I'm not sure how many times this has to be restated.
It's car manufacturing. Everything that could be done by a purpose specific robot arm bolted down to the factory floor is already done by a purpose specific robot arm bolted down to the factory floor.
What remains unautomated, then?
The long tail of tasks that are too minor, too finicky, too open-ended or too reliant on manual dexterity to be offloaded onto traditional robots.
This is where this new generation of robotics comes in. This is the kind of task they're designed to do: "a task that's still done by a human in a high automation environment". Universal robots are angling for the tasks that are impossible or uneconomical to automate with traditional industrial robots.
> Everything that could be done by a purpose specific robot arm bolted down to the factory floor is already done by a purpose specific robot arm bolted down to the factory floor.
Hah! Hardly. I say this as someone whose first "real job" was in applying robotics research to automotive assembly - there are still a ton of assembly tasks that could be performed by a fixed-base robot arm, or a robot arm on a linear rail/fixed gantry. Wheeled mobile manipulators are only needed in a few cases, and humanoid form-factor is only "necessary" in very few cases (and I don't think the current crop of humanoids is particularly suited to these tasks).
In my opinion/experience, the impediments are that (1) the system integrators that are usually responsible for assembly-line robotics are too stupid to figure out how to apply robots to the problem, (2) the automakers themselves are often too short-sighted/stupid/unwilling to invest in increased automation (and particularly in building the in-house competency that they really need), (3) the hostile/exploitative relationship between (most) automakers and their main suppliers means that low-hanging improvements to parts/assemblies are a non-starter, and (4) the automaker C-suite (and investors) are too drawn to silver-bullet solutions (e.g. humanoids) than practical automation improvements.
"Could be in principle" and "could be in practice, under technical and economical considerations in play" are two very, very different beasts.
Everyone in the industry learned that the hard way.
At a certain point, the tasks that remain stop being "dexterity" problems and start being "AI" problems. That is: a robot could do the task - if you either spent big $$$ on redesigning the entire task around the robot's intellectual limitations (uneconomical), or if you had an incredibly advanced AI capable of problem solving driving that robot (impossible with 00s AI).
The "universal robot" bet is the "incredibly advanced AI capable of problem solving" bet. That in 2020s, AI is finally capable. The body only has to be "good enough to make most tasks possible".
> "Could be in principle" and "could be in practice, under technical and economical considerations in play" are two very, very different beasts.
> Everyone in the industry learned that the hard way.
The auto industry is notorious for making incredibly myopic choices to save money/make money in the near term versus long-term investments. The relationship between automakers and their suppliers/vendors is basically a century-plus of the automakers trying to (1) outsource anything they can for a quick buck, and (2) grind the supplier/vendor margins down to nothing. (This is part of why the newer Chinese automakers with much greater vertical integration are such a threat to the traditional automakers; vertical integration has a high up-front investment but the payoff in flexibility and speed is significant).
Vertical integration is superior if you can pull it off. Big fucking "if". There's a reason why automakers don't usually do it.
The name of the reason is: corporate rot. They don't have the organizational backbone that wouldn't let their "in-house manufacturing" rot away into inefficiency and waste.
Not that it has much to do with why automation fails to penetrate certain tasks. The reason why "long tail" tasks are often beyond automation is: piss poor ROI, calculated correctly.
You go out of your way to automate a certain process with traditional robotics, and it'll probably pay off in 15 years. The chassis this applies to is going to be in manufacturing for 10 years. At least half the systems work you've done there would have to be redone for the next chassis. Fun.
The bean counters counted their beans, and found out that using traditional robotics there is a losing game. Thus the search for better options. And the humans performing the tasks in the meanwhile.
> Not that it has much to do with why automation fails to penetrate certain tasks. The reason why "long tail" tasks are often beyond automation is: piss poor ROI, calculated correctly.
I actually don't think any of the big automakers have ever really, in-depth considered the ROI of "traditional" assembly automation (i.e. anything SoTA pre-2020), with experts in all parts of the process in same room. It's easy to assume that these companies must make careful measured decisions based on evidence, but in practice big decisions are made by small groups within the C-suite, often pretty divorced from the reality on the ground.
For example, many of the big asian automakers seem to have completely ignored the well-understood effects of their demographic crises (i.e. significantly aging population) on the future of their workforce (i.e. they are having trouble retaining and hiring new workers as the older generation retires) and this totally changes the economics of automation! Now they are all having to play catch-up, having realized that they must automate, at whatever the cost, because the issue is not "robots must be cheaper than human labor" it is "we might not be able to afford human labor at all".
