Former SpaceX engineer invents a “Robotic Blacksmith Army” | Hard Reset

- This could be the factory of the future. Actually, this could be several factories of the future. - We have automotive parts manufactured in the morning, and then aerospace parts manufactured in the afternoon.

- This is called 'roboforming. ' It's a new way of shaping metal by using robots and artificial intelligence. Why do this?

- We're still manufacturing things the same way we were making them a hundred years ago. To make the next leap and keep up with the digital world, we need to accelerate the physical world, and that requires for manufacturing to go to the next level. - Right now, most metal manufacturing is sheet metal forming, basically, stamping metal into a shape.

- Sheet metal industry right now is a 250-billion industry. Most metal parts you see day-to-day are sheet metal parts. You're driving in a freeway in your car, you're basically sitting in a sea of sheet metal.

Or you're sitting in an airplane, it's basically a sheet metal aluminum can. Even electronics, all sheet metal, sheet metal, sheet metal. - Sheet metal forming is fast and cheap, but there's a downside.

- You have to build a factory to manufacture the parts, and most factories are very specifically built for the design you're trying to manufacture. It's not easily changed; it's very expensive. - The uniformity of our world is an artifact of the assembly line and the way we manufacture things.

But this technology could create a world where custom manufacturing could be just as affordable and efficient. - I'm working on a product that allows you to design a new car and start manufacturing it today if you want. - Our scanning technology allows us to look at the part and form those without the need for stamping, the same agile way that craftsman did it.

- This is some serious metal. Actually, it's heavy metal. No, I mean it's really heavy.

It's, it's like a 1/4"-thick steel plate. Stick around to meet the robots that will build the future. This is "Hard Reset," a series about rebuilding our world from scratch.

We came to Chatsworth, which sounds like a fancy butler, but is in fact a city in the greater Los Angeles area. Why here? Because that's where Machina Labs is, a company that is trying to change the way we make things.

This is Ed. He founded Machina Labs because, basically, he wanted a robot blacksmith. I mean, who doesn't?

- Manufacturing used to be arts and crafts. So back in the day, they used to 'bomb form,' and bomb forming is a process where a craftsman sits behind an automated hammer and slowly deforming a sheet of metal into a shape, kind of almost conquering physics, right? They can incrementally deform with a limited tool set that they have- a bunch of hammers, maybe few handheld tools- but how can they use the set of tools that they have in a creative way to get to a final part?

So what we are trying to do now is can we do what those craftsmen did back then, but using robotics and our system, and form those without the need for stamping, the same agile way that craftsman did it in the past? Yeah, so- Can you tell Cell 1 folks, when they open the clamps, they don't drop it? - That's okay.

This pause gives us a moment to appreciate Ed's magnificent beard. It's just a masterpiece of beardsmanship. Yeah.

Anyway, let's move on. - I come from the additive world. I was in charge of a team that was building a very large envelope metal 3D printer for aerospace applications.

- Ed is being modest. He was helping build a rocket for a little startup called SpaceX. - So when I was at SpaceX, we had this very large, complicated problem: We were building large tanks.

Think of a tank for a rocket that's 22-feet-tall and 10 feet in diameter. And when we were working on Falcon, and Falcon 9 specifically, the diameter of Falcon 9 tank could never increase, because there was so much tooling went into that shaft floor that was specifically built for that diameter. So if you want to add more fuel to that vehicle, you had to just make it taller.

Well, you can never make it larger in diameter; you couldn't make the rocket fatter. So even a company like SpaceX is this challenged in terms of like once you lock in into that design, you kinda have to stick with it. In order for us to build process, we have to take that concept of craftsmanship and see if we can scale it.

And once you're looking at the detail, it's actually very complicated. 'Cause what happens in the mind of a craftsman or a sheet shaper is pretty complicated. It has dexterity of the robot, so we needed robotics to replicate the dexterity of their hands, but then we also need to replicate what's happening in their mind.

