In The Arena by TechArena - Achieving Sustainable Building Construction with Urban Machine

Episode Date: February 15, 2023

TechArena host Allyson Klein chats with Urban Machine CTO Andrew Gillies about how his company is achieving a breakthrough in sustainable construction through the AI and Robotics powered Machine.

All right, let's get started. Welcome to the Tech Arena. My name is Allison Klein, and today we are delving into a topic that is near and dear to my heart, sustainability, and in particular, sustainability of trees. I'm very glad to be joined by Andrew Gillies, co-founder and CTO of Urban Machine. Andrew, welcome to the program. Thanks so much for having me. Excited to do this. Andrew, why don't we just get started with an introduction of you and your role at Urban Machine and how you came to become the CTO and co-founder of Urban Machine. Yeah, sure. So my background sort of bounced around through startup land in Silicon Valley here, PhD out of a robotics lab at Berkeley. And before starting Urban Machine, I was doing sort of freelance contracting work. And through some of that work, I met my co-founder and the CEO of Urban Machine,

Eric Law. Eric is really the construction industry veteran and wood products guru. He had been sort of tinkering on this problem at his last role. He was working on innovation and sustainability at Swinerton Builders. And he recognized through that initiative that there's this huge waste stream of lumber in the U.S. We're effectively landfilling every single year. So the EPA studies, the most recent numbers on that is around 37 million tons of lumber. We're shredding and we don't really have a good use for that shredded material right now. It mostly gets spread on landfills to keep birds off of garbage or suppress smell. Some of it gets used as, you know, mulch for agriculture or some amount might get burned for fuel in specific cases.

But there's really not a good answer out there right now what to do with this huge volume of material. And the construction waste in general is a huge part of our overall waste stream and somewhere around 25% of our overall waste stream is just construction and demolition waste. And if you go and speak to the operators of the recycling centers and the landfills like we did when we were starting this, they're just overwhelmed by the pressure of all this material. There's a lot of changes in the construction industry, a lot of these urban areas now. And they actually call it woodageddon because they're getting so much of this material, they don't know what to do with it, they have no answer. Because once you shred that material, it doesn't have a lot of value. So when Eric and I were looking at this problem,

he identified that, well, when it first comes out of the building, it's not shredded, it's still in its beam form. The main reason in that state that it's not valuable is because it's full of metal fasteners. It's full of nails, screws, staples. And it's not very sellable when it's covered in bent nails and screws and staples. No one in a lumber mill is going to want to take the risk on their saw blade of passing that material through their line. If you hit a nail with one of those saw blades, your whole line is down. It's just not worth the risk.

So the mission we're on at Urban Machine is to intersect this material while it's still in its full board form and figure out a way at scale to remove the metal fasteners. That's really the crux of what we're doing is getting the metal fasteners out of the lumber. Once you can get the metal fasteners out of there, as long as you don't shred it, it's still very high value material. And we've been working with early customers and early partners to validate that.

We can sell it either at or substantially above the commodity price of lumber. I came at this from an automation perspective saying like, okay, well, we're talking about a repetitive task, scaling a repetitive task. You know, it's a perfect candidate for automation for robotics, especially where robotics are today. So that was really my role is working with Eric. He had the concept and then trying to connect the dots of like, okay, so we, we have a system change here. Let's provide a technology stack that can implement that system change. Yeah. So I don't talk about this on the tech arena, but one of my passions in life when I'm not talking about technology is actually remodeling homes and trying to use architectural

salvage as much as possible in that remodeling. And I know how much of a burden it is to remove nails physically or manually from wood, even if you are a well-intended sustainable person. It is a tremendous amount of work to do that. So when I saw that you've invented a solution that combines artificial intelligence and robotics to bring that same capability to market, my antenna raised and was very excited because you've combined my two areas of passion. Tell me about the solution that the team has invented and how you've tapped technology, both from the world of robotics and the world of artificial intelligence, to bring to bear to this labor-intensive challenge? Yeah. So at the heart of this problem, really you can think about it as sort of a reverse Amazon pick-and-place system. Amazon has this huge challenge of trying to move objects from A to B,

right? Or like place an object into a box. We're trying to pull nails out of wood. So it's, the fundamental layer of this is the same two challenges. There's a perception challenge, which is, you know, given a scene, where are the objects that I care about, what's their pose in 3D space, their position in orientation, and then what's the classification of that object. And then the other challenge is it's a manipulation challenge, which is okay, now I know where the object is that I care about. How do I go in there? How do I grab it?

