We're sunsetting PodQuest on 2025-07-28. Thank you for your support!
Export Podcast Subscriptions
cover of episode Machina Labs’ Edward Mehr on Autonomous Blacksmith Bots and More

Machina Labs’ Edward Mehr on Autonomous Blacksmith Bots and More

2024/9/4
logo of podcast The AI Podcast

The AI Podcast

AI Deep Dive AI Chapters Transcript
People
E
Edward Mehr
Topics
Edward Mehr 阐述了 Machina Labs 如何利用人工智能和机器人技术来革新制造业。他指出传统制造业的局限性在于为每个部件都需要专门设计和建造工厂,这导致成本高昂且难以改变设计。Machina Labs 的目标是构建一个“软件定义工厂”,通过人工智能和机器人技术,实现工厂设计和运营的灵活改变,而无需改变昂贵的机械设备。他以 SpaceX 的 Falcon 9 火箭为例,说明传统制造业对硬件的依赖性,以及改变设计所带来的巨大成本。他认为,3D 打印技术虽然提供了一种可能性,但仍存在几何形状限制和经济性问题。Machina Labs 的解决方案是构建一个名为“机器人技工”(robo-craftsman)的系统,该系统结合了传统手工技艺的灵活性与自动化生产的可扩展性。这个系统能够模拟工匠的技能和思维过程,并通过 AI 技术来优化制造流程。Mehr 还详细介绍了 Machina Labs 工厂的布局,与传统工厂不同,它采用模块化的“单元”(cell)结构,每个单元由机器人组成,可以编程执行不同的操作,物流系统与制造过程解耦,提高了工厂的灵活性和可扩展性。在技术方面,他们使用工业机器人手臂,并通过收集和利用数据来训练 AI 模型,逐步提高系统的自主性。数据收集面临的挑战在于模拟制造过程中的物理现象非常复杂,因此他们结合了实际操作收集数据和利用 GPU 加速模拟的方法。最后,Mehr 对制造业的未来发展趋势进行了展望,他认为人工智能和自动化技术将深刻改变制造业,推动制造业回归美国,但这一转变的速度取决于地缘政治因素等多种因素。家用人形机器人普及还需要更长时间。 Noah Kravitz 主要负责引导访谈,提出问题,并对 Edward Mehr 的观点进行总结和补充。他从机器人技术在制造业中的应用、软件定义工厂的概念、AI 在机器人技工系统中的作用等方面进行了提问,帮助听众更好地理解 Machina Labs 的技术和理念。

Deep Dive

Shownotes Transcript

Translations:
中文

Hello, and welcome to the NVIDIA AI Podcast. I'm your host, Noah Kravitz. One of the most fascinating, dynamic, and visceral applications of deep learning, machine learning, computer vision, all these terms that we now sort of put under the AI umbrella, is the world of robotics.

Robotics has been a vibrant space for years and years now, but since the advent of large language models, the interest in robotics, the exposure it's getting to the mainstream audience, and certainly some of the ideas about how we can communicate, train, and leverage robotics

has expanded. Here today to talk to us about the world of robotics, specifically in the manufacturing industry, is Ed Mayer. Ed is co-founder and CEO of Machina Labs, and I am absolutely delighted to welcome Ed onto the podcast. This has been in the works for a little while now, so I'm excited to hear all about what Machina Labs is up to. And trust me, it's some really, really cool stuff. Ed, thank you so much for taking the time to join the NVIDIA AI podcast, and welcome. Thank you. Great to be here.

So let's start at the beginning, if you will. Tell us a little bit about what Machina Labs does, maybe how it got started.

And we'll go from there. Sounds good. So at Machina Labs, we're trying to, or we're building the next generation of manufacturing floors. And the main enabler is artificial intelligence and robotics. The main challenge that we're trying to solve is something that I learned in my past careers. And it is that every time you have to build a physical part, a part made out of the material, a design that's made out of physical material, you pretty much have to build a factory around it. There's

There's a lot of machinery and equipment that you need to build that are specifically designed for that part, for that geometry, for that material that part is using that cannot change, cannot easily change. Every time you want to change the design and material, then you have to go change these machineries and these toolings that you have to deploy in your shop.

