Transcription
I'm at the BMW San Luis Potosi car manufacturing plant in Central Mexico. Here, some 700 robots are working around the clock to lift, bend, fold, spray, and in general, construct the next generation of cars.
But if these robots are so good at manufacturing, why does this place still need some 3,700 humans? What are the limits of automation?
The world's first automobiles were one-off art pieces crafted by a single engineer. However, by 1913, interchangeable parts and the moving assembly line made the car a mass-produced commodity. Thousands of human workers were given simple, highly specific tasks, conducted in sequence in order to produce the final vehicle.
While that was great for many of the workers, for the few who were now being constantly exposed to hot metal or toxic fumes, they ended up suffering near-constant workplace injury. A solution wouldn't exist until 1947, when a man named George Devol Jr. released his Speedy Weeny onto the streets of New York.
George realized that busy commuters wanted freshly cooked hot dogs, but unlike at a Bunnings sausage sizzle, he didn't have a team of Boy Scouts to do the cooking. So instead, he got a simple linear hydraulic actuator, mounted it inside a vending machine, and got it to push sausages from a fridge to a microwave to the consumer, all in 20 seconds flat.
With money made from his device, George added in some extra motors and a more powerful pusher to create Unimate, the world's first industrial robot. Unimate could move 200-kilogram loads, make repeated movements with sub-millimeter accuracy, and didn't need silly things like a breathable atmosphere or room temperature.
In 1961, General Motors purchased the first Unimate, using it to move hot metal castings and to weld car bodies. Importantly, it could be slotted right into their existing production line, replacing human workers on a task-by-task basis.
While GM bought theirs outright, other manufacturers rented theirs, effectively being paid like a human worker, although without the human risks of injury, death, and unionization. A key part of our operations is this, the mechanical arm.
I borrowed this guy from my lab at university, so it's not quite as big as the ones we'll see in the factory, but it can do the job just as well and has all the right parts. These bits are called joints. They're controlled by an electric motor and able to spin independently a full 360 degrees.
Linkages link those joints together. The original Unimate used an extendable hydraulic linkage, but that ended up being a pain to operate and maintain, and turns out you can do the same thing by just adding in more joints.
At the end of our kinematic chain is the end effector, which in this particular case happens to be a knife, but in practice, can be whatever you like. There are 30,000 parts that go into a car, made by suppliers using simple mechanized processes.
Those parts get packed up, sent to logistics hubs, and forwarded on to the actual BMW plant. Originally, BMW let their suppliers use whatever packaging they wanted, which then needed to be Tetrised its way into long-distance shipping containers.
However, in 2024, they introduced a new universal standard, which exactly tessellates into the crates. Arriving at the factory, they're unpacked and prepared for assembly. This facility was not built for humans.
As we've been going through, I've had to snake through this rabbit warren of tunnels as we move underneath the moving robots. The entire operation runs on a single production line. Three classes of vehicle, left and right-hand drive, auto and manual, and all the colors of the rainbow, are being produced one after the other.
They move first through the body shop, then painting, and finally, assembly. The biggest robots live in the body shop. Although robots do a lot of the actual heavy lifting and dangerous welding operations, we still need people to keep them fed.
Here, Gabriel is loading components that came in from storage into the robots. He's not just managing this machine; he's also got four of them across this part of the facility. The main body moves along using tracks.
Additional parts are held in place and welded together using arms with custom end effectors. This is the most complex set of robots in the facility. 16 of them welding in parallel in order to make together the main structure of the car, as well as the outer surface.
This incredible amount of robots ensures that the process happens fast to prevent the production line from being backed up behind it, as well as to ensure that any expansion caused by uneven heating is mostly mitigated.
We now have the backend, made of steel, and the front, made of aluminum. This section is where we merge them together. Since you can't weld different materials, instead, this component here is using a structural adhesive, ensuring that they have a tight bond.
Raw metal doesn't play nicely with the environment, and who wants a boring, steel-colored car anyway? Our next stop is the paint shop. Painting requires four layers applied one after the other.
Any contaminants in one layer will cause defects to magnify to those above it. In order to stop anything from getting into the facility, the cars are dusted with an ostrich feather duster. I'm wearing a full suit to ensure that my own body doesn't contaminate the environment.
I've got a hat, these things spray me with air, and my boots have a very sticky pad on the bottom which removes any contaminants and dust. It's a big process. This is the preliminary part of our paint process.
We apply heavy metals in a water bath to the surface of the car. This ensures that the paints will attach to them later on. These aren't robots, but rather simple machines, which can ensure regular operation as it moves through the baths.
This whole thing is about 200 meters long. Unlike primer, automotive paint requires several even layers, which you can't get by just dunking it into a bathtub. Here, we're giving the robots a massive airbrush and wrapping them in a protective plastic apron.
