We'll know this works when it starts replacing Amazon pickers in quantity. Amazon has been trying to automate that for years, with many demos and contests. So far, nothing can quickly and reliably take random products out of one bin and put them in another. Amazon's robotic systems move larger containers and shelves of bins around, but do not yet pick individual items.
You raise a great point. And the Amazon picking staff are onshore in wealthy countries. I guess the minimum wage paid by Amazon is around 15 USD per hour.
I wonder: Is the task of automating this work primaryly difficult in vision or dexterity (motion)? Or maybe they are equally difficult for different reasons.
Probably both vision and dexterity, and the first mistake we make as roboticists/engineers might be to distinguish the two like they're separate problems to solve or that a solution exists where the two live a separate life.
Agreed. The solution will likely be some vision foundation model that directly sends controls to the robot ("move here, grab, move there"), trained by Amazon with RL to integrate collision avoidance, object detection, grasping point detection, grasp verification etc.
If we're talking about picking objects at random from one bin and putting it in another, I don't need my eyes to do that. Proprioception (shape and location) and touch (texture) are enough to do that.
There's a lot more money being thrown at this than in previous years. Seems to be growing beyond corporate R&D labs and university research towards startups trying to productize it.
>A humanoid robot takes roughly 5,000 steps per hour. Each step sends a shock of 2–3× body weight through the leg actuators—forces that would be fine occasionally, but become destructive when repeated thousands of times without pause.
As someone who comes from the world of running and knee problems, I feel this misses the issue. Normal walking should not produce these kinds of shocks unless your gait is really jumpy or otherwise screwed up. You only start to see these forces when running and that's where technique becomes important even for humans if you want to prevent damage to your joints over long distances. But at least for walking I suppose that a fully articulated humanoid with all the degrees of freedom of human gait should be mostly a control problem, not a mechanical engineering one.
The force an impulse generates on a contact depends on the speed of deceleration. It's just F=m*a
Slow deceleration leads to low forces. If you have a contact event with a hard substance, like a rigid metal for accurate kinematics, the deceleration to zero upon a contact has to happen instantly. Meaning the deceleration is incredibly high, resulting in extremely high forces for a few milliseconds.
Human bodies are made out of a flexible and impact resistant material: water. When a contact event happens, your body deforms, which means that the deceleration happens over a longer time frame with less force. Not just that, your muscles also have a certain amount of flexibility in them and basically zero internal inertia. All the inertia is in the limb as a whole, whereas for a robot there is a spinning motor and gearbox that needs to slow down as well.
You could solve this as a control problem by adding series elastic actuators, which means you need to change your mechanical design.
The human body goes further than that too, when you're out jogging - as your foot approaches the ground for a stride, you slow the velocity of your foot downwards towards the ground so there's less of a sudden deceleration.
Imagine when you throw a tennis ball high in the sky, and then you catch it on your racket without bouncing by matching it's velocity, your feet do the same thing with the ground on a smaller scale.
As others say, not necessarily. The breakeven point for jobs like Amazon may be quite low (or high? I mean simply "not yet there").
I'd say that we'll know it works when robots with those hands start turning out on the Russo-Ukrainian frontline en masse, because it is there where the lack of manpower has the most pressing and brutal consequences, and cannot be mitigated by usual peacetime incentives (e.g. better benefits).
That frontline has already sucked in all the automatization innovations of the last decade, as long as they proved themselves in combat.
> We'll know this works when it starts replacing Amazon pickers in quantity.
That doesn’t follow. There are plenty of tasks that can be fully and reliably automated but aren’t, for the simple reason that human labor is dirt cheap compared to advanced robotics.
I disagreed, then re-read your post, then re-read the OP, and now I've come full circle to apologize; I think you make a fair point.
I work at a biotech. We spent who knows how much time and money trying to develop a 'lab technician bot' to automate one of our critical assays. Turns out, a 6-figure machine still isn't as economical as my coworker Y, one of the veteran lab-technicians. Sure she takes the occasional sick day but even at our volume (and we do industrial-level, multiple clients batched into a single assay pass) it won't be economical to replace her for a very long time (if we even reach that scale).