So what we're trying to do is, can we build a truly software-defined factory? A factory that can change its design, can change its operation without having to change its machinery, which is very expensive. It takes a long time. Right.

And, you know, these challenges have been prevalent in manufacturing for the past, you know, 100 years. And I think right now we are at the point where both robotics and AI are mature enough that we can kind of rethink this paradigm of how manufacturing floors and shop floors have been built. So when you mentioned learnings you picked up in your previous jobs, you were at SpaceX and also Relativity Space, is that right?

Great. Great. And were you working on manufacturing and sort of the, everything surrounding the manufacturing process there as well? Yes. Yes. Got it. Okay. So I,

I have little experience with manufacturing myself. I've been living in the digital world as a content creator for a while now, but it makes sense to me that if you want to create something physical, you have to build the physical tooling and such around it to create it. And obviously, tearing that down, reconfiguring would be an expensive process. So when you talk about the idea of a software-defined manufacturing floor,

What does that mean? Because obviously I would think we're not able to turn zeros and ones into physical output as such. So how does that work? What does that mean? Yeah, no, absolutely.

So I have a little bit of a kind of an interesting, not interesting, maybe a more hybrid background. So my education... I bet it's interesting, but we'll see. Yeah, I appreciate it. My education was mostly focused around robotics and software. You know, when I went to school, academically, did computer engineering, and then focused more specifically toward machine learning, empirical modeling, and AI. I kind of spent early days of my career at companies like Google and Microsoft.

But, you know, always had a bend for manufacturing since I was a kid. I did a lot of kind of like welding and carpentry when I was in school. So I always wanted to be able to combine and bring this world of software and robotics into manufacturing. So when I went to SpaceX, that's when I kind of started to learn how tough it is to manufacture parts.

So to illustrate that, I'll give you an example from one of the projects we were working on in SpaceX around, you know, 2020, 2010, 2012 timeframe, which is Falcon 9. Once we decided on the diameter of the Falcon 9 rocket, a lot of tooling, a lot of equipment in the factory were configured for that diameter.

Right. Meaning that, you know, for example, the tooling that you use to build the tanks for that rocket are all fixed at that diameter. The moment you want to go get a fatter rocket and you want that because a fatter rocket means you can put more fuel in it and it can go to higher orbits.

but it's just not an easy change. It means hundreds of millions of dollars investment in the factory to change that. So actually, if you follow Falcon 9 throughout its life or just Falcon family, you will see that the rocket kept getting taller but could never get larger in diameter.

And as a matter of fact, if you look at SpaceX in the past 20-something years that the company has existed, they have two rocket families that they have developed. Actually, one and a half. So there's a Falcon family and there is a Starship family, which is a significantly larger rocket and still hasn't become productionized yet. So one and a half, that's why I call it one and a half. And in order to build Starship, they had to go start from scratch. New facility, a lot of new tooling out in Texas, right? Yeah.

So that's what I mean when we say manufacturing is very hardware specific. Sure. Now, there are technologies that have been developed in the past few decades that allows us to slightly move away from that. 3D printing is one of them.

That's why I've moved from SpaceX to Relativity Space. And the goal at Relativity Space is can we 3D print as much of the rocket as we can to get rid of this limitation, a limitation where we cannot change the design of the rocket if we want to.

So over there, I was in charge of a team that was building a very large envelope of 3D printing, right? You could print structures that were like 15, 20 feet wide and, you know, 20, 30 feet in length. And this was basically a robotic arm with a welder attached to it that was welding layer by layer, you know, basically what 3D printing is, layer by layer, a very complex structure that could be very large.

So for the first time, techniques like 3D printing kind of give this promise of now you can actually turn a design into something that incrementally can build a geometry and it's not tied by a specific tooling.

