How Protolabs turns CAD files into parts in under 24 hours

0:09Gene Demaitre: Welcome, everybody, to Episode 257 of the Robot Report podcast. I'm your host this week, Mike Oitzman, Senior Editor for the Robot Report and Automated Warehouse. Thanks, as always, for tuning in. and today I'm joined by Gene Dimitri, Editorial Director, and also Brianna Wessling, our Associate Editor. Gene, Brianna, welcome to the show. Thanks for having us. Yeah, I'm excited to be back. Yeah, Steve's on vacation this week enjoying some family time before school starts again for his kids, so he's off getting a little relaxation. Maybe he'll listen to this show on his drive home. Who knows? He needs to escape, but it's fun for the three of us to get together on the show.

0:52Gene Demaitre: that doesn't happen all the time. So, yeah, let's get caught up. And, you know, I want to start with, you know, the work that we just published today, Gene, the big announcement with the Humanoid Report. We've been working on that for a couple months now, it seems like, between all the content behind it and the layout and everything else. So, Gene, this was your baby. You pulled this all together for us, and I know it was a ton of work, but always great to publish a document like this, right? Yeah. Most people know us for the podcast and our articles on the Robot Report and Automated Warehouse, but we do have downloads, which are free.

1:30And the one that we just posted is around the state-of-the-art for humanoid robots and mobile manipulators. And I want to make sure that we mentioned the mobile manipulation part because we curate an interview that we had with Jan Ziska, right, pick, who we all know and we want to make sure that we're talking about the technology that's already available right now and it's starting to roll out into warehouses and factories as well as looking ahead and what i enjoyed as i helped put together this report was really revisiting conversations that we had around what are the technical challenges what are the use cases what are the integration opportunities around humanoid robots.

2:21There's so much attention around the money going into it and around, oh, they're going to either replace us all as workers or they can't even walk and chew gum at the same time. There's such a range of opinions and hype around this that it was really nice to just focus on a couple of really clear stories that go through issues like cybersecurity, clarity issues like um you know where are the robots actually being developed and i think that was uh hopefully we're providing a service for all of you out there who are listening

2:54Gene Demaitre: and brianna you had contributed to us about your contribution to the report yeah well it's a fun project to work on because we do we cover the day-to-day of humanoid robots feels like we write about them every single day lately but it's nice to take a step back and really get that overview of what the whole landscape's looking like, where everybody falls, where the real challenges are, and the opportunities. So it's really fun to just get that overview and to step out of the minutiae of it all. Yeah, and it was a fun to compile the main table here of that list that all the humanoid manufacturers that I think are relevant right now.

3:32Gene Demaitre: And, you know, I've been compiling this list religiously for the last two years. Gene, I'm an Enneagram type five, which for those that know that I like data, I like to collect data, I sort of get lost in the data that served me well as a product manager for 25 years. So anytime I discover a new company, I have to drop what I'm doing, go over and throw it in my database, make sure it's listed there. So I hope that this is one of the definitive sources of companies, short of those that we just can't know about. Because, for example, some startup Chinese companies who haven't popped their head up yet, we don't cover Chinese language news until it goes a little bit more global.

4:19Gene Demaitre: So certainly there are some companies flying under my radar yet, just because we haven't seen them. But as soon as I see any inkling of something humanoid or mobile manipulator related, it goes into this table. And so that's sort of the front centerpiece of this, Gene. Thanks for pulling this all together, formatting it. It came together. It's great. Well, Mike, I thank you for the work that you did in the data collection here. I think it is worth noting that these are over 200 verified humanoid robotics developers and providers. And we know roughly what stage they're at, where they're located and that does not necessarily count all the people who are developing just software for human robots or all the people who are developing just one component a knee or an ankle or a hand for uh humanoid or semi-humanoid robots so there's a lot going into what's out there there's probably two or three times as many companies that are in that some ecosystem if you will uh but you know we're trying to follow all of them and uh you know there are events now popping up increasingly around the world focusing on humanoids uh we're going to have a humanoid track in our own show but we also want to make sure that people understand the majority of these companies are still at a very early stage and uh it's exciting it's a lot of opportunity but it's also has a lot of shaking out you know i would not be surprised if we're asked to do an update to this or refresh to this in six months or a year because i think there's so much that's happening in terms of the dynamic nature of this space.

5:56Gene Demaitre: Yeah, and as you said, there's over 200 companies in the table. I have over 250 companies in the database. Some of them couldn't verify, as you said. You know, they've come and gone. Some of them are already obsolete or out of business. I'm still tracking them as a tombstone. But, yeah, it's an interesting look. So, again, kudos. You can go download this at therobotreport.com. It's a free download, as Gene said, so just look for that at the top of the webpage and go download your own copy today as soon as you hear about this. We've got a great show for you today. Our guest today is Mark Kermish.

6:34Gene Demaitre: He's Chief Technology and AI Officer at Portal Labs. Portal Labs is a contract part manufacturer, and they have rapid prototyping and manufacturing capabilities. And, you know, Gene, I was unfamiliar with them before this conversation. And they have an interesting business model to help anyone who's manufacturing physical parts, any of the entrepreneurs in the audience or students that are thinking about, they want to start a business, have an idea. ProtoLabs is a great way to get those parts built if you don't have the capabilities yet. So it's definitely a conversation worth listening to. I just want to remind everyone about RoboBusiness, which is coming up on October 20th and 21st in Santa Clara, California.

7:17It's our 20th anniversary show. It's hard to believe, but there have been a lot of folks working in robotics for that entire time. We're going to talk with some of them about what they've seen in terms of evolution. We're expecting over 100 exhibitors and more than 2 ,010 needs. Our call for speakers has ended, but we are going through and reviewing them all. Thanks to everyone who submitted their proposal, as well as people who submitted proposals for our startup alley. We're already recruiting first speakers for feature panels. But in the meantime, if you go to our websites, you can see our opening keynote is going to be on the future of robotics with Rodney Brooks, Ken Goldberg, and Jonathan Hurst.

8:01So these are people who are household names in robotics. And if you're not familiar with who they are, definitely go to our sites and look up. We're also going to have a keynote on AI featuring Andrew Davidson, Chief AI Officer at Universal Robots, Michael Vogelsang, who's the head of AI Foundation Models at Collaborate Robotics and Amazon. Save money now by registering during our early bird discount. To learn more, go to robobusiness.com. We've got you covered with all the latest news and information on our network and websites, including the robot report and Automated Warehouse Online. Mike, tell us about the news this week.