- Instead of just stamping out a shape onto these metal sheets, these robots push on it to create whatever shape you want. Actually, one robot pushes and the other supports. The support part is super important, because if you just pushed from one side, you'd get a big stretched out mess.

Using two effectors allows you to control exactly how and where the metal bends. This is Michael. No beard.

- It's a bit of a dance between two very synchronized robot fingers that are slightly offset with each other, and they're kind of gently pinching and rolling metal and extruding it. And when we actually form parts on the cell, we're not just moving the metal out in space, we're actually thinning it out to achieve the angles that it needs to achieve to form a part. - This beautifully orchestrated dance allows the robots to slowly nudge the metal into whatever shape you want.

You might think this is pretty straightforward, but people have been trying to do this for a long time. Well, not quite that long. - This method of sheet forming, you know, people have started looking into it since the invention of CNC.

Have been doing research in this area for the past, I would say, actively, maybe past 20 years, but loosely, the past 30, 40 years. - A blacksmith doesn't just hammer on metal. They are constantly observing how each blow has changed the object, and adjusting it, changing the way they hold it or strike it next time.

This is a feedback loop that requires tremendous skill, patience, and intelligence. And until recently, robots weren't really intelligent enough to both manipulate the metal and observe and adjust to how it has been manipulated. - From a hardware standpoint, what we're doing has been possible for a long time, but from a software standpoint it was very, very hard.

So our access to GPUs, our access to neural network type models were incredibly helpful for us in sort of being the brain of the craftsmen. And we will even touch material with our robots, sense what's happening, and then feed that back into models that can essentially do what a craftsman is doing. It's like, "Okay, this alloy is feeling a little bit rigid today.

We're gonna have to compensate in this direction. " - The robots here are constantly sensing and compensating for super complex forces, which has only been made possible by machine learning. In fact, the intelligence that makes this happen is the real star.

These robots are actually just industry-standard robots. - What we do here uses a lot of off-the-shelf and commodity hardware. In fact, one of the really successful cell systems that we have in the building, I'm not gonna point out which one, but we bought it on eBay, actually, the robot.

- I was told you bought these on eBay. - I didn't buy 'em on eBay. - You didn't sound defensive at all when you answered that, by the way.

- Well, no, I mean I could have. I didn't buy 'em. The person who was selling these also sells stuff on eBay.

I think that's where probably the confusion came from. - Someone was just selling one of these on eBay and we're like, "Okay, like, let's save some money. Let's try it out.

" One of the joints has a bit of a like a lunk in it and it's a little like dangerous to use, but it works, right? The darker orange ones are really old. I think they started out roughly this color.

- But yeah, we didn't buy 'em on eBay. - I got the sense that the guys from KUKA see what you're doing with their arms, and they feel like you guys are sort of like well beyond recommended use parameters. - These robots are not created for their intended use, but yes, definitely.

It's not an application where they're like, "Oh, you can use robots for sheet forming," you know? And I remember in the first early days, when I started talking about that's what I wanted to use these robots for, they're like, "Not doable. Don't do it.

It doesn't make sense. " - So an off-the-shelf robot arm isn't quite everything you need to do this. You also need an end effector that is strong enough to survive the tremendous amount of pressure being put on the sheet metal.

So you make all these in-house? - Yes. All of these are made in-house.

So we are, to some extent, we are vertically integrated- but allows 'em to basically iterate on designs really fast. The substrates, usually it's out of very hard material that doesn't deform under pressure. So think of like carbide and tungsten.

- Wow, that's heavy. - They're a pretty heavy material. - Wow.

That is not light. One thing I would love for you to show me, actually, on the arm, is sort of like where does the standard KUKA arm end and your hardware begin? - So you see in this robotic arm, anything after the orange is basically ours.

Now, there are a little bit of modification we might put in terms of sensors that added to the rest of the robots, to capture some of the things we need to capture. Obviously, we also manufacture the full frame. So everything on the frame, hydraulic clamps, the cooling, everything is also built by us and designed by us.