The challenge with 3D printing was that there's a lot of geometries that are still not accessible to it. Either physically not possible to print in, that's a lot of physical challenges, or economically, it's just not feasible. And that kind of gave a way toward the thinking around how do we build a little bit more overarching automation

that can do different types of processes, different types of material. And it's not specific to just maybe 3D printing or machining or one process. It can do different types of manufacturing processes that need it in the shop floors without having to be tool specific. And that was the genesis of Machina.

And the core idea is to answer your question, like, how is this going to be possible? The core idea is, you know, we used to have this flexibility, right? If you go a couple of centuries back, manufacturing used to be arts and crafts. People actually manufactured things. And they could one day, you would go to a blacksmith and you say, okay, build me a sword.

and they will start from a rod of material use their hammer and apply it incrementally to that material and they had a creative mind so they could you know apply different set of steps and be creative with it to get it to a sword and then the day after that you could go say oh can you build me a sheet and it would start from a flat sheet of metal and then they apply the same hammer but in a different way and it will give you a shield right out of a sheet metal

So they were super flexible. The challenge was they were not scalable. So with the current manufacturing paradigm, we kind of traded that flexibility for throughput. But I think now if we can replicate what a craftsman does in an automated fashion,

And then we can maybe get the best of both worlds. We can get the flexibility that the craftsman had, but also scale it. And that's basically what we're doing at Machado. We're building what we call a robo-craftsman. And the core components is, can you replicate the dexterity of the craftsman, which is the robotic comes into play? But more importantly, can you replicate what happens in the mind of the craftsman?

How do they come up with a set of procedures and steps using the very simple tools to get to a part that is accurate? And then because it's a robotic system, you can easily scale it. So I want to ask you to dig into that a little bit. But before I do, are there particular industries, sub-industries,

and types of projects or other parameters that define what Machina is working on and then also what you are not working on? Is it specific to, you know, are you building robotic blacksmiths for that matter? But, you know, are they aviation industry, different industries? What's sort of that end of the approach? And then maybe we can dig into some of the

the technical bits about how you're doing it. Yeah. So the robotic craftsman cell that we build, this is a robotic system. We call it a cell. It's actually industry agnostic, right? You could build parts for aerospace. You can build part for automotive and that's the beauty of it, right? Like in the morning you can do automotive parts in the afternoon. You can do airspace parts. But in order to build a good business, we always want to start from somewhere that we are very good fit based on the current economic conditions. Sure. The

develop it there, and then from there, expand to other verticals. So the focus that we have today is mostly on aerospace defense. We have some focus on automotive, but dominantly, our business comes from aerospace defense. And the reason for that is that the traditional factories are very good at producing the same thing over and over again. So we don't want to necessarily start competing with them on making the same thing over and over again. We want to compete in areas where design is constantly changing.

the parts are constantly changing and what we call is a very high mix environment. That's where aerospace and defense is a very good fit.

And they're also traditionally good adopters of new technology compared to maybe some of the other industries that have lower margins and cannot necessarily risk testing a new technology. So aerospace defense is where we start because of market factors and go-to-market considerations. But the goal is that this can apply to any vertical in the future. Yeah, makes sense. I should ask, when was Makina founded? Yeah, so we started the company in 2019. Okay.

So the company's right now like four years old, four or five years old. But yeah, but the team comes from, you know, manufacturing background. You know, my co-founder, Bobak, is a material scientist. He comes from aerospace and automotive in the past. Lots of SpaceX alumni, some of the folks from relativity space. So a lot of people who have been in the agile manufacturing space and robotic space. And then we started in 2019. And then I think we have our first manufacturing cell in 2020.

Cool. And in the interest of full transparency, this would be a good time for me to mention that NVIDIA is actually an investor in Machina Labs. All right, so let's dig into it a little bit from whatever angle and to whatever extent you'd like to, Ed.

I guess I have sort of two questions, and you can pick the one that makes more sense. One is kind of how does it work in terms of what does your manufacturing floor look like or the best that you can describe that over the radio, so to speak, and kind of, you know, how have you built the cells to be able to accommodate this mix of different manufacturing tasks?