8:42Gene Demaitre: Yeah, so news continues even though it's August and some people are on vacation. Things continue to happen. It keeps us employed, certainly. And one of the things that we reported this week, Gene, was about the Q2 2026 robotics demand report from A3. And they talked about some of the numbers in terms of robots ordered. Tell us a little bit about some of the numbers here, Brianna. Yeah, well, let's dive into the overview numbers first. So for Q2, 2026, North American companies ordered over 8 ,900 robots valued at$622 million. So this is about a 4 % increase in units and a 20 % increase in revenue year over year.

9:29So not a massive increase in units, but still an increase. We love to see that. For the entire first half of 2026, we reached 17 ,995 units valued at$1.166 billion. So that's a small increase, 2 % in units and 6 % in order value year over year. So we're not seeing huge exploding numbers, but things aren't going down either.

9:53Gene Demaitre: Numbers are going in the right direction. That's good. Gene, tell us about some of the other trends that the report brought to light. Yeah, it's noteworthy that the Association for Advancing Automation pointed out that early in the year or late last year, things were kind of flat. So even a modest increase is good news. although it also means that we're still dealing with uncertain economic times. One side of that is the decline in automotive OEM orders for industrial robots. And I think that's pretty significant. Orders declined 25 % in the first half of this year, which offset the growth in the other sectors.

10:32Now, even though the other sectors, not automotive sectors, have surpassed automotive manufacturing orders, It's still a big deal because, as we know, the majority of robots that are in use in North America are in automotive production. And for all the talk about humanoids and cool next generation robots, existing robots are still largely deployed in that sector. And so that slowdown is significant. but non-automotive buyers accounted for 56 % of the units ordered in Q2. And again, I think that's good news. We're definitely seeing increases. I think the most significant ones were in semiconductors, electronics and photonics or in 35 % growth, life sciences, pharmaceuticals and biomedical 32 % growth.

11:27And here's where the number about automotive might be a little bit deceptive. Now, automotive components improved by 24%. So maybe the slowdown is that the big automotive plants have retooled or changed in direction from what they were going to do with electrification, and then they have their stable installed base. But the component providers are still modernizing. They're still adding robots. So again, there's some nuances there that we want to keep an eye on. But overall, as Brianna noted, the improvement in orders is a good sign. We also saw modest increases under 20 % for food and consumer goods, plastics and rubber, metals and metalwork.

12:13So I think, again, the manufacturing industry is still adding robots and still modernizing maybe their existing fleets. The year-over-year numbers are also positive, led by electronics. I think the data center boom might have something to do with that. Automotive components, and then again, food and consumer goods being the leaders. So that's mostly what A3 reported around industrial robots as a whole. Brianna, did you want to talk about collaborative robots for a moment? Yeah. So in recent years, I think this is only maybe the second year that they've started breaking down their totals between industrial robot arms and collaborative robots.

12:55So all of these numbers are for robot arms. They don't include autonomous mobile robots, things like that. But when it comes to collaborative robots, there were 2 ,774 cobots ordered in the first half of 2026. So that's$114 million. But that accounts for around 15 % of total units and 9 % of total revenue. So the vast majority of the robots that are working in the U.S. are still the industrial robots. For Q2, we did see 1 ,137 cobots ordered. That's$44 million, 12 % of the quarterly units, and 7 % of revenue. Where we're really seeing the traction with cobots is in healthcare. Over 43 % of the The first half of 2026 orders were co-bots for healthcare and electronics as well.

13:50We saw 36 % of first half of the year orders going to co-bots. So I think that data center boom is affecting the co-bots as well.

13:58Gene Demaitre: Yeah, I think you're right. The data center thing is one of the biggest investment trends going on. So it's all those industries that circle around that are investing. You know, Gene, it's interesting. we've watched these numbers for decades now come out of A3 and the RA before that. AMRs, as Brianna just said, aren't tracked. It'll be curious to see if they have enough leverage within the industry to ask some of those manufacturers to start reporting numbers on AMRs and ultimately humanoids as that growth takes off and whether we'll be able to get data in those segments that we can trust like we do with the industrial and cobot arms today.

14:42Yeah, I would say anecdotally, I mean, we know the leading mobile robot providers like Locus, for example, aren't doing pretty well. They are scaling their fleets. They are growing. But that's not necessarily indicative of the entire category. And warehousing and supply chain, obviously, we have a whole site that covers that. But it is interesting to see how that is also having its own ups and downs relative to the broader industrial space. And it's funny, we're talking about$100 million being a good number, but then when you compare that to how much do cell phones make or how much do these cars make, all the products that are made with robots, right?

15:20And to me, it's interesting that robots are such an economic enabler, despite growing only 1.1 % year over year in June. We have to really look at the bigger picture and how robots contribute to that. So, you know, A3 knows that there's a lot of quarterly fluctuations, seasonal fluctuations that we have to take into account as well. They're going to take a conservative view about long-term growth. And so I think we have to be optimistic. But at the same time, I can understand if you're a startup in this space, you've got to be cautious. You got to know your commercial plan because if you're bringing a robot to market and the market is tough, you have to have a real game plan and some good corporate partners.

16:13Gene Demaitre: Yeah. And as long as I've been in this industry and the capital equipment business, you know, we just talked about the first half numbers. the numbers are always back and loaded right capital equipment always uh peaks near the end of the year as uh we you know the industry's trained their buyers to wait until uh they have too much production and they can get a deal on uh you know buying industrial arms and cobot arms and what have you so that's just the cycle we all get it we all understand that so you know who knows if this is what the first half of the year looks like uh let's hope that the second half of the year is lining up to be even better.

16:51Gene Demaitre: So we'll come back to that in a bit. So Gene, another interesting story this week. We didn't report on it, but talking about Uber is selling its stake in CERV Robotics, a company that we have followed quite closely and had Ali on the podcast, the CEO of CERV in the past. Tell us a little bit about this story. Yeah, well, we've We covered Uber and Serve Robotics in the past and various partnerships around delivery robots. And we're talking about what's related to Uber Eats. According to Bloomberg, they parted ways over differences in delivery strategy, which indicates that that delivery market might be in for some change.

