to? We want a unicorn immediately. Yeah.

And that was actually that's a perfect one for us with the expansion, right. We knew that having been in robotics for a while, like, staying in the Bay Area was the best way for us to move really fast and iterate, right? And we knew if we started shipping these across the country or the world, we're spreading ourselves thin. And now you're dealing with operational problems, as well as technical problems. So it's like, iterate here, the market is big enough here. And so we lay that in front of our investors from the beginning, like, Look, we're not going to be that grow everywhere, quickly. Nope. Because that one area Exactly. Nailed it. And then almost half the customers be like, Why don't you come over here?

I'd be wonderful. Yeah, that funded growth would be fantastic. When they're like, We have a big project over here. We need you over there. It's like, I don't know, it's logistically challenging for us.

And then we can support it right, then your support team is not eight engineers that have to your right your time. Yeah, exactly.

I'm always surprised. companies doing that, when they're like, we're here and there. And they're there. It's like you, there's 10 of you. It's like how, but

that has been the approach. I think that a lot of firms and venture capital firms have pushed and for some models, I think it makes sense, right? The first set the market is more important than anything. But once you when do you have hardware? It'd be that easy. Yeah. And it's

logistically a lot of challenge to coordinate. I mean, you have to like so even if you're not going to fly one of your skilled engineers there when the robot is down for some reason. It's still like, Okay, now you need to train a few very highly skilled technicians in the area, and coordinate with them and give them updates and all sorts of things. So that sounds hard to me. It sounds like a smart approach to stay just in the Bay Area. That's a hell yeah. Let's see. So now the other really interesting thing that I was thinking about with the Segway earlier, you are working directly with unions with your robot, and then just tell me a bit about that.

Yeah, when we first started writing, we were learning about construction and found drywall finishing. It was actually interesting, because we weren't thinking as roboticists, I would hear drywall screenful We were thinking of being in the sheets. And it was talking to construction. People were like, oh, yeah, that's hard. But let me tell you. We don't think we had even thought about that. And no, it's like this perfect. That was it. As we looked at a right one was look at your market, right? And we want to we just talked about it, we want it to be in the Bay Area, state type. commercial construction is what makes sense for robotics, right? Unless you have a pretty big house, maybe the logistics of moving the robot might not make sense instead of the square footage your house has to do. So we knew that there was like a minimum project size that it makes sense to maneuver a robot. And so that was like, okay, commercial area. It's a Union Market. Right. And I think that could maybe upstairs a lot. People will be like, Alright,

1 post with the drywall tag