And then the other question I have, of course, is how has and is AI playing a role in informing, you know, and I'm hoping that we get into the sort of the creative mind of the, what was the term, robo-manufacturer? No. Robocraftsman. Robocraftsman, thank you. I knew it was a little warmer of a term than manufacturer. So where should we start to dig in a little more here? Yeah, no, I think let's

starting from the kind of the layout of the floor as a good start. It's because it is different than traditional manufacturing floors, right? We have a facility, we have a couple of facilities here in LA around like 110,000 square feet now. And it's very different than, you know, what you would see, for example, if you go to a automotive factory. The traditional paradigm of manufacturing, like I said, is very much has been focused on making the same thing over and over again. And,

And the way we've been able to achieve that is dominantly because of an invention that actually we did in America, done by Henry Ford in the early 19th century, called assembly line. Where you have a product moving in a linear fashion in the factory, and each station, or along its way, people are just basically installing different things or doing different manufacturing operations on it.

So if you go into a high volume factories of today, you will kind of see that assembly line flow where material comes in from one end, they start getting put together on the assembly line. Each operation is getting done at each station. And again, you get a call or whatever product that you get coming out of it. And for my sort of non-expert point of view, the thing that I always think of is each station is doing exactly the same thing or more or less exactly the same thing over and over and over again.

Yes, and that's how we have been achieved to get very high throughput. But then the Achilles heel is that you cannot change. The moment you want to change it, basically you have to overhaul the whole factory, right? So that's why it becomes very expensive. So our paradigm kind of flips that a little bit on its head. So it says, okay, let's have these what we call cells. And these cells are enabled by robots. And each robot can be programmed to do a different operation.

And now we're decoupling logistics also from the manufacturing, meaning that now you have a facility that actually more looks like a data center for folks who are familiar with the compute world, where you have basically the same way in a data center, you have a whole bunch of computers and you can program these computers and these computers are connected together. You can program these computers to do different types of operations through software. That's the same kind of paradigm we have in our shop floor. There is array of manufacturing cells.

These cells are robotic cells and can be configured to do different operations in different parts. And then through a kind of centralized system, you can basically program what each cell does. Maybe one cell forms a hood for a car, another cell maybe it's welding it, another cell is trimming it, and the logistics completely decoupled.

because we want to get this flexibility. Now, there's a whole bunch of other benefits as well, right? Like not only you get flexibility, but you get kind of rolling updates. You can kind of like, you can bring down a cell and the whole manufacturing doesn't need to get overhauled as it is in the assembly line, right? So there's a whole bunch of other benefits that come into play. But yeah, our shufflers look very different than what you would see in a, let's say, a car factory today.

When you're talking about the robotic cells, the robots themselves, are they arms? Do they have different physical forms depending on the task? And how much does or doesn't the physical form of the robot itself, to call it that, the Robocraftsman, how much does that dictate? You know, I'm going back to your SpaceX example about the diameter of the rock being fixed and imagining, well, there's got to be some constraint on...

what one of these cells just physically can't do. Yeah. This is actually have been a very big subject of debate within our company since early days. How do we make these cells as agile and as flexible as possible? There's, I think, two considerations. One is the size of the cell and the other one is what are the components?

If you look at the industry today, there are multiple options we have had. We can go very simple automation. A lot of people can think of like Yantries and systems that are kind of basically XYZ movements that you can have in them. So that's a very simple system. And then you can go all the way to a very complicated robotic system like humanoids, right? Where they have a lot of degrees of freedom and they can do a lot of things automatically.

But then the downside is as the system gets more complicated, it's harder to manufacture. The accuracy maybe is not as good. The amount of forces they can apply is not as good. Versus on one end, you have a very simple system, can do a lot of accurate stuff on the gantry side, can apply a lot of force. And then you have on the other end, very complex system, let's say humanoids, which companies like Figure and others are trying to do. And you get a lot of flexibility, but maybe you lose power.

precision, accuracy, and the amount of load they can apply. So we kind of chose to be somewhere in the middle. So we use robotic arms. We use industrial robotic arms to kind of have the benefit of both worlds. Get enough flexibility in our manufacturing selves because we can kind of replicate almost the same dexterity as a human with, say, SACSEX or 7-axis robotic arms.