17:38We know, for example, that Starship is getting out of college campuses, right? And that's a big deal as well. So we're seeing some evolution, maybe some economic contraction in that space. The fact is that most delivery robots have had difficulty branching out. I mean, they might be in multiple cities, but they generally have an easier time on a campus, whether it's a corporate campus or a college campus or retirement community, than on public streets. And I would not be surprised if both the delivery services and investors aren't getting perhaps impatient. But, you know, Serve originally spun out of Uber as Postmates.

18:25And that's sort of interesting because Uber had acquired Postmates for roughly$2.5 billion in 2020, spun it out in 2021. then they have continued to work together. But the volume has been dropping. The strategies were obviously not in sync. And even though, again, I don't want to say that delivery robots are not successful because service delivery volume reportedly grew 50 % in a single quarter with a non-Uber delivery partner. And maybe there's some disagreement over exclusivity there as well. So the delivery robot space is interesting, even though it's a little bit opaque because it's clouded by where they're delivering and how much a portion of the ride-hailing services or the delivery services are using that.

19:19The payload capacity may be part of the issue. Local regulations are certainly part of the challenge. So we want to see these things succeed. but you know there's been revenue declines uh serve had um it had 400 percent revenue growth but then the removal of demand from uber is not going to hit it this year you know and

19:47Gene Demaitre: interestingly enough i had a briefing this week with another startup called no tips They are going after the space for suburban and rural delivery of food. So I was excited because I've been a rural community, but their solution, which I'll report on here in the next week as I write up information from this briefing, they basically, you know, they're attacking the fact that the add-on costs for human delivery of Uber via Uber, you know, adds, almost doubles the price of the food. You know,$30 for food,$30 delivery fee. Who wants to pay that kind of thing? So they're coming up with an automated solution to get food out into those areas without driving along the sidewalk.

20:31Gene Demaitre: So more to come on that. But it's still an interesting story and one, you know, worth watching. So, yeah, we'll keep following this arc and, you know, stay tuned for more. So, hey, the last story here is an interesting one, one that I think could spiral out of control, but it's a little bit reactionary. We didn't report on this. It just came over the wire today. And it's the fact that San Mateo County might be the first county to regulate humanoid robots for commercial use. So all the things that are happening here, here's an example. KQED reported this and various other local sites have reported on this in the Bay Area.

21:19Gene Demaitre: And the story goes that San Mateo County Board of Directors, now San Mateo is right middle of the peninsula, south of San Francisco, north of San Jose. There's a lot of tech companies there. They were one of the ones that banned, I think, sidewalk robots, you know, a decade ago, coming off that last story. Gene, here they are again taking note. So they adopted a resolution on Tuesday that could make it the first to require companies to pay for permits before humanoid robots entered the workplace in San Mateo County. And, you know, it's interesting. They're basically directing their legal staff to draft a formal ordinance that would regulate the robots, you know, in areas like hospitality, logistics, customer service, emergency services.

22:08Gene Demaitre: So, you know, I'm curious, you know, your guys' thoughts, whether you think this might become a trend, whether it's warranted or not. The idea here that counties could make it a permitting issue for these types of systems to be deployed in the commercial and ultimately, I don't know, consumer use. Thoughts, Gene? On the one hand, it seems a little bit premature, right? Yeah. There are widespread deployments of humanoids yet. And humanoid in human workplaces is also still a little ways off. most of the humanoid trials that we've seen right now are in manufacturing settings that are pretty well controlled and they're cordoned off from being near humans despite what you might have seen at trade shows and um you know i appreciate that they're trying to get ahead of it especially because they are in a county where very likely there will be humanoid robots before the rest of the country.

23:10But at the same time, I'm not sure what this signifies other than maybe some politicians trying to get on the public notice as being forward thinking on this. The bigger issue that I think this illustrates is the need for national standards around humanoid robot safety, around humanoid robot operations or metrics on measuring them. We talk about this all the time, Mike and Brianna that there's obviously a need to get ahead of that aspect of things as these robots are being developed and tested. But having a law in the books that says, oh, companies have to pay for permits, it's shaking them down before they've even made any money.

23:57And so I'm not so sure it's the wisest approach. I'd have to read the ordinance more closely. But I do think it does bring up an issue that is likely to happen. We saw that with delivery robots, as we were just saying. We've seen that now with autonomous vehicles as they're starting to roll out. And humanoids, which can go in so many more places than those other kinds of robots, are going to need some ground rules. The question is who's going to define them. That's true. I think generally regulations can be a way, a pathway for the industry to sort of gain public trust for robots. Because I think right now there's so much mistrust.

24:38A lot of people feel very nervous about seeing robots in their day to day life. So they like to know that somebody else is keeping an eye on it who's not necessarily making money off of it. But I do think ideally at state level or federally, there would be some regulations that would be a lot better for the industry to move forward and to navigate those regulations. so I definitely think it's a sign to regulators at the state level at the federal level that this is going to be a really big conversation and it's important to start thinking about it and getting really educated about it because I do think this one comes a little bit from not understanding what robots are actually in the world today like there are other robots that are operating on the streets of, you know, San Francisco that aren't humanoids.

25:29So it's interesting. I definitely think this one, I don't know how far it's going to go, but I think regulation can be a good thing if we're working with the regulators and we can build jobs that way.

25:42Gene Demaitre: You know, again, Gina brings to light the California regulations on autonomous tractors, right? The fact that you can't deploy an autonomous tractor quite yet. And so that's still, you know, trying to work its way out as again, that industry tries to work out the rules. But, you know, in all those cases, there's a constituency that's, you know, alerting the, you know, the government officials to a potential issue. And I think one of the things cited in this case is the, you know, the local firefighters union has said, you know, hey, these things have huge lithium batteries in them. And so, you know, what happens, you know, when these things catch on fire?

26:22Gene Demaitre: How do we make sure that these things, you know, meet some basic requirements from, you know, quality and what have you? And being aware of where these things are deployed. At least that's one thing that a permit does is it lets you know how many of these devices are in the county, where they're operating. so that if, you know, if there is a fire and the firefighters are going into a facility, you know, they could know, hey, along this wall, there's going to be 20 humanoids being charged in this factory or warehouse or whatever, right? And just to be aware of what type of equipment they might need to deploy to a situation.