This is over the last decade at one of the largest automakers in the world. Naturally there is significant variation between individual lines and plants; some are newer and more automated, some are older and much less automated. Are some cars being built on more automated lines? Yes. But a great many, probably the vast majority, are being built with fairly low assembly* automation.
* There is a significant split in automation between "body weld" stages and "assembly" stages. Body weld is very heavily automated basically everywhere (although there are some surprising exceptions in places), while assembly is much less automated.
Agreed, and hence I suggested an amazon warehouse tour (they offer one for their flagship robotics 'research' warehouse) to anyone, or a Tesla factory tour (might need to talk to someone, fairly manageable).
This reminds me of the quote, "the future is already here – it's just not very evenly distributed."
“One of the largest automakers in the world” makes me think of very low-tech companies like Ford or whatever. I can’t imagine this would bring much actual experience with this new generation of robotics.
Ford doesn't even make the top 5 - and however "low-tech" you think these companies are is the point, the overwhelming majority of new cars are being built by those "low-tech" automakers. The problem is not the limits of current technology (or even of the state of the art 10 years ago), it is the lack of vision and will within these companies to invest in using it.
> I can’t imagine this would bring much actual experience with this new generation of robotics.
Luckily for you, my job has always been within the robotics research side of the company, so I am very much aware of the strengths and weaknesses of the current technology.
This gives me vibes of "... but they're dinosaurs" that pervades Tesla-type discussions. I've watched several extensive discussions here with some Tesla fans breathlessly announce "really cool", "futuristic" "new thinking" functionality or features for their cars that "the dinosaurs just aren't even considering"... except to their initial skepticism, disbelief, actually, the dinosaurs often do have those things (examples including "adaptive blind spot", where triggering of those alerts is highly dependent on speed differentials, so if the vehicle is more rapidly approaching you, the alert fires earlier, dynamic traffic sign recognition - my car doesn't just recognize school zones, but can see when they are active if they have flashing lights, and, based on equipment installation, will actually count down to when my intersection light will turn green, or road mapping radar, where it actually scans the road surface ahead, avoiding potholes when safe, and adjusting ride height dynamically, not just 'press a button or choose a mode to raise or lower').
Indeed, Ford. Leaving aside the "old school" six axis robots that have been around for decades, Ford absolutely uses UR10s collaborating with humans to sand the entire car body in about 30 seconds, and to fit shock absorbers. They're also used at the engine plants. They also use the Symbio platform for transmission assembly, and fully autonomous forklift robots throughout their Tennessee plant.
My point isn't that they're dinosaurs, it's that they don't really develop robots. Some automakers do, rarely, but they mostly buy them from specialized robotics companies. It's not knowledge-gathering to buy a sander, nor even really to use one. Not on the same way it is to develop that sander. If they were all making their own robots and competing in that way, maybe my response would be different.
I don't need robotics experience, I need automaker experience. Their software is universally terrible.
Ford isn't a tech company, they don't even make their own robots. They buy them from someone else. What... positive experience do you have to assume that one of the lowest-tech industries in existence is somehow giving experience with some of the most advanced tech in the world?
> I don't need robotics experience, I need automaker experience. Their software is universally terrible.
If you had any auto industry experience, you would know that the people responsible for the design and build of the physical car and the people responsible for the user-facing software are very separate (in fact, the user-facing software might be entirely contracted out).
> What... positive experience do you have to assume that one of the lowest-tech industries in existence is somehow giving experience with some of the most advanced tech in the world?
You do realize how laughable this position is, commenting on an article about one of the largest automakers in the world buying out the remaining stake in the robotics development company that they already effectively owned. Do you really think that somewhow between owning BD and their partnership with GDM that no-one in the entire corporate structure of Hyundai is aware of the state of the art in robotics?
>commenting on an article about one of the largest automakers in the world buying out the remaining stake in the robotics development company
enhance:
>commenting on an article about one of the largest automakers in the world buying
ENHANCE:
>buying
It does not support your point to show that Hyundai is purchasing the company that actually built the robots.
If you want to make your point about how laughable it is that companies don't tend to be in the business of making highly advanced robots, you should probably not prove it with a company that essentially achieved world-largest-status before even finishing their straight-up purchase of this knowledge and tech.