But then at the same time, you're using a kind of slightly commoditized system. So we don't have to build a wheel from scratch. So robotic arms already have existed for a few decades. There are multiple vendors that are offering the same thing. So it's more commoditized. But then it still gets us that benefit of, so it has flexibility, but it has the benefit of applying a lot of force. It's still very accurate. And we can kind of get rid of a lot of its limitations through our software stack.

So use robotic arms, but ideally you might want to have multiple configuration of these robotic arms, right? The same way in a data center, you have, you know, one system that has, you know, X amount of RAM and this type of a CPU and other system might have much larger RAM and our system might have a GPU, but hopefully we can only have like few five, six different configurations. And then we do this five, six different configuration. We basically can do all kinds of different types of parts, right?

Right now, for example, we have very strong robots that can do very thicker material. We have thinner robots that are slightly more accurate, weaker robots that are slightly more accurate, but then they cannot apply as much force. So we have a few different configurations, but together pretty much can do different types of geometries. Maybe 80, 90% of geometries out there is accessible to us with different configuration, different sizes.

I'm speaking with Ed Mayer. Ed is the co-founder and CEO of Machina Labs, who, as Ed has been detailing, is revolutionizing the manufacturing industry by combining robotics with AI and a new approach to what the manufacturing floor itself looks like,

How it can be reconfigured, what it's all about. So let's talk now about how you get the robotic cells to do what you want them to do. And maybe even past that, I'm curious when you talk about the creativity aspect and, you know, combining that sort of old world human world. There's still artisans out there making things. Putting that into this robotic process, what that's all about. Yeah.

So ideally, what is a system that you want? If you want a system that actually works like a robotic craftsman or a true craftsman, you can go to the system and say, hey, here's a design. Here's a design of a part that I want. Basically, input, design, intent. And then the system can come up

with how it will direct its kinematics, basically robotic arms, to do the right set of operation, pick up the right tool, apply it the right way to the material, and then get you the right part in the end, right? Sure. So that's the ideal solution that we want. Today, those steps are done with a lot of experts.

So there's an expert designer that looks at the CAD, maybe modifies the CAD or the design intent to get to something that's manufacturable. It goes back and forth maybe with the person who brought the design intent to figure out, okay, what are compromises you can make to make this manufacturable? Then there's another expert that turns the finalized, the negotiated design into a set of

that the robotic system can do and then programs that robot to do those steps. And at each point in time, if the robot is not doing well, maybe we do a QC and say, okay, the robot is not doing the right thing. We have to iterate again and then kind of update the robotic instructions and then see if we can get to a better part. And then in the end, you have to do some kind of a QC. The part comes out, somebody looks at it and say, okay, is it actually toward the specs that we want? Now, these are the steps that are currently done by experts. So,

So what we want is kind of combine all of those into an intelligence system.

Now, what you see has been done with ChatGP, these and others, you would say, okay, yeah, it seems like we are very close to be able to do these things. Like now we have these robotic systems that can ask it to do something that's very complex, and it will give you an answer. You know, there are news now that like, you know, the latest iteration of training of these models is as smart as it may be a PhD, right? So we have these very expert systems that can reason through with AI through these steps.

The main challenge we are facing is that, you know, these systems that you see like ChatGPD and others have been trained on the data that has been publicly available, right? You can use internet to train it. With manufacturing data, the data has not been traditionally available.

So we have that extra challenge is that, okay, yes, seems like the models are capable enough to be able to decipher and turn a design into steps that the robot needs to do, but we don't have enough data or available data to train things on them. So the extra additional complexity for us was we need to build systems that can generate data and incrementally get better. So we need to start with systems that use heuristic methods first.

not model necessarily, to generate a lot of data. And then we use the data to build models and incrementally improve its capabilities over time to get to a point where you have that robotic craftsman. You can literally put the design intent, input the design intent, and it comes out the other end, the actual part that these robotic systems are building. So do you generate that data in a simulation? How did you go about building the datasets? Very good question.