Sequential layers of color base coat one, color base coat two, and a clear coat are applied to the vehicle. These robotic arms are able to dexterously reach all the hard-to-get areas of the vehicle to ensure that the entire thing is colored.
Four robots with eight cameras each and their own special lighting system take 1,000 photographs of every single panel on the car to ensure that there are no scratches and it's been painted to the highest possible quality.
These robots are pretty complicated to program. Not only do we have the regular six degrees of freedom of an arm, but they're also mounted onto tracks and learn to move up and down to reach the entire vehicle.
At this point, we have a very pretty, if functionally useless, shell. A flyover takes our vehicles to final assembly, where trim is added, and they get a drivetrain.
So far, our robots have been doing great with the lifting, welding, and spraying. However, that starts to fall apart here on the assembly line, which is where the majority of the human workers live.
In this part, they're putting in seats, but in other places, they're doing other very manual operations, putting in wires and other forms of assembly. Robots really struggle with that.
The first problem is that our parts tend to be soft, bendy, chaotic objects, hard for robots to track. 3D camera systems do exist, and this is a professional-grade one I borrowed from my lab.
It uses a left and right eye to build up a stereoscopic view of its environment, just like a person would. Although the resulting image isn't great, with the objects jumping back and forth several millimeters between frames.
Humans can still see 3D even with one eye shut, by knowing the relative proportion of objects of known size. If they're smaller, then they're probably further away. Robots can do the same thing using AprilTags, patterns of known dimension, which have an added benefit of having nice, regular lines, allowing for us to also work out the orientation.
They're a bit like QR codes. In the factory, I saw several of these AprilTags in place on different objects, although typically, for scenarios involving vision, humans are just the better option.
Electric motors work best at high speed and low torque, which is actually the opposite of what you want out of a robot. Using an insane gearbox reducer with a 1,000 to one ratio will increase torque by 1,000 while reducing speed by the same amount.
That's great, right up until you hit something. While torque increases in direct proportion to the gear ratio, inertia becomes squared. That means if we hit something with a force of five newtons, 5 million newtons get reflected back into the robot.
Robots don't just bump into things. They annihilate those things and themselves in the process. Teleoperation is one good solution. The leader arm has the position and velocity of each of its three joints being recorded and sent to the follower, which strives to match those as precisely as possible.
This allows me to do some pretty fun movements. That follower also sends its information back to the leader, allowing me to feel a virtual force as it interacts with the environment.
By tuning some of our parameters, I'm able to work on things that are much larger and heavier than I'd normally be able to interact with. At the same time, if we use a really small, precise follower, then I can do operations on things that are really small, like surgery on a grape.
Often, we need humans and robots working directly together, in which case, we use collaborative robots, or cobots. To keep the human workers safe, we limit the maximum amount of torque that motors are allowed to exert, and also use relatively low gear ratios.
They counteract the worst effects of the squared inertia turn. They're programmed to exactly counteract the weight of the objects that we're trying to move, meaning you can push them on effectively weightlessly.
The way you program that is by changing from position control to torque control, and then back calculating all the resistances that we expect our object is going to face. We can also add in virtual guide rails or restrict it to certain planes of movement, further helping out the workers.
While that's great, now not only do they need to know what gets plugged in where, but also how to use, tune, and debug their robotic companions. Fortunately, BMW has massively invested in an onsite robotics training academy for just this purpose.
I had a go with using some of their robots, and yeah, things can get pretty difficult. This is one of the cobot stations. The crew here are on lunch break at the moment, which is why I can stand here.
We have some components which are done entirely by hand, fitting pieces into the engine, as well as using the cobot in order to increase forces and torque in order to bolt this part of the assembly together.
One of the cool things about having humans and robots working together is they need to have some form of communication between the two. In this station, they've got "Pac-Man" music in order to indicate what new components are coming and give feedback on how production has gone.
At the final stage of the assembly, it's time to attach the roundel. Now, this could almost certainly be done by a robot, but in a way, it's a final human stamp of approval.
From start to finish, it takes 48 hours to build a car, with a new one rolling off the line every two and a half minutes. Along the way, they've been interacting with ever-more complex machines, from mechanisms to robots and cobots.
Our 3,700 humans play support roles in logistics, with loading of non-standard parts. They oversee robotic operations, jumping in to fix mistakes. And of course, there's final assembly, with both cobot-supported tasks and those that are still complex and fiddly enough to need a dedicated human.
Maintenance engineers and programmers get the robots running. While site support, among other things, run a closed-loop recycling plant and a solar farm to ensure that everything else runs smoothly.
For hundreds of years, car manufacturing has been an orchestra of craftsmanship and precision, originally as one of human endeavors, then as mass-produced devices made by humans acting like automata, and today, with a mix of man and machine.
To get a car to actually drive itself off a production line, well, that's a whole new field of robotics, and at least until a future episode, I'll gladly do the honors.
This has been James Dingley from the Atomic Frontier. Keep looking up.