26:58Gene Demaitre: So I think that's the other side of the coin here that, again, we don't often think about, you know, as we're technologies and building new products, the impact on all the other community services and things that have to happen in the case of an emergency. So I think that part's valid. But again, we'll have to see how this happens. On the one hand, I'm sad to say I live in California where there's a lot of regulation. On the other hand, you know, California is leading here. You don't have other counties yet or governments asking these questions about the technology. So I guess it's because, you know, they don't quite know what to expect and need to be prepared.

27:37Gene Demaitre: So a lot of CYA going on here in some aspects. I think the fees from the permits are going towards the fire department so they can get special equipment and things like that. So who knows how much the fees are, but it does at least appear that it comes from a practical place. And just a reminder, we've got you covered with all the latest news and information on our network of sites that include the Robot Report and Automated Warehouse. All right, on to our featured interview this week. Joining us today is Mark Kermish. He's the Chief Technology and AI Officer at ProtoLabs. Gene, tell us what we're going to hear.

28:16Mark at ProtoLabs is a contract manufacturer. And so one of the things that I think is really interesting is normally, traditionally, when you create a robot, you have to prototype parts, and it takes a long time to turn them around and, in fact, to design them in such a way that they are strong and light and affordable and ready for manufacturing, and you go through lots of iterations. And so one of the things that ProLabs is offering is multiple ways to produce parts and rapid turnaround of those parts. And I think that's pretty important for all our developers in the audience, because whether it's 3D printing, mill the machine, injection molding, when you send a CAD design off to a contract manufacturer, it can take weeks or longer in order to turn those things around.

29:10And Mark tells us about how he works with the clients to not only design things as a nuclear as possible, but also to make sure that they are turned around in a speedy way.

29:26Gene Demaitre: Well, Mark, welcome back to the show. I think we last had you here in episode 138 about two and a half years ago. And at the time, you were working for a different company. So thanks for being here. Well, thanks, Mike. I really appreciate it. And thanks for having me back on. And certainly a lot's changed in the world of technology, let alone my own life in the last two and a half years. And looking forward to sharing it with you guys and catching up. Yeah. Well, hey, so now you're at ProtoLabs. So why don't you start there? Tell us what you've been up to since the last time you were on the show.

30:00Yeah, so I joined ProtoLabs last October as their chief technology officer and artificial intelligence guru, so to speak. And what really brought me to ProtoLabs is back in 1999, ProtoLabs solved a problem in the industry around how do you create rapid injection molding parts? Really, how do you create rapid injection molds? Back then, it took sometimes three to six months just to get a mold made. And our founder, Larry Lucas, wrapped software around the concept of design for manufacturing and machining, enabling you to use that technology to take a CAD design, analyze that design for manufacturing, build the tool pathing and machine code, and be able to submit that directly to mill for manufacturing.

30:50And so now we're able to produce injection mold, CNC milled parts, 3D printed parts, and sheet metal within a day. So if you order on a Monday morning, you can have your part by Tuesday morning. And because a lot of that software is what allows us to do it, it was a fantastic physical, I'm going to call it physical AI machine learning problem set to dig my hands into.

31:15Gene Demaitre: Interesting. Well, I mean, what's your role as a chief technology officer and especially this new title of AI officer? What does that mean at a fast custom manufacturing shop? Yeah, so a handful of things, right? You know, first and foremost, it's, you know, how do we evolve this proprietary technology that essentially automates the concept of design for manufacturing machining in today's world? especially as we're going through this resurgence of manufacturing in the United States demand is spiking there's continues to be a lack of labor in the manufacturing workforce and so robotics and automation become more and more important for our own plants to operate efficiently and so a lot of my time is spent between you know how do we create a better customer experience for the engineer, providing them real-time design for manufacturing feedback.

32:11And if you have kids that are millennials or Gen Yers, nobody wants to talk to a sales guy or anybody anymore. So what can I do on my phone or my computer? And so it's creating that full service experience. And then following that digital thread all the way through to our manufacturing, right? So it's creating tools that enable us to drive efficiencies in manufacturing to improve quality and on-time delivery, you know, at the pace that we have to operate on. And I can talk through some examples of how we've specifically applied artificial intelligence to accelerate innovation and accelerate quality or accelerate part preparation before the manufacturing processes, et cetera.

32:55And so I get this big playground to play with to apply the latest technologies to create a super fast manufacturing process for our customers.

33:30Gene Demaitre: One of the things I'd be fascinated to understand is, you know, at the actual shop that you guys operate, what's your favorite manufacturing process to watch in action, whether it's, you know, all the various tools that you have to share with production? I'd say there's, I mean, there's probably two that I really enjoy. Like, there's nothing better than watching a CNC mill with essentially milling hard metal. Right. Right. because you're taking what came in as a rectangle or square block of metal. Typically, it's six to eight inches tall and six to eight inches wide and watch a very intricate part being created out of that.

34:11I think a fast follower second would be what's called MJF printing. This is really lightweight material that's typically used in, right now, drone applications are both fixed wing drone as well as quad copter drones and the material is light as a feather right and it's super strong it's got the same strength characteristics as carbon fiber and so as you kind of get this powder bed and you slowly see that part emerge from the powder is super fun to watch something come to life well mark you just mentioned that hardware design the eucentate months can now happen in days which is good for roboticists But how does Protolab optimize that path from CAD file to physical testing?

35:01Yeah, it's a great question, Gene. A lot of it is really around how we've built machine learning models to assess the manufacturability of the part and then to simulate the tool path that is required to mill that part in particular. We've actually had a GPU farm for 20 years. and we were, I'd say, ahead of the curve in that regard in how we used it to do this tool pathing assessment and creating a simulation of that. And so I can actually go into our backend systems and I can pull up a CAD file and see exactly the tool paths that would be run from a particular machine. We then take that tool path and we generate what's called the G-code, which goes right into the controllers of these mills.

35:49And that really is how we create the digital thread from taking it from the CAD file all the way to the machine. And, you know, one of the benefits for us when you talk about just efficiency is we don't need skill labor to run these machines. Right. So on a daily basis, we have operators that can, you know, they've got to clean the machine. They've got to put the new block of metal and set the fixtures and the jigs. Then they close the door and they just click a button. Right. They don't have to worry about programming that controller. And that way, most of our machinists are really focused on preventative maintenance, right?