So when we started, it's always the question in these, no matter who, right. Talking to you, talking to somebody about wildlife conservation and tracking animals. It's always about the data. Yeah. Yes. Yes. So here's the challenge, right? Simulation is a good way to create the synthetic data, right?

The challenge is even our simulations today are not fast enough for our specific processes. So if you look at, for example, in some of the other robotic tasks where you do manipulation, like you move stuff around, then you have simulations that are fast enough, right? They call it these robotic gyms, right? Where the robots basically can do a whole bunch of different things and kind of explore the space and kind of get training data. Because the physics, the kinematic physics is actually pretty simple, right? To simulate.

The challenge with our process is that we are simulating a physical phenomenon that's much more complicated. You know, when we are forming these sheet of metals, we have to simulate the deformation. So plastic and elastic deformation, we need to simulate friction. We need to simulate material cohesion. There's a whole bunch of things. We need to simulate tearing. Yeah.

So when we initially, we started to do some of this work, for example, like try to simulate it, we realized that actually if we properly want to simulate the environment and our processes, it's actually very difficult.

expensive and it takes a long time. A part that would take for us to form with the robots for 15 minutes at the time on the differential equation-based simulations, we call them finite element analysis models, it would take one week on 27 cores of a CPU machine. So it's like, okay, it's cheaper maybe to just run it in real life. Just capture the data, yeah. Capture the data. But we took both approaches. So we said, okay, let's build systems that we can do a lot of trials in real life.

and make sure we build them in a way that we can capture the data, a state of this process every four milliseconds. That's what we were doing. So we captured terabytes and terabytes of data with that over the past four or five years. We built thousands and thousands of parts. But also, we are working on making faster simulations using GPUs, right? Can we actually expedite the speed of simulation so that we can also augment it with data that comes from synthetic work? Oh,

So we have to do both. But yes, it has been a huge challenge. These are the two challenges that are kind of ahead of us that maybe other companies like the ones that are working on natural language models don't have because internet is giving them user-generated data for years. Yeah, no, I don't want... You mentioned QC and it's stuck in my head. I wouldn't want the robotic QC craftsman, so to speak, hallucinating based off of at least the stuff I read on the internet. Let's put it that way. Yes. Yes.

So you mentioned, you know, the data and the training and then the speed of everything being kind of two big hurdles that you've been working to overcome. There are other either hurdles you've overcome or perhaps just, you know, moments along the way that were surprising or just kind of stand out as kind of like these big milestones in the development of, you know,

everything Machina Labs is doing. And then sort of the follow-up to that is, I guess, beyond the training and the speed, which are obviously huge. If you look ahead to the next, you know, couple of years or whatever the best timeframe is, maybe what are some of the things that you and other robotics and manufacturing companies are trying to clear to get to sort of the next stage of your evolution?

Yeah. So there's an interesting challenge when you want to generate your data, which has been probably the biggest challenge for a robotic company. I remember there was a podcast from Ilya talking about why OpenAI dropped their robotic arm back in the day. Okay. Right? And the main challenge is that, yes, you say, okay, I have to generate my own data. But you're generating data with a very complex physical system. These robots can break and they might need maintenance.

So you almost have to figure out a way to be very good at operation. Right.

You need to be able to run a large fleet of robots very smoothly to generate data. And that actually has been one of the main challenges, like I said, even OpenAI had at the time, where we're like, I don't know if we are in it. We don't have the right expertise to create this operational rigor to operate these robotic cells. Now, they were funded by billions, actually, of kind of they could go and dedicate themselves to do this. We also don't have the luxury of that. We need to create

We're a venture-funded business, obviously we have a good amount of funding, but we need to also generate results for our customers. I think one of the biggest challenges for us was in order to enable this AI-powered world, we need to actually become very good operators and create value for the customers fast. A lot of challenges that we had in the early days was like, how do we create a robotic facility that can work around the clock and generate data and resolving a lot of those challenges?