Absolutely. I worked at a gene sequencing company and I led the software side of making a robotic product[0] to automate the 20-30 minutes of sample preparation time. It's great for lots of uses, but for anything outside the exact thing it automates, it doesn't cover it. For that you need an expert human.
But that does follow. The economics working is not some outside factor. If the robot “could do the task” but would cost more than paying a human to do the same task then the robot “does not work”. It is frequently because the robot would be too slow, or not reliable enough, or could only handle certain types of items. But ultimately all of these boil down to cost.
We have seen lab demoes of robotic manipulation for decades. The reason why they stay in the lab (when they do) and don’t become ubiquitous is because they are not good enough. In other words they don’t work. The economics and “does it work” is not two separate concerns but one and the same.
It's a continuum, not binary. The same robot that doesn't financially "work" for replacing a manual scavenger sorting garbage in an African slum might be quite cost-effective sorting recycling in Switzerland, and would likely have a niche regardless of price if used to (say) sort biohazardous or radioactive materials. And there are already millions of robots out there assembling cars etc.
What is the point of humanoid “general” robots then?
We already have pretty reliable ways to make and train humans.
Humans are cheaper and better than robots.
I could imagine robots for some specialised tasks where you don’t want to use a human for eg security reasons, but you don’t need general purpose robots for that
technology gets cheaper over time. If they were always going to cost the amount they do now, you might have a point. But they'll eventually get much cheaper.
In natural ecosystems, nobody beats the apex predator directly, and nobody beats the hyperspecialized niche critter at their own game. The new species has some advantage that’s different than what is there.
If a humanoid robot is slower dumber human that is expensive, requires power, can’t get wet, falls over, and doesn’t understand stairs. Is not sleeping and being radiation tolerant enough of an advantage to be worth it?
The nature comparison doesn't work on a fundamental level because you're only getting a fraction of the human's power based on how much they're happy to sell.
They already are, the problem with humanoid robots is that people think that adding legs to the robot will somehow fundamentally make it more intelligent.
People see a robot arm attached to a stationary platform and understand it requires integration work to perform a single task.
But when those same people see a humanoid robot, they think they can just talk to it like a real human and it will do what you told it to do.
They don't think about the fact that the humanoid robot has to be programmed exactly the same way the stationary robot arm has to be programmed and that programming the legs in addition to the arms is a much more challenging problem.
Robots can be optimized for tasks and if they are, their benefits are greater. When cars replaced the horse, it was because they didn’t poop, and because a car designed only for transport would not suddenly have a heart attack and stop working.
A friend who works at Amazon made the same point: "We don't really need robots in the FCs urgently [other than the Kivas], because it turns out you can just pay people $17/hour"
I was thinking this week that AI token costs are probably going to get so expensive soon that bright spark CEOs are going to realise “why am I paying for such expensive coding agents when I can pay people from the third world to code!?!” and announce outsourcing like it’s some kind of stunning and innovative revelation.
C-suite has been saying this for 30+ years. They never tire of it. Ask yourself: At this point in time, why aren't all programmers working from low cost jurisdictions?
I think you didn’t grok the hidden punchline - this is the stage after they’ve replaced all their third world coders with AI agents, until one day a C-suiter has the revelation that humans are cheaper and better, and the company then starts toting its humanistic credentials all over LinkedIn.
Mechanical picking has been too slow. It's not a problem with the robot mechanics. Here's 300 picks/minute from 2012.[1] The parts are all the same, so the vision problem is simple.
But picking arbitrary objects from fulfillment bins is still running at a few picks per minute.[2] As the speed picks up, humans become less necessary.
That's the point of the test condition. When running a robot becomes more economical than paying full-scale humans $17/h, something important about robot abilities will have changed.