- The sensors on these arms make them sensitive to the amount of force they're exerting and where they are in space. They can even scan the geometry of the object they're shaping to make sure it's coming out correctly. I'm surprised at how like gentle they seem, although I know there's tremendous force being applied here, but it's sort of, it's almost hypnotic to watch them.

- Yeah. I think right now, we're applying 2,000 newtons on one of the sides. So 2,000 newtons is around, basically, like 500 pounds.

And then on the other side, you have another 2,000, 1,500 newtons. - A newton is equivalent to about 102 grams of weight in your hand in Earth's gravity, which is equivalent to about 6. 58 Fig Newtons.

So 2,000 newtons of force is about 13,161 Fig Newtons stacked on top of each other. I am hungry. When is lunch?

Anyway, so now that these robots have the intelligence they need, what will they build? The world runs on heavy metal. Oh sheet, I mean sheet metal.

It runs on sheet metal. Actually, heavy metal is a contaminant in groundwater, you don't want that. Look at the appliances around you.

Or look at cars, or airplanes, sinks, bathtubs, ducts, facades, roofing. We use it to make so much stuff. We've been doing it for a long time, and we're pretty good at it.

- Manufacturing technologies that are used widely, most of them haven't changed in the past, I would say 60-70 years. The way we manufactured Model T at Ford, the first mass-produced car, is the same way we're manufacturing Model S doors. Make it very giant, archaic, press that just does this with a lot of force.

Put a sheet in between it and just stamp it. - Each of the parts that get stamped out requires a huge machine, which requires a huge investment and a long time to set up. - When we were talking with some of the folks in automotive, a plant that manufactures a vehicle, we're looking at a $150 million investment in just equipment.

These are plants that are like a half-a-million square feet. And then you need to also put in stamping presses that are like three-stories-tall, and then three-stories-down in the ground. - A hundred and fifteen million dollars seems like a lot of money just to keep Eminem busy during the day.

I don't even want to try and convert that to what it would cost in terms of Mom's spaghettis. Still hungry, guys. Oh my God, when is lunch?

- You kind of need to be a multinational corporation to make something out of sheet metal. That's boring to me. I don't want the design power for the things that I use to sit in the hands of probably some of the organizations that might be least ready to design the next thing.

- Ford famously said, "Everybody can have any car, as long as it's black, because I need to change my factory if you want something custom, and I will never be able to do that. " - Manufacturing on an assembly line locks you into a specific way of doing things. Making adjustments like changing colors means changing or stopping the assembly line, which is expensive.

The uniformity of modern manufacturing is what has made it fast and affordable. - If you look at World War II, what enabled the United States to win that wasn't necessarily superior technology, but much faster rate of production. We could make tanks faster than they could destroy it.

- But with technology like this, we could see fast, affordable manufacturing that was also customizable and flexible, even on something that rolled off of an assembly line. - One of the things we're working on now is the Anvil project: It's our platform for custom car body designs. So what you're seeing is like a very early fit up of this hood that we formed.

And you can see it's still very raw, but it is our first attempt at actually getting something on the truck. - The team here is building a totally custom vehicle, replacing all the visible panels of this truck as a way to see how their manufacturing process can integrate with existing products. - We're not redesigning the car from the bottom up.

We are doing the thing that most people that are interested in custom cars wanna do, which is, "I wanna make the outside look badass. " So we're going from a truck that you would see on the road to a truck that you would see in like a sci-fi movie, basically. - So if you're making a new model of car, you could send the part specifications to a roboforming factory and they could start making parts right away, instead of taking years and hundreds of millions of dollars to set up a new factory.

And if you want those parts to be customized or changed, that's as easy as loading in a new file with the custom shape. - By using this type of robotic technology, we want the cost of this custom car to essentially approach the cost of a commercial car. - This project is also a way for them to let the machines practice making parts like this- and I mean that literally.

Because these manufacturing cells are powered by AI, they learn in a way that is similar to us, trying, failing, and trying again. And once they've practiced it a few times, they begin to make it perfectly. - Whenever I interview candidates to join our team who come from like traditional aerospace companies that are more research-focused, they are appalled that we don't like simulate stuff before we do it.