36:20Making sure those machines and the tools in those machines are rotated and maintained such that we can deliver the quality and tolerances for our customers. Well, and speaking of quality, you mentioned, obviously, working with metals and complex geometries. I think that's one of the things that's really interesting. On the one hand, you're not requiring as much cognitive load from the operator. But with the machinery and with automated design for manufacturing, you are able to maintain that quality. Can you talk a little bit about how things like ProDesk can catch errors before they manifest or before, you know, you're running a 3D printer or cutting this metal?

37:05Yeah. So I think it's a great question. And so, you know, with ProDesk and the design for manufacturing capability, you know, we take a look at that tool pathing, right? And one of the things that allow us to move with speed is we've standardized on block sizes. and so you know unlike a lot of smaller machine shops they kind of cut the raw block to fit the part but because we standard on the side of block size I know exactly how I can run and orientate that block with the simulated tool pathing we then take that tool pathing and where we run into issues right if we find the tolerance that's too tight you know inject your sorry injector gates or ejector locations for a mold are in the wrong spot, right?

37:51Or you've got a specific seam that may be an issue on a mold as you start to inject the material into that mold. We can highlight that in 3D, you know, with a rendered part given to us in that CAD file. And then so when a user logs back into ProDesk, they'll get that feedback and it'll actually have it broken down into six or seven feedback sections. And when they click on a section like ejector locations, If there's an issue with it, you'll see a highlight on that CAD model with an explanation of what the concern is and a potential recommendation for a solution. Or a call to action if they need some direct consulting help or they can reach out to one of our application engineers at no charge to kind of have a conversation.

38:35Threading is a good example where certain threading needs to have a discussion sometimes versus not. or if you have a two-sided part, how they want that part oriented as the part gets flipped to make sure they get the right quality. There's also certain scenarios where we can't automate the removal of all the material, right? So if you've got a block and somewhere in that block, you've got a groove right in the middle, our tools may not be able to get down there to remove that. And so, yes, we have to ask two questions. Hey, we can deliver it to you without that material removed and you can mill it in your own shop, but at least you get the part quickly.

39:10Or we can run a secondary operation and we can take that metal out after the main part's been milled. But that sometimes adds time. So depending on their desire, if it's like, hey, I need the part in 24 hours, most likely we'll get that part with the material still left in. If we need that part in three days, that gives us the time to run that secondary operation and give them a completed part. As Mike mentioned before, with your AI role, are you deploying AI as part of this process as well? We are in a couple of different ways. So we've created some internal AI tools that our application engineers are using to essentially assess the manufacturability, especially taking 2D technical drawings and be able to essentially take those technical drawings and run it through a similar process for manufacturability.

39:56And also getting that DFM feedback where then we can annotate that 2D drawing and then repackage that and send it back to the customer. We've also done what we call part similarity search. and so internally as we're evaluating a particular part you know our internal team can go in and say what are similar parts and geometries uh metal metal types or material types that we've made what was the design for manufacturing feedback what was the tool pathing we talk about everything in milling minutes right so how many minutes were used to mill this part what tools were used to that part um so we can give a more comprehensive you know review of feedback for that customer and then we've also kind of traced it down where like especially in 3D printing optimizing what I call the print box in a particular printer of how many parts can I fit in what orientation does that part need to be in how do I build the scaffolding depending on the the process of 3D printing we're using so the part has the support as we're printing it we've created AI capabilities to help us optimize that part placement and maximize for quality as well as part count so we can deliver the most cost-efficient print experience for the customer.

41:10Gene Demaitre: You know, I find that such an interesting response. And one of the things I've learned coaching high school robotics is trying to teach the kids about design for manufacturing because we 3D CAD things and then we 3D print them. And the kids design these amazing parts and And I look at it, I said, you know what? How are we going to orient that on the printer, right? You got six different ways you can orient it. What's the best way? What's going to take the fewest amount of supports and all of that? And a lot of times it's like, let's have a discussion about redesigning so that it's easier to print.

41:45Gene Demaitre: It's more successful in printing as well as, you know, of course, orientation of layers for strength and those types of things that a young designer doesn't bring into it. And maybe you have the same sort of experience with experience designers about understanding the different sort of characteristics of the different things they're trying to print, right? I got this great idea, but, you know, now it needs to become something real and strong. So how do you work with, you know, your clients, roboticists, to choose between, you know, whether a part's going to be 3D printed or CNC machined or other types of processes based on the part's, you know, use case?

42:32I think you hit the nail on the head where we got to really understand that use case to start. I'll give a couple of examples, right? If I'm in a new product innovation cycle and I'm thinking about a robot in particular, I'll use a medical robotic part where I may have a combination of plastics and metals as they're kind of managing this part or this robot. If it's in the very beginning, it might be, hey, just to get an idea of the shape and geometry and how this fitment might work, 3D printing in a basic polymer might be the fastest way to get that part in your hands. and we can print a couple of those parts where you can print dozens of those parts that are pretty low cost.

43:15But if they start to move up that what I call technology readiness scale, they're now kind of hitting that TRL five or six or moving more towards what I call product introduction. So they've got to think through the manufacturability. It might be like, look, this part is most cost-effectively done through an injection mold instead of through a 3D print. We can control the tolerance a little bit better. They control the smoothness. We can heat treat that part differently. And why don't we think about moving it down that path? Or again, if it's a metal part, it might be like, okay, let's move from 3D printed metal, which can be a slower process because you're only moving as fast as the printer, versus CNCing that part through the traditional CNC machining.

43:59And again, it often comes down to what's the material type? What's the tolerance you're trying to hit? What's the quality that you need that factors into all of those decisions?

44:09Gene Demaitre: Yeah, you know, it's interesting. Folks on the podcast who've been longtime listeners know that I'm a fan of a show called Smarter Every Day. The host is Dustin. Maybe you know of him. You know, one of the things that he's positing to them, so I'm curious, what you're seeing is that our manufacturing process today is lacking that skill set to create tool and dye, to create injection molding plates, right? It's a skill set that's retiring out of the industry and not being passed along to this next generation. So I know you do injection molding. It's a complex process to build a mold. What's been your experience in terms of finding that skill set, bringing that skill forward so that you can help your clients to build molds, to build new dyes, that type of thing?

44:59Unfortunately, I would say we don't find a lot of depth in that skill set in the United States. It's something that we have to train in. And so it's bringing somebody who may be coming at it. If they're right out of school, we get to mold them kind of from the beginning. They may be like we're as much apt to hire an application engineer out of the CNC machine shop and then retrain them in injection molding. Because at least they'll understand the concepts of how to mill the mold. and then we can train them on what the characteristics of how you design a mold is and how that actually flows into the implication of the actual injected molded part.