I dunno, I worked in an Amazon Warehouse for a year part-time (and a couple of weeks full-time when in-between jobs) --- on one occasion, I pulled up to a bin full of non-descript cardboard boxes near where a group of trainees were working their way through, grabbed one box, spun it around for the six-sided box check, scanned it, confirming it was the right one, and before I could move on to my next pick, a trainee asked, "How did you know that was the right box?", which required a several minute explanation of how the item description and the slight differentiations of the boxes led to that conclusion.
The big win would be training the folks doing stowing to not create such situations and to put markedly different things in each rainbow bin.
This would be a more convincing take if reasoning LLMs didn't already exist. Given the growth in capability over the last few years alone nothing about your description "several minute explanation of how the item description and the slight differentiations of the boxes" seems beyond an artificial intelligence to solve by the time humanoid robots would be ready to physically traverse a warehouse.
Your last point is also interesting given perhaps a robot is more amenable to such instruction, thus creating cascading savings. Each human has to be trained, and could be individually a failure. Robot can essentially copy its "brain" to its others.
Or likely more accurately, download the latest brain trained from all the robot's aggregate experiences from the amazon hivemind hq
The "Markedly Different things" in each bin was a big Amazon Warehouse advance in warehousing. Traditionally - things that were "alike" were put on shelves/bins - but (according to Amazon) it was far more efficient for pickers (at least back in the day - may have changed since then) to have random things on shelves located near each other to allow for equal access to popular items by pickers.
I have high respect of Tuomas and his work around SAC for RL in robotics.
But this is slightly unconvincing, most because of the author
>They spend thousands of computer hours practicing movements inside simulated worlds and inventing their own solutions.
This is exactly what almost every other picking startups have been doing for the last couple of years.
I can think of at least a dozens, some even making their custom gripper hardware. It still relies on sim2real transfer and then there's a bottleneck of things such as representing deformable objects. And that's still just scratching the surface of it.
I can definitely see that they have the right team. But the claim made by this author is far removed from the actual demo he describes. I've seen same demo for years, last one was in CoRL by Google (Gemini) and even then you could see clever robotics guy (some Boston dynamics engineers) that came by and gave it a clever task it failed on.
> But the claim made by this author is far removed from the actual demo he describes. I've seen same demo for years
The article describes multiple demos. Are you referring to the chicken nuggets one? That sounded pretty impressive to me. Is there publicly available videos of this?
They today have similar system that can quickly sort dumplings (more sensitive than chicken nuggets) ob conveyor belt.
No sim2real even needed. That haptics sensor is dirtcheap; camera based haptics sensor are today even available as open-source hardware that you can assemble for cheap.
If we don't limit to company demos we can dig up demos from I think almost a decade ago, and at least ~5 years ago for company demos.
Universal Robots ( https://www.universal-robots.com/ ) force sensing collaborative platforms were very advanced years ago, but like most bot firms small market demand made retail consumer pricing unsustainable.
>I can think of at least a dozens, some even making their custom gripper hardware.
The simplest solution sometimes is more robust in practice:
Too many edge-case failure modes in an uncontrolled setting. Building platforms that could seriously harm people by just falling over is an inherent design risk. =3
All my life I've loved robotics, so I was very eager to get things in the house, but my primary problem with humanoid robots is that they're very different from my Roomba-successor Dreame vacuum in a crucial way: they can fall. The Dreame can occupy the same space as my toddler, but the more industrial grade robotics machines cannot. The Unitree Sun Wukong is unbelievably impressive and I could completely imagine a world where it replaces humans in existing dangerous spaces without requiring the spaces themselves retooled. But in my house, perhaps the future will be like what these guys say and I'll have an Eka Claw on my kitchen counter and another by my washing machine, and so on.
In the classic example of old-guy-gets-surprised-by-new-tech, I bet people will find a way around the problem: but the thing has to be powerful to be fast, and if it's powerful it can hurt.
Who can tell. It was just prior to the pandemic when I was showing my wife talktotransformer.com and thinking about how much needs to be solved before it's useful. More fool am I HAHA!