How do I manipulate it? How do I move it from A to B? So those are sort of the two main technical challenges we're working on. On the perception challenge, that's where a lot of the magic has happened for us. That's where the AI lies, which is like looking at these complex scenes, looking at the wide array of fasteners we see there. And it's not good enough to just pick them out in a 2D image. We need to know the 3D position and orientation of these objects. So that's where we're applying AI right now. It's like leveraging the breakthroughs that we've seen in computer vision over the last several years. Yeah, so we've developed a special pipeline and a data set. We joke that we've now generated like the largest data set of

reclaimed lumber in the world. So the amount of lumber we've put through there and the number of wild fasteners that we've seen. Yeah, and the AI is the unlock there is instead of going in and and and doing you know painstaking feature engineering across this like long tail of fasteners we we have a model that we've developed and then a pipeline after that to post-process the results of that model um so that we can output uh pick points that we that hand off to our manipulation stack uh so that that's really the heart of the AI. And then when you look at it from a standpoint of the robotics, you're passing on the information of here are all the fasteners in this piece of lumber. Now go at it. What does

that look like? Yeah. So the go at it part, that's more my area. So I'm much more on the hardware side. So the the end effector, the part of the robot that's going to actually try to grip the fastener. And then there's how to get that end effector to the pose, right? So that's the robot arm part of it. Our solution, we decided to go with the robot arm that's a gantry format. So it's a three axis with a rotation on the end. It's sort of a glorified 3D printer in a sense, a very fast, powerful 3D printer. We did some initial experiments. The loads that you need to pull a nail out of wood can be up to several hundred pounds.

So there's not a lot of off the shelf gantries that can work on the cycle times and apply the loads that we need. A lot of off the shelf robot arms that you might think of like a universal robot, a cobot, or even a light or medium duty industrial robot arm like a KUKA. Those might have a payload of around a few kilograms up to like maybe 50 60 kilograms for like a reasonable light to medium duty robot arm but we need several hundred kilograms so that kind of ruled out those off-the-shelf solutions uh so we've actually developed our gantries in-house so that they can be not only super fast but also have these like

tremendous load capabilities um so that's what the gantry that's what the robot arm looks like and for the end effectors we took a lot of inspiration for other hand tools that you might use to remove nails there's also some fun tools that you can find in the hardware store to help solve this problem for people they're all frustrating in their own ways and they're all awkward in their own ways so we kind of did a design study across what was out there and arrived through iteration at a solution that we think is a really healthy compromise across all those tools. We call it Big Bird because it more or less looks like a sharp little beak.

And we can pull out, with that same beak, we can pull out all manner of nails, staples. We can actually pull out a big 16-penny, like a framing nail. That nail could be hammered all the way through two by four, folded on the far side. We can actually get under the nail head into the wood, penetrate the wood a little bit, get under the nail head, grab the shaft, and just pull that nail up and bend the nail straight to pull it out of the hole. It's quite the scene. Amazing. Yeah. So this is fantastic. When you look at actually entering wood into the machine, what is the time frame that it takes to scan that piece of wood, remove all of the nails, and exit the lumber out of the machine?

Yeah, so there's a lot of moving pieces of this. We have targets on the business model side, so we know if we want to hit a certain revenue, a certain rate with the robot, we want to hit a certain revenue a certain certain rate with the robot we need to hit a certain throughput and we actually created a full 3d model of our system that we could input certain characteristics and parameters of the wood so that we could estimate we could break down into how quick each of these actions needs to happen so we could say like you know if we have a pick time of X and if we have a movement time of Y and we have this number of fasteners and a piece of wood and what is this dimension? Throw that all at our 3D simulation and we get a result.

This is how much volume of lumber we get a day. We know how much we can sell the volume of lumber. And so all that backs into these sort of each of these modules, we get a cycle time for them. Our target, our initial target was 15 seconds per nail removal. That was sort of like the initial. And to date, currently with our third generation prototype, we've got that down to about 4.8 seconds. That's amazing. We're like well under initial targets. Yeah.

That's really incredible. We're getting better. Each generation has got better and better and better. Yeah. And we can move lumber through the system. It works out to be about 10 meters a minute. We don't do it continuously. It's sort of a semi-continuous process. Yeah.

But we can effectively cycle the wood with our conveyance system. We have a cycle time for, I think it's like every 0.25 meters we can move. We can move that in about three seconds, a little under three seconds. So I know that Urban Machine has been getting a lot of attention since you've been public with your product. What customers are you going after with this? Is it builders? Is it places where lumber is disposed of today in landfills is who are you working with and what has there been has excuse me what has their response been when you presented the machine yeah the response has been really strong a lot of folks you know the story you told the beginning of working on the home having leftover lumber and the pain of what to do with that.

We hear that all the time from an individual level to a company level. So they immediately recognize the problem. They immediately want to be involved. It's been a really strong response. Partnerships with the company, you can almost think about it like a two-sided marketplace, what we're doing. We have the input side, and on that side we have the demolition partner. So they're feeding us the material. And we have a few key partnerships that we started starting in the Bay Area.

We have a few key partners, and they've been really excited to work with us because they're also feeling the pressure of what to do with this lumber. Right now they're paying to dispose of this lumber. There's tipping fees at landfills, and then there's the trucking fees of hauling it off. And then there's regulation coming in that's saying that they have to find a better use than it getting shredded and ending up at a landfill. So they see the change coming, and they want to find other avenues of what to do with this waste stream. So that's on the supply side.

On the demand side, we've seen a huge spectrum of folks interested in it. As we're getting going, as we're small before we scale, we have just like little sort of sips of material that we've been putting through the first couple of prototypes. And we've been working with partners that don't need huge volumes of material.