> companies don't tend to be in the business of making highly advanced robots
I think you've gotten hung up on the idea that making robots is the essential part of the problem. I'm not going to go as far as much of the ML community and say that making hardware is secondary, but the software side is where most of the latest and greatest work is happening. Better hardware is not going to magically solve all the outstanding problems of manipulation. Better software might solve them entirely independent of hardware so long as the hardware is "good enough". I think there is a general sense that the field that robotics is approaching the equivalent on the mid 2000's with respect to computing architectures - there are still first-party RISC UNIX workstations on the market (e.g. Apple, IBM, Sun) but the incoming tide of commodity x86 platforms is clear for all to see. There are still some gains to be had by designing everything in-house, but the marginal gains versus off-the-shelf components or even full systems are steadily narrowing. There is every reason to believe that COTS hardware capabilities will continue to mature and become more commoditized.
Of course, the purchase here that actually mattered happened several years ago; this headline is just the final piece. BD and Hyundai have been working fairly closely on Altas applications to auto manufacturing for some time, never mind the additional research being done at BDAII/RAII.
You have to have the will to be a vertical company how many companies want too in the west? Not many do.. BYD, DJI, Huawei, maybe Apple a little there are few others but most are partial but the Chinese are the leaders in a willingness to do it all, note Ford Motor at one time did it all for a short time at their beginning.
I have visited factories for work and my experience is the same. There is so much stuff that could easily be automated but is not because it is too expansive for too little value to make a custom one off machine. The big high volume things will be automated but these machines will have 90% success rate and lot's of stuff that needs to be done by hand. You can search for factory tours on youtube to get an idea. Here are two videos, an Amazon warehouse and a Tesla factory. the big heavy stuff is automated but lot's of work is still done by hand.
https://www.youtube.com/shorts/-R6cBkza17khttps://www.youtube.com/watch?v=45slYC99uUg
Do you think humanoids would be a fit on assemblying the assembly line itself? To my limited knowledge a lot of setting up the factory is making sure your line works as expected with X 9s reliability. Here dexterous humanoids are this _universal_ virtual-to-physical interface and, akin to Auto Research, could run assembly line experiments autonomosuly?
automotive workers unions started around 1918 and became major political players in the 30s -- a fact that i'm sure is wholly unrelated to why there are so many un-automated tasks in that industry.
It’s not really a secret that most new auto factories serving the US tend to open in places where those unions are not active, like the South or Mexico
Pretty different tasks, environments, outcomes, metrics, goals, and other things in a warehouse vs. a factory… really have no clue how this is supposed to be relevant. Why not mention farms or libraries?
Libraries are typically either governed by a municipality's rules around employment and treatment of employees, or part of a school/etc where there are again additional guidelines about these things to be sensible and not leading people into unsafe behavior.
But if these tasks are too minor, too finicky, too open-ended or too reliant on manual dexterity for a purpose-built robot, how can a general purpose robot perform them better? If anything, they should be doing worse.
The only thing I can think of are tasks that are so rarely done, it's not economical to build a robot for. But I then I also don't see how another robot solves this problem.
These robots operate on completely different principles.
One can lift insane weights, has insane torque, and absurd precision, and can do the same movement millions of times with virtually no deviation. You program these with an exact movement plan, just like you would programm a CnC with a tool path. They are basically cnc machines.
The other one is a inacurate, unstable, dynamic system controlled by neural networks and heuristics. It has massive deviation over each run, but that means that the programming must be able to account for it. Which makes it suitable to operate on problems that are messy, unrepeatable and human-shaped.
Pretty much. It's a total paradigm shift from how industrial robots normally work. A pre-planned motion executed carefully and precisely vs open-ended "do this thing" powered by a very large bag of opaque neural network heuristics.
A robot that has to be carefully adapted and set up for the task vs a robot that you can point at a task and have it figure out how to do it. A robot that doesn't deviate vs a robot that absorbs all kinds of deviations.
It's a bet that The Bitter Lesson will win over Moravec's Paradox, in the end.
It's also worth noting that when e.g. inputs to a stage might have unpredictable defects or alignment, a robot arm utilizing neural networks for planning and analysis might still be the best way to handle that - without the extra degree of freedom of movement-relative-to-floor, planning can be done more rapidly, and movement can be executed more aggressively and quickly.
If I were Hyundai, I'd be looking at this as buying a significant amount of vision, dynamics, and integration systems expertise, not necessarily the dream of self-motive walking systems.
> movement can be executed more aggressively and quickly
That's exactly the reason why it's usually a bad idea to run a classical robot on a neural controller. If atlas bumps into something you get a small bump and maybe a broken atlas. Your average industrial robotic arm will happily yeet whatever it bumps into across the room.