45:35Injection molding also has an aspect of material science that I think is important for the application engineer to know. What's interesting is, again, we've built a handful of internal tools, mainly using machine learning, where we can simulate the viscosity of a particular material type And we can show how that viscosity goes across the mold based off the temperature and expected quality outcomes or constraints that the customer gives us. And so we can, you know, not in every case, but probably 80, 70 percent of the cases, I can have that simulation where the application engineer can see what may happen.

46:13They'll have a conversation with the customer before they even actually shoot the first part mold or even before they make the mold to kind of have a dialogue of like, hey, here's what we're seeing in our simulations. Let's talk through the implications of that where we may need to see some design changes.

46:28Gene Demaitre: So at Protolabs, an application engineer is the first contact with your client, potential client, about talking about their design, talking about how it's going to be produced, and sort of engaging with them to talk through the processes of what they want, what they need, and what you can deliver and that type of thing, right? Yeah, there's always an account executive that's typically in the middle of that too. What's interesting is a lot of our account executives have become very technical. And so, you know, they can often be a first stop for basic information. Some of our account executives, I'd say, have become borderline application engineers just based off of the necessity.

47:06But you're absolutely right. That application engineer becomes a core part of the experience for that customer. and as we think about artificial intelligence we're really looking to build a lot of these tools to using us internally as the first customer as we refine these tools and we feel a level of confidence that we can expose them to our customers and get the same service you know we'll start to expose them to the customer as well so they may be able to do some more self-help so today like we know there's a high percentage of our injection molding customers that use design for manufacturing before they're ever ready to even talk to us about making the part, right?

47:43They're just trying to get a sense, will this design work? And what do I need to think through? Especially because of the complexities of injection molding, we're finding a lot of engineers just don't understand the process well enough. And so they become reliant on our DFM to help teach them along with some of our supplemental educational materials you can find on our website. Well, Mark, you brought up an interesting point about using AI internally, even before you're dealing with the customer, even before the customer is dealing with the machine learning aspect of the tools. And certainly we hear a lot right now around how AI is enabling a lot of things, but we're still sorting out where it's best applied.

48:23Beyond that hype, where are you seeing AI actually delivering a return on investment in custom manufacturing? Is it in the tool path generation, the anomaly detection, the quoting? And there's a lot of different elements here to which you can apply various flavors or aspects of AI. But where are you seeing the most value right now being delivered, given that human talent is very scarce? I think I'm going to start out with a negative before I get to the positive. Right. Because this is one of my biggest pet peeves in manufacturing is the concept of preventative maintenance. Right. I haven't seen AI make an impact yet on preventative maintenance.

49:06And part of that is the lack of data that we really get from the machines themselves. It's a problem set we're trying to work on ourselves again, right, and how we connect our machines and use AI to help us ensure we've got highly maintained machines. Setting that aside for a second, when I flip back to the positive side, there's a couple of really interesting areas that I think are really exciting. first and foremost is with, you know, agentic coding capabilities, right? If it's cloud code or if it's OpenAI's codecs, we're able to actually empower application and manufacturing engineers to write applications that solve problems on the line, right?

49:47Sometimes it's a quality problem. Sometimes it's, you know, capturing capacity versus demand and getting a better capacity model. Sometimes it's managing tool life, right? At one point, do I need to replace this tool? How far can I run the tool before I start to see quality degradation? One of the more unique things that I thought was a really cool use case was managing visual quality inspection. And so we had a handful of application and manufacturing engineers come together and they designed with Claude Code an AI model. They downloaded an open-wave model to a laptop. They used some of our internal tools to do simulations.

50:31So instead of trying to find defective parts that we've made, they just started simulating all the different variants of defects that we could see on particular parts. Then they built a cage with a set of serial cameras and fixed lighting where now we can run parts coming off the line through. and they're comparing that part back to the CAD design and then using the trained AI model on all the simulated defects that we had created to determine is that part of quality or not of quality. What's really cool is this was probably a side project. Maybe there's 40 hours into it, but if we were going to buy this off the commercial shelf, we would have spent hundreds of thousands of dollars doing this versus maybe tens of thousands of dollars that we spent.

51:18But that's where I really see the excitement around the applicability of AI and manufacturing and starting to see impacts. The other one is being able to expand the number of folks that can program a cobot. Again, because we can apply a traditional AI coding agent where now they don't have to know Python or the machine code to control that. They can just kind of say, hey, I need the robot to do X, Y and Z. Here are the measurements of the cage that robots in. here's the part type. Oh, by the way, you know, I've got a claw on the end or I've got a pincher on the end or whatever it might be. And they can start to program that.

51:56Now, all of a sudden, I don't need to have a application engineer with that skill set necessarily on every cobot, right? I can train an operator to do that work. But a lot of what I see is AI is still very valuable where there's an aspect of a human in the loop, right? And I think because there's so many complexities and variants in manufacturing, making sure that there's somebody that's constantly validating the output and being able to retrain those models on an ongoing basis, which is why we're starting internally before we get something to go external to our customers. We have to have, I think, at least a 96 plus percent confidence rate in the output, because the worst thing we could do is have the wrong information go out or a hallucination go out.

52:41And what's interesting is we're still seeing in some of the examples we talked about, you know, design for manufacturing, part orientation, material selection, right? Some of the material simulation, depending on the particular process we're talking about. Well, looking ahead for a moment, because you said that you've seen a lot of changes in the past two and a half years. So looking ahead, you know, three to five years, how do you expect these tools to continue evolving to help roboticists optimizing topology or lattice structures or managing assembly weight? Because it's something that right now, you've told us what the state of the art is, what you're able to do, but where is there room for even a higher level of assistance?

53:30I think there's, you know, thinking more in the future and being a bit futuristic about it. I can see a scenario where the robot has models on the hardware itself embedded in its software that's gathering data that tells you how the efficacy of the robot is moving through its jobs and tasks. And I can collect that data back and I can use that data to kind of improve the design of the machine. And so I do think there's an opportunity where the machine, the robots themselves become a self-feeding loop of education and training. to help all the way back to the manufacturing capability, right? Where a robot could literally communicate, hey, I've been told to do X, Y, and Z movement with my hand.