- Custom parts have more uses than cool hot rods. This is also important for repairing things like exotic airplanes or equipment that they don't make parts for anymore. If I brought you in a part where I'm like, "I need this aircraft wing.

I can't get it made anywhere," can you scan that and then kind of create a mesh that then can be duplicated? And what does that process look like? - That's exactly one of the things we're doing for the customers right now.

Sometimes there's a part that comes in that they don't even have a drawing for, we have to scan the part and rebuild the CAD, and it's part of our stack. You can scan it, it automatically generates the CAD, and we use that to basically pass it through our path-planning software to figure out what kind of a path will generate that part. We lost a lot of these manufacturing technologies.

We don't have those manufacturing plans. We went more toward bespoke planes that are very expensive, and million-dollar fighter jet that it's very advanced, it can do a lot of things, but once it gets damaged, very hard to fix it; it'll take years. So what we are working with a lot of times with DOD is can we bring this industrial capability back so that we can maintain, like, we can fix a plane in two days, three days, as opposed to four years.

- At this stage, there are still limitations to roboforming. It's not as accurate as it needs to be for every job, but it's getting better all the time. It's also getting faster.

Beyond just being quick to adapt to new forms, roboforming can also work with materials like titanium and other alloys that were never compatible with sheet metal forming before. That means it'll open up manufacturing capabilities we've just never had. I mean, if you ever wanted to drive above the speed of sound, you kinda need a titanium truck.

So I mean, really, for science, I think it's a necessary step you guys need to take. - I think so. I think we need to design some sort of demonstration of titanium and do something goofy with it.

- Yeah. I think it's only, it's only fair. - Yeah, I agree.

- Yeah. - Right now, we work with a lot of big marquee customers, but ultimately, I think what our team is excited about is what we call 'democratization of manufacturing. ' Get this in the hands of the people who couldn't make these parts possible without joining, like, somebody like a SpaceX or Hondas of the world.

Can we enable those guys? Because those are the guys who's gonna have the next generation of big ideas that, you know, that we want to look into and make, hopefully make. Those tools are gonna be very crucial and will change the landscape.

- So what will it mean when our factories are as capable and adaptable as a blacksmith? So, picture a scenario where we rebuilt our manufacturing capabilities from scratch. That world would include a lot of different ways of making things, 3D printing, machining, injection molding, casting, you name it.

But sheet metal forming will always be a massive part of that world. So, how would it all work? - You know, I imagine a world where, in the future, you as a designer, you can go onto a portal, upload your design, get guided through how you can manufacture it in the most efficient way, hit submit and say, "Okay, I want 20 of these in 20 days in Los Angeles, California.

" And the right facility, that doesn't look like traditional factories, it more look like data centers that has bunch of these robotic systems that can be programmed to do multiple operations. I think that would be the ultimate vision where we, I think manufacturing needs to go. - But manufacturing this way is about more than just speed and cost, it's also about proximity.

Roboforming factories are smaller and more adaptable, and that could change where we build them. - If you go to Midwest, there are areas where we had large manufacturing plants, manufacturing certain type of the car. Once that car was obsolete, that whole economy around that whole town died, 'cause that factory could not easily be retooled.

It was cheaper to just let it go and go start somewhere else from scratch. But once we move to this new paradigm where you can configure the factory to do different things, you can distribute it for one thing, it can be very closer to the consumers because it can be very adaptable. You're gonna have much more stable communities.

They're gonna manufacture the things they need very close to where they live, and they can constantly be adaptable to changes in needs in those communities. - When it doesn't take hundreds of millions of dollars to set up manufacturing, smaller companies will be able to make products like cars or appliances. It will mean more ideas, more creativity, and more personality is infused into the world we live in.

- The reason I care about manufacturing is at the end of the day, manufacturing is still an art. You know, it's a form of self-expression. You are conquering the physics of the world to create something that truly expresses what you want and what you are about.