A) the idea is that these robots do have dexterity capabilities a lot closer to human hands
B) there’s a long tail of individual tasks it’s uneconomical to build purpose-built robots for each individual task. But it’s economical to have 1 robot that can do all of them.
Do they? A human can both chuck kilograms of stuff across a room or kick in a door, but then pick up a single hair off the ground, or feel and manipulate (things even lighter than) a literal feather.
Robots can certainly do things more repeatably, if not more precisely.
Robot arms can have reproducible movement with error bars measured in double-digit microns. Where they lack is in the force feedback, and even then they have had huge strides.
The human arms and hands are very versatile, and imitating them is a good choice for a universal robot, though 3 or 4 arms are definitely better than 2, and the hardest to imitate are the sensors, not the actuators.
But the rest of the human body is not useful in a factory environment, so the arms could be mounted on a mobile base that does not have any resemblance to a human.
> the rest of the human body is not useful in a factory environment
I dunno, a legged design is pretty useful for navigating complex environments. The arms and legs have to attach to something so you've still got some sort of torso. About the only thing you can easily do away with is the head I think.
But certainly a bipedal design seems unnecessarily complicated unless you need it to climb ladders inside narrow tubes or something similarly specific. I feel like a quadruped with 4+ arms mounted on top and many-jointed fingers might be ideal (both in terms of utility and also creepiness).
And C) they don't always have to be at parity with human hands to be good enough because humans are flat out expensive. Humans need health accommodations, have sick days, vacation, and make mistakes too. The bar is much lower and the incentives are much higher than many people probably think.
Because it is general purpose. We did not have the ability to create a single robot form which could do all of these minor, finicky, and opened ended tasks. Now that seems within reach. The nice property of humanoid robots is that the world is already made for human form, and so if you're trying to replace people naturally this is what you'd want.
Well, humans obviously do those jobs, so a clearly a general purpose robot (in this case, a biorobot) has been found to do the job better. Don't overthink it.
It's not a "general purpose" robot, it is a "human replacement" robot, with similar skills and shortcomings to a human. Humans are not general purpose.
All you need to do is look at a recent video of car manufacturing process, and watch what the humans are doing.
>> how can a general purpose robot perform them better
Better than what? It seems that as long as they perform the tasks "better" (cheaper / faster / lower-error) than the humans that are currently performing them, that is an improvement for the factory owner.
>It's car manufacturing. Everything that could be done by a purpose specific robot arm bolted down to the factory floor is already done by a purpose specific robot arm bolted down to the factory floor.
>What remains unautomated, then?
Stuff that can be done by purpose specific robot arms on wheeled platforms, which is very difficult, but will be much more feasible than a humanoid robot doing anything.
No one ever considers the physics of this situation.
You can't put a robot arm on a wheeled platform without making the platform very heavy otherwise the whole thing will topple over. This gives your entire assembly a mandatory floor space requirement that may be quite large, and severely constrains how much reach you have (see the Handle robot from BD itself).
A platform like the Segway with a self-balancing system can help with this, but since it doesn't have legs it has very little ability to keep the top of the platform steady - all it can do is accelerate around to try and accommodate wobbles, whereas a bipedal robot can simply shuffle it's legs around and keep the top of it's body stable.
It is difficult to build a control system which can do this, but once it's done it's done.
VRML may not have taken off in the 90's, but I can, today, go to https://play-cs.com/ and have a have a jaunt in a virtual world. There are tons of other virtual worlds to explore. Second Life is still around.
It took 3 decades to get to where we are today witht that. If it takes until 2066 for humanoid robots to arrive, I won't be around to see them, but it also won't be a fantasy by then. I'm hoping it won't take quite that long, but we'll see.
You might have forgotten some of the most-funded ventures in VRML. It wasn't just "exploring virtual worlds" or whatever, they wanted VR to be the way you interact with the whole of computing. Checking emails, web browsing, chat, spreadsheets. Second Life at al are cool, but a fraction of the vision of VRML as it was presented in the '90s.
Apple quite literally reheated the idea of VRML with Vision Pro, and we all saw how that went.
My comparison to VRML isn't about technical feasibility; sure, on a long enough timeline, anything is possible. The problem is market fit relative to technical complexity.
On a timeline long enough that humanoids become practical, it is almost certain that we will come up with other technologies along the way - technologies that will do the envisioned jobs of humanoids just as well, if not better than humanoids, and at substantially lower cost.