So that's been one big area of work for us. And the other piece, I think, you know, kind of like maybe a little bit away from technical work is, you know, because you are operationally heavy company, you also need to deal with a lot of people. One challenge that people... How big is Makita? How many people? We're roughly now 70 people. Okay. But across many skill sets, right? We have technicians all the way to robotic engineers, to AI engineers, to software engineers, material scientists,

And if you traditionally want to build such a multidisciplinary team, you have to balloon up, right? You have to create all these different expertise that kind of work together. So we have to also figure out a way, how do we not balloon up while we operate, um, you know, a, a huge operational facility and then generate data while we also maintain a rigor and expertise in the fields like AI and others. So I think the biggest challenge for us has been kind of navigating those waters. Um,

You know, as a technologist, you always underestimate how important the people factor is. Yeah. And I think, you know, that's something that, you know, I kind of, I can have learned over the time that it's probably the most important factor, right? A lot of technological developments are already there, but having a team that is excited and can operate efficiently has been one of the biggest challenges here. Absolutely, yeah. Yeah.

So, to put you on the spot here a little bit, how do you see manufacturing changing, you know, because of AI and the things that we're talking about? And as we go forward, I mean, five years, 10 years, 20 years down the line, is all manufacturing going to be...

you know, look very different and have this sort of flexible, scalable, reconfigurable, you know, type of design that Machina, you know, has and is continuing to build? Are we going to have to sort of knock down or gut, you know, the existing manufacturing facilities and rebuild them with this in mind? And then this may be out of your wheelhouse, so please feel free to demur on this. But

Do you see these types of things coming to, you know, sort of consumer and household robots going forward? Will there be a time where, you know, I'll be able to have my my home robot, you know, build me a new desk for that, you know, whatever. Yeah. So it's an interesting time right now. Right. Yeah.

We are in a situation where I think there's a lot of tailwinds for changing manufacturing. As a country, we develop a lot of previous manufacturing methods in late 19th century, early 20th century.

As we developed these technologies, but then over time, as the cost of labor went high in the United States, we started kind of moving these things into other countries. So a lot of manufacturing moved out of the United States. But now we're at the point that we're realizing that, oh, maybe that was not such a great idea because now we have dependency on

And these large centers of manufacturing that might not necessarily align with our values, right? You know, they might be a little bit more authoritarian or they might just not want to have the same interest as we do. So now we're thinking about, okay, we need to maybe be a little bit more self-sufficient.

The challenge is that the core technologies that are underlying these manufacturing techniques does not allow them to come back to the United States. We talked about a factory right now needs to be built for every part. So what does that mean? That means that that factory needs to be building a lot of the same part for the rest of the world before it's economically.

So it's very hard to replicate that same factory in a smaller version and still be competitive in the United States. The nature of the technology lends itself to centralization. So we're kind of finding the nature of the technology at this point. It's not just like, you know, with Gusto, we can bring manufacturing back in the United States. Maybe some of it we can, but for the most part, unless the technology doesn't change, I don't think we can be able to bring manufacturing back. But that being said, like I said, there's a lot of tailwind. People want to bring it back.

Because now it's becoming an existential threat. So I think because of these tailwinds, there's going to be a lot of changes that are going to happen in manufacturing. And automation is one of those biggest things that's going to change. In order for manufacturing to come back to the United States, it's not going to look like what we had in the 60s and 50s. It's going to be the new paradigm of manufacturing that requires less labor.

or maybe requires a more sophisticated labor with higher productivity than we had before. So automation, flexible manufacturing, easy to reconfigure shop floors are going to be big part of...

bringing manufacturing back to the United States. You know, you look at some of the towns in Midwest that used to be manufacturing towns. You go there, it goes to downtown. You know, the whole economy around that factory died because that factory, you know, the product it was making became obsolete and it just, it was not flexible enough to change. It died. So this will change. Now, the question is how fast? That's the question.