So we see folks in either their boutique architectural applications or they're creating furniture out of the lumber or floorboard, sidings, things like that. But then on the other side of the scale, we've also had interest from developers, sort of at the top of the pyramid, where they're coming into a site, and they might have a client that has their own sustainability initiative, and they're going to redevelop some site. They're going to turn it into a large,

let's say something like a large tech campus. There's a bunch of old buildings there that have a surprising amount of lumber in them. So they're actually specifying from the developer level, hey, we want this site to be almost fully circular. We're going to involve an urban machine from the earliest stages. They're going to clean the lumber coming out of the old buildings that we're tearing down. And we're going to directly use that material on the same site in the new structures. Oh, that's wonderful. So that, which is, which is pretty wild. So, so we come in, we, we clean the lumber as the demolition crew is taking it down.

And then that lumber just sort of hangs out on site or nearby storage until it's ready to be built into the new structure. It, it never, it never leaves the initial site, which is pretty wild. It's really the most sustainable way you could go. And you get this great story in your new structure that you can say like, hey, this is actually built. There's a story here. A lot of the cities involved, they want to maintain some sort of connection to the history of the place. And we can provide that as well. That's fantastic fantastic so if a company is going after lead certification or whatever for their building this gives them a direct path um for reuse of materials for the exact same building that they're remodeling or or you know uh

retrofitting that's fantastic um When you look at broader sustainability in construction, wood is obviously a huge component of that story. But, you know, I think about the fasteners that you're taking out of the wood, other applications. Does Urban Machine have broader ambitions to be part of a broader response to sustainability and if so what do you see are the opportunities um to apply technology in this field yeah there's tremendous opportunity there's a there's a huge sea change coming here one of the exciting things about working on this problem and and starting urban machine is that we just feel like we have a lot of wind in our sails. There's a lot of change.

So if you think about it, we're really starting in this very narrow wedge, which is dimensional lumber, get the metal out, right? We see a lot of opportunities in adding technology sort of upstream and downstream from where we are today, demolition is going to change to more of a deconstruction style, right? So instead of just coming in and smashing and... Yeah, I think he froze. So we'll just see if he comes back. Okay, so you froze at demolition.

So if you just want to pick up from that, and the question was, we get the metal fasteners out of dimensional lumber? That's really a starting point for us. We see the opportunity for technology to impact this industry upstream and downstream from us. On the demolition, on the upstream side, there's really a lot of pressure to change from a demolition style to more of a deconstruction style. So right now, the status quo is mostly going in and flattening the building as quickly as possible. That's not going to be the way it is in the future. Regulations are coming through. Waste pressure, like I mentioned earlier, is a huge problem. And deconstruction is just a different beast.

Currently, it requires a lot of human labor, but we see a lot of opportunities to jump into that process, work with their demolition partners as well, and provide technology upstream as part of the demolition process directly to help deconstruction move faster, move safer, and require less human labor because the industry as a whole is just severely under-resourced on the labor side.

So that's a bit on the upstream. We see opportunities there. And then on the downstream side, we are capturing so much data on this wood. We have it registered in our machines. So doing value-add processes to the wood while it's still under our control is sort of a no-brainer. We can start looking at post-fabrication, even forming structural elements,

which taps into the other huge movement that's happening in the U.S. right now, which is this movement to mass timber, where you're seeing steel and concrete skyscrapers transitioning to full timber buildings. I think the record in the U.S. now is 18 stories of a full wood structure. These are incredible, beautiful buildings, far lower carbon impact than steel and concrete. And there's been a trend in Europe for much longer. I think it's about 25 years.

Europe is way ahead on this trend. Finally, the U.S. is starting to catch up. We're starting to see more. They're called cross-laminated timber production facilities where they make these huge timber structural elements. We're seeing more and more of these in the US right now. So we also see opportunities there where automation is going to be a huge lever that's going to unlock that even further. That's so exciting. I can't wait to hear more. I am going to be following your story to see how Urban Machine scales and introduces new capabilities in a very exciting place.

One final question for you, Andrew. If folks want to connect with Urban Machine and learn more about the technology, talk about a partnership, where would you send them to connect with the team and learn about the machine? Yeah, so the most direct place to reach out would be on our website, which is urbanmachine.build. So you can go there. And then we also have links to various social media outlets. We're very active on LinkedIn as well.

So we have an Urban Machine page on LinkedIn. You can reach out there. Yeah, yeah, and we'd love to talk about partnerships. And we're also hiring. So we're hiring across the development side, operations, and on the technical side as well. So if you're interested in working on cool robots with a sustainability mission, please reach out.

Thank you so much for being on the program today. It was really interesting to learn more. Thanks, Allison. All right. Did that go well for you? Did you think we covered all the topics you wanted to talk about? Yeah. Yeah. That was great. Okay. Jory, any feedback? You're on mute. All good? Okay, so next steps on this is, I'm just going to stop recording, hold on for a second. Next steps is I will edit, wrong button, sorry about that.