Ah, I see what you're saying. The three features of a humanoid robot are two legs, two arms, and a head. Without introducing technology we haven't even thought of yet, the head is a bunch of complexity that is unecessary if designed from scratch. One 360 degree camera means no eyes on a head that needs to rotate and move with 3 degrees of freedom. As long as we're changing vision systems, cameras and lights embedded on the "hands" would make it easier to do certain kinds of jobs. The number of hands, the number of fingers on the hand are up for optimizing, along with the number of arms. Three might be optimal vs two. Maybe six? And a shelf in the middle of the chest. The other thing that defines humanoid is bipedal motion, a.k.a. having two legs. From liking around, the other options are three or four legs for six and then also wheels a.k.a. rollerblades.
It's impossible to predict the reality of product market fit for the future. It sucks doing the dishes, even with the washing machine doing the work, so how about the cabinets themselves are the washing machine so you just put dishes back where they came from and they're cleaned by the cabinets. No humanoid robot maid needed.
If we look at cost as the optimization driver, we really have to dig into the nitty-gritty of robotics. The big one is the amount of energy that we exert just standing there versus a humanoid robot just standing there is ridiculous. So then a centar looking thing. Or maybe one point of contact with the ground which is a ball and the machine runs balancing and hopping algorithms. Pogo stick traversal might be most efficient, given enough compute to make it feasible.
What you need then is a better arm (or even just hand), not a human.
Or a new take on car design with automated production in mind regarding all the wiring and what not (easier said than done, I'm sure many have tried and failed, but eventually someone will succeed).
If the task is too finicky, too open-ended or too reliant on manual dexterity for a purpose-built robot arm to handle, why would a walking, humanoid generaic-use robot do any better?
Naw, the real answer is that factories have been built around human labor - they weren't built to be forward-compatible with purpose-built robots, so during the transitional period where we build these purpose-built robots, you need humanoid robots to fill in the parts where the factories were geared for human labor.
So we want to replace a lot of cheap human labour with expensive robots?
Even if I think this has legs, where do the cheap humans go to work after? Where/what are the remaining jobs for all of this displaced labour in both white collar and blue collar worlds? It basically screams UBI. And in a UBI world, the economy looks pretty different and humanoid esque robots start to look either very altruistic or very dystopian depending on how hard the oligarchs don't want things to change.
Like, what's the end game for humans in this path we're embarking?
>Like, what's the end game for humans in this path we're embarking?
The market becomes more efficient as fewer human beings are needed to create value and move capital. A lot of them are going to die, surplus to requirements. A lot more will be stuck in lives of grinding and meager poverty, probably doing gig work acting as "flesh AI" for less expensive robots or "blood boys" for the rich. But the rich will be very rich indeed.
It won't just be end-stage capitalism killing people, either. The collapse of the knowledge economy, scientific and research institutions and the mass adoption of AI to fill the gap will kill tons of people too, as will the return of diseases like polio and smallpox, and mass starvation as climate change destroys global agriculture, and the normalization of christofascism.
We're almost certainly not getting UBI, at least not in the US. It would help too many black people and immigrants, half the country would secede. We might get something called UBI but only so long as it isn't universal, and has tons of racially biased and religiously motivated means testing and plenty of carve-outs that keep that money flowing to the top, and out of the hand of the "useless eaters."
We were alive in an interesting time where a clerical caste of society was needed to get the most out of capital. Once this is no longer the case it could be that we return to something feudal.
'Everyone over 30 knows this' is a prior assumption (it is not necessarily correct; and nothing is said about shame).
The comic strip is saying if above is true, then people still have to learn at some point so on average it would be around 10k people per day.
I think the math is this:
For people born in a given year: 4000000/365/30 = 365 people per day
but you have 30 sets of those people (those born this year, those born last year, those born two years ago, etc.) So 365 * 30 = 10950. 10k is easier to say for viral purposes.
It's car manufacturing. Everything that could be done by a purpose specific robot arm bolted down to the factory floor is already done by a purpose specific robot arm bolted down to the factory floor.
What remains unautomated, then?
The long tail of tasks that are too minor, too finicky, too open-ended or too reliant on manual dexterity to be offloaded onto traditional robots.
This is where this new generation of robotics comes in. This is the kind of task they're designed to do: "a task that's still done by a human in a high automation environment". Universal robots are angling for the tasks that are impossible or uneconomical to automate with traditional industrial robots.