That's a harder one to predict because I think there's a lot of geopolitical parameters there. Interesting times, as you said. Yeah. If you enter a conflict with China, that will change very fast. Yeah. If you don't, maybe it's going to be a little bit more paced. But yes, it'll definitely change. Timeline is tougher to predict. Now, will we have these systems in our homes? I think that timeline is slightly higher.

longer. You know, there are people predicting at some point everybody's going to have a humanoid robots that's going to help them with things. I think the lower hanging fruit is getting robots to use AI and automation in factories because we already have figured out a lot of other stuff around these robots and we have a very good supply chain around them. But I do see a future where we're going to have humanoid robots in every home. Everybody will own one. Maybe a little bit longer. It's harder to predict anything beyond 10 years, but I would put that in more 20, 25 years out there.

Fair enough. For folks listening who might be aspiring entrepreneurs in the manufacturing or robotics automation space, or maybe studying and interested in robotics and automation, but not quite sure what path to take, what advice would you give either from sort of the technical side, or you obviously have an entrepreneurial background and spirit yourself here? So

I think the point you made earlier about the people being at the core of it and sometimes, you know, that being the harder thing for technologists is a salient one. But what would you give as kind of advice?

I think one thing that has changed since I came out of school is we need way more multidisciplinary people. You know, when I, you know, I came from a family that like valued education a lot. My mom was a teacher. My dad has his PhD. So it was very focused on like, go to school, get expert in one area. And that's, that's going to be your thing. I think that paradigm slightly is breaking now.

where we need people that can connect different components more. We're always going to need experts, but we also need more people that maybe they know mechanical engineering, but they also know software. They also know robotic. They also maybe know a little bit of material science because we have these kind of breakthrough technologies like AI, and we need people who can connect the dots between multiple disciplines and build something new.

Yeah. So that is becoming more and more important. So I would suggest for people who are going to school now, get exposure.

Maybe become an expert in one area, but get as much exposure as you can in other fields of science and engineering. That makes you a much more attractive candidate for the companies of next generation companies, right? So that would be one main area of, you know, I would say advice I have for people who want to be in the technical field. On the entrepreneurial side, I think, you know, it is becoming more and more important to build mission-driven companies. Hmm.

there used to be a time where there was a lot of kind of incremental improvement companies that you can build. You know, the system's already there, you just want to make some incremental improvement. Right, right. I think with AI, now the shifts are going to be more fundamental. There's going to be always incremental thing, but incremental things are just easier to build. So I don't know if companies are necessarily going to be well-equipped, startups well-equipped to do that. As a startup, you want to kind of go after something more fundamental. And

And for that, you need more kind of mission-driven companies. So find something that you're really excited about. You know, you become a big believer that that change needs to happen and go after that. And I think there are going to be less companies that are just going to optimize this and that because it's just easier to do those things for the companies themselves, for the larger companies themselves.

Sounds like sage advice to me. Ed Meier, for folks listening who want to learn more about Machina Labs, your latest innovations, what you're up to, perhaps more technical-oriented research papers and blogs and stuff like that, where can listeners go online to learn more? Yeah, obviously our website, machinalabs.ai, is a good place. But also we're very active on social media.

on LinkedIn. I think we put both even technical and commercial milestones on it very often. Twitter as well and Instagram. Yeah, so I think follow us on those channels. I actually like, you know, very openly, one of the main kind of philosophies we have at Machina is like build out in the public.

So we actually share even our technical challenges online and, you know, people respond to it and we are very open. So if you follow, I think for even for the technical folks, you will see a lot of interesting kind of content that we put out there that they can contribute.

Fantastic. Ed, thank you so much for taking the time to chat with us. I could binge your ear all day or ask you questions all day, I should say, and listen to you talk about it. It's fascinating, fascinating stuff. And as you said, interesting times we live in, and it's probably only going to evolve faster than it has been. So I look forward to keeping an eye on what Macintosh Labs is up to, and maybe we can talk again down the line. I would love that. Thanks, though. Thank you. Thank you.