54:14I don't have the strength to grip, or I've got a digit that isn't performing to the expected tolerance. That feedback could come back. It could either autocorrect, or it may need to go back into the product development or software development lifecycle. I do think that computer vision is going to get more and more prevalent in manufacturing. I can see a world where essentially my whole manufacturing plant is fully in instrumented, where I've got IoT data coming back, I've got visual inspection data coming in, where I can make adjustments on the fly. I think the tool sets will just get more and more advanced as compute power continues to expand around simulation.

54:56right so if i can actually simulate not only how the part is going to be manufactured but i could actually say okay i've got a digital twin of the robot right so if i'm making a part change or material change i could run simulations around the physics of how that robot's moving right and get a pretty quick feedback loop in the early stages of design instead of having to wait until i've got the physical hardware fully assembled and so those those learning cycles should get shorter and shorter in my mind as we continue to go. And I think a lot of the areas of focus are really going to be around, you know, call it design for maintenance of some of these machines.

55:35So after I built the machine, how do I maintain that machine over a lifespan? Because I don't think we know yet what is the average lifespan of a robot, right? So I think about like, you know, figures robot that's come out and it's starting to get into manufacturing scenarios. You know, It's great at pulling and pushing, but it's got a weight limit, so it can't carry a lot. But how do I know when I need to replace the exoskeleton fabric exterior? How do I know when I need to replace a joint? How do I know when the software's end of life? All of those things are interesting challenges where AI can potentially help in simulating that and bring it back into the design for manufacturing phase as you're building your product.

56:18Yeah, humanoids are particularly complex machines, so they're going to need that kind of insight for maintenance. But you mentioned the lessons learned. What are some of the most common mistakes that roboticists make today when it comes to transitioning from those low-volume prototypes to production and tooling and molding and all of those processes? Yeah, so I think where to start is a good question, right? I think there's a handful of things, right? If you break it down, like in the upfront design, DFM mistakes that often happen, especially like if you're playing around with a printing, like a 3D printing versus molding, and it's assuming they're one and the same.

57:05So if you know you're going to have to have an injection molded part, but you're experimenting with 3D printed parts just to kind of lock in that design, the characteristics as you move from 3D printing and injection molding are not small differences, right? So you got to make sure you're continuing to build that design process back into the permanent part that you're looking for. I think there's things around like non-uniform wall thickness, missing draft angles, sharp internal corners. Those are types of design components that are pretty minute details that show up very differently when you injection mold a part versus when you're printing that part.

57:44Um, also things like not accounting for shrink or resin switches, right? So you may be printing on, you may be injection molding on a resin that goes end of life, or there's been a chemical change. Um, you know, understanding, uh, the characteristics are on abrasive or heat tolerant or cold tolerant plastics. Um, as you move into like tooling, uh, some folks completely skipped the concept of tooling and how they design that, that for the manufacturing. I think the other piece is you've designed for the prototyping versus the end state. So in prototyping, you've typically got a higher level of tolerances that you're dealing with.

58:25I've got more tolerance drift, so to speak. I'm okay if a part wears out after 1 ,000 bends instead of 10 ,000 bends. And especially with complex robots, that can actually become a major defect issue if you're not accounting for all of that. and it's not like we have robust data or tools for testing of all the physical aspects of a robot right the delivery robots on college campuses for food are very simplistic versus you know the cobots or humanoid robots that i think will come i think there's a lot around volume and costing right and so i think that's something that we don't know yet with our robotics industry especially i think it'll be interesting to watch in the humanoid space where you know if tesla really is going to produce tens of thousands of robots that'd be the biggest scale we've seen and i don't think we know yet exactly how does that impact manufacturing the costing the quality uh and the life cycle yeah so it's been a bit of

59:25Gene Demaitre: our conversation talking about prototyping helping uh young companies early projects you know get get some parts built so that they can assemble them, test them, you know, do all of their verification. But I know you guys also help companies move from that into, you know, early volume production. So tell us a little bit about the opportunity there with ProtoLabs. And I think you've got your ProtoLabs network with a variety of partners around the world to help companies, you know, produce where closer to where they may be located, that type of thing. Talk a little bit about how you can help them de-risk their supply chain.

1:00:05Yeah, I think it's a great question. And we often are partnering with our customers. When you start to move towards production, there's three or four things that they're looking for, right? Typically, it's now how do I cost down the cost of my manufacturing and the part in and of itself? And that's where the ProtoLabs Network, which is made up of independent manufacturers around the world, that their specialty is typically running production runs, right? But they may be in India or China or South America or Mexico, right, where you're just getting some cost arbitrage. Sometimes that can be gated, right?

1:00:38If I've got a defense or a aerospace product, I've still got to stay in the United States, right? So then if it's in the United States, it's often working with them. And we do have a partner network here, but it may be working with them on designing, you know, how cost and complexity in the part, especially when you start to bring in the assembly side of it right so we'll go through a process where we kind of call it manufacturing design for assembly and so the less joints the less number of screws right all those types of things you know go into consideration there's also you know material consulting right where it's like hey stainless steel is great and it's got a longer shelf life but the cost of aluminum is half that and the cost to replace that part might be worth it so i get call it two years of lifespan instead of four but even two you know replacing that part twice is still less expensive than stainless as an example as you get into the polymers plastics and resins it just gets more complex and there's lots of choices and sometimes it's hey you know you need to be thinking about buying ahead of your needs for your your materials such that you can lock in your costs you're not dealing with the constant commodity price change and we'll even store materials long term or they can run short runs against those materials using their that we hold on to for life in that regard.

1:02:02And then we'll often even set up specific lines for customers, right? If they're able to commit to say, hey, we've got a 1000 part run a year or 10 ,000 part or 20 ,000 part, those still fall within our sweet spot for our factories. And so we can often fill in that production form because if you're going overseas, most of those competitors or companies are looking for 100 ,000 parts to a million parts to be efficient in pricing. And they'll turn jobs down around that are small. So we still play very heavily in that low volume, high mix space, which we can help kind of fill in the gap.

1:02:38Gene Demaitre: Mark, the other thing that I, in researching for this, I noticed that you're obviously You serve in a lot of industries like defense and medical and aerospace that have different documentation requirements, compliance requirements. Talk about how that may be a part of the process and how you have those conversations with those types of clients about how you can help them, you know, achieve their compliance goals. Yeah, and I'll start out maybe qualifying that a little bit is we believe, I believe, that there's a huge opportunity for AI to actually ease this compliance burden around documentation.

1:03:18where I can now feed in if it's an ISO standard compliance or it's a federal regulatory like CMMC compliance. I can feed all those requirements in. I can then spit it out based off of the type of part we're producing and saying, hey, here are the things that we have to ensure the qualities, inserts, and documentation that you need. I can then run the documentation we're producing back through that AI model and making sure I'm hitting all the requirements, right, as an augmentation to the manual review. But it is certainly not for the faint of heart in regards to managing all of these compliance and regulation components.

1:03:56For us, we've really built those capabilities and those checkpoints into our manufacturing execution software. All right. So when a part comes in, we tag that part. Hey, this is an ITAR part. It's built for aerospace, which then kind of dictates what are the documentation needs that we have to follow. and then we can you know also how does that part and the data around that part need to be treated and so if it's itar based right you've got citizenship requirements you have to adhere to if it's cmmc and cui based i just have to have a need to know documented i need to know that this part is moving through so if i'm a quality uh operator or if i'm a machine operator you know they've got to have the right and intent to to see that part so using our technology tools internally and we're continually optimizing it as the regulations change is important for us, right?

1:04:49So we can fulfill the need of the customer with the easiest burden possible. And so we haven't done a ton yet around AI. It's one of the use cases we're actually currently trying to build out to see how could that actually help accentuate the process and ease that burden a little bit more, including potentially taking the raw data and automating the generation of that documentation. And then verifying the quality and efficacy before it goes to the final manual review step. Mark, what's the one piece of technical or strategic advice you give a robotics founder launching a physical hardware product today, given the tools that are available and also the challenges that they're trying to solve?

1:05:31Yeah, I think the first piece of advice is be really clear on what is the job to be done by the robot you're building. right is uh you know general purpose robots without that real defined job to be done make it more difficult to design for purpose right it's one and you know the example would be as i've designed this robot and on the end of it i've got a claw where i need a rubber gripper instead you know that that's the first piece of advice that we go through the second piece of advice is really thinking about

1:06:06all the different fitment needs that you have as you're building this robot, and what are the tolerances and the types of movements, because that's where your wear and tear is going to happen. And if you don't understand where that wear and tear could occur as you're building and designing, you may end up with material choices or design choices that come back to bite you in the end, because you're not getting the lifespan of the part or the movement of those parts around the robot or you expected it to be right. I think we've probably seen it, right. You know, a robot, you know, breaks in a manner that we didn't anticipate, right.

1:06:42Or the software glitches and all of a sudden the robots doing a movement you didn't anticipate and it breaks apart. And so a lot of that, you know, there's a little bit of chaos theory of, of having to think through like what are all the things that can go wrong that I have to factor into the design.

1:06:56Gene Demaitre: Well, Mark, it's been great catching up with you. I think you've landed in an interesting role at an interesting company. What a great place to sort of keep your thumb on the pulse of innovation and what's happening. I'm sure you've got lots of interesting war stories, talking to clients, as you said, looking at parts. So once again, thanks for joining us today and best luck with everything going forward. Yeah, thanks, Mike and Gene. It's been great to catch up, and I look forward to doing it again soon.

1:07:34Yeah, Mike and Brianna, it was really interesting to listen to Mark discuss the role of AI and robotics in manufacturing, right? Because we often think about the robots doing the manufacturing, but we don't think about the robots being used to make more robots. and it's an area where there's obviously still room for improvement and AI is already delivering value in areas like manufacturability review, part and similarity searches, visual quality inspection, optimizing print box placement and helping non-programmers control collaborative robots. So I think there's a lot of opportunity still for AI to move into new areas but it's also worth noting that Mark is skeptical about AI's impact in preventative maintenance because there's simply not enough data or the right data available yet.

1:08:25That will no doubt improve as we get more sensors out into the world and more equipment has instrumentation, whether it's an autonomous forklift with sensors or robots with sensors on their wrists and arms, and again, not just on their heads. But we're hearing similar things from manufacturers, that there's a need for the right data so they can not just deal with problems, but look ahead, anticipate, even in the design stage.

1:08:54Gene Demaitre: Yeah, and Gene, I love the discussion where we talked about prototyping and moving prototyping to production. Obviously, Proto is in their name. So, again, for folks that are in the audience, ProtoLab seems like a potential vendor for you to help get your first set of parts built. But, again, this moving from that initial prototype to production, I love the discussion we had about design for manufacturing and the types of conversations that their advisors, their application engineers have with clients about taking an idea and actually making it now producible in volume and some of the things that they have there, issues like tooling constraints, draft angles, and what have you.

1:09:41Gene Demaitre: So that was cool. That hit home for me because I've lived that as a product manager, building physical products, the whole process of going into dance manufacturing to build the first couple devices and then actually building a line to build your production devices is something every hardware manufacturer has to deal with. And again, ProtoLabs can help support that. And on the other side of it, Gene, I like the discussions about, you know, how they help support low volume production in their supply chain to build parts for their partners through their partner network and to be able to build anywhere around the world, you know, that their clients need, you know, manufacturing done.

1:10:24Gene Demaitre: You know, the last couple of weeks, we've been talking about the FCC ban and getting parts built, you know, in country. They can do that for any partner in any of the various countries that they're in. So very cool to see how they've built that whole network out across the world. So I want to thank Mark for joining us today and for his background on the company. So before we go, we have just another reminder about the upcoming RoboBusiness event. It's going to be taking place October 20th and 21st in Santa Clara, California. Like Jean said at the top of our show, it is the 20th anniversary show.

1:11:04So we'll be looking back at 20 years of RoboBusiness, looking forward at the next 20 years of robotics. We're expecting over 100 exhibitors, over 2 ,000 attendees, and we are still having our early bird discount. So register now to save with that early bird discount. to learn more, you should go to robobusiness.com. Great.

1:11:25Gene Demaitre: Well, that's going to do it for episode 257. Thanks for everyone for tuning in. Thanks to Mark Kermish, BrotoLabs Chief Technology and AI Officer for joining us today. And Gene and Brianna, thanks for joining me on the show today. Appreciate your voices. New episodes of the Robo Report podcast drop each week. You can find us anywhere you get your podcasts. Please subscribe, leave us a rating. Most importantly, tell your friends about us. This episode was produced by Mike Witzman, edited by Tyler Kubisti. Our executive producer is Amanda Buhner. Thanks for listening.