The question is do they fall and can't get back up
The main issue is how heavy duty they are, because they operate on lithium batteries you can't make them too heavy otherwise it burns battery. So these humanoid robots durability will be closely aligned with innovation in lithium battery tech, or having larger and expensive robots with lots of battery.
That's not a robot problem, that's a toddler problem.
We don't leave our young toddlers to roam freely around the house for a reason. Our homes are full of hazards to these risk-seeking small people and a robot is just one more on the list.
Not that big of a problem, right? Just put a lot of power sockets throughout the workspace. Robot gets to its work station, can be tethered and recharge when it's operating there. Similarly in a household.
I got given a Huawei one for free which I found useful while I had a housemate with a cat since the place needed a vacuum almost twice a day. But after he moved out I just went back to vacuuming manually since it’s easier than having to scan the floor for every cable or throw rug it might get jammed on.
I don’t want to say home robotics will never happen since it seems likely eventually it will. But I think the deployment will be much much slower than entirely software based products like ChatGPT.
You won’t have an eka claw. You will have a humanoid. It’s a no brainer. You will get used to the “danger” just like we got used to the danger involved in driving a car or ceiling fans or propane home heating. Every year you’ll have a handful of injuries/deaths but eventually because of how useful they are no one will care and rightfully so.
what's the danger involved with ceiling fans (unless you are Korean)?
now that I think about it I can only remember videos of people doing really stupid things with them, then being surprised by really bad results, but never heard about any of them endangering anyone during normal operation
In rooms with low ceilings people can sometimes get hit if they’re standing on their beds. Basically the point I’m trying to make is you have something with metal blades spinning really fast only a few feet above your head.
It seems silly to be talking about a “ChatGPT moment” for a piece of industrial hardware that no regular person will ever have any cause to consider buying.
The ChatGPT moment was when they launched a product that was generally useful to the average person. Something that isn’t a consumer product at all is very unlikely to achieve success in the consumer market.
I'd consider hands to be more important than bipedal mobility.
I work in R&D, supporting a high-tech factory. The factory has already been laid out so that the entire place is accessible for materials being moved around on carts. The worker could be replaced by a cart with hands. If we could solve the hands problem right now, we'd be buying robots by the dozens.
Also, lots of things could be done right now by stationary robots. But at the present level of technology, what we really lack are programmers. Naturally what I'm saying could be overturned tomorrow by AI, so I'm talking in terms of how things work today. I'm actually one of the few people at the site with experience at industrial automation, but it's not part of my job at present.
In a sense, the hands we lack are hands on keyboards.
While I agree we'll see millions of bipedal robots, it won't be because they're doing our chores. People will buy them for the same reason I'd want one today: They're fun toys.
Even though today's robots are vastly more sophisticated, the progress of the last decade shows we shouldn't expect a sudden revolution in their abilities over the next ten years. As is often the case, solving those final few challenges that really make a difference always takes the longest.
Computers were nowhere for ever, then everyone had them.
The internet was tiny, then everywhere.
Smartphones were a teensy market, then everyone had them.
GLP1s were for a small group of diabetics, now a significant portion of the population take them.
This is how things playout time and time again.
Does it mean the commentors 10 years is correct? No. But it also doesn't need to be incredibly optimistic. All it takes is getting the robots right, and there are multiple companies who seem very close.
Mostly just the cost, yeah. It will be like buying a car. The economics will have to make sense for regular people, while it starts popping up in tons of places and become a status symbol.
Digital computers existed for ~10-20 years before hitting the consumer market. It took almost a half-century for the microprocessor to become a ubiquitous appliance.
We’re already seeing huge progress in humanoids coming from china. The big problem is software and world understanding, but the data collection from today’s humanoids and the rush to capitalize on their potential now that manufacturing their form is largely solved (save for hands) will see these problems overcome.
I expect it will be common to see them make deliveries in five years. Regular people don’t have to buy them for them to see widespread use.
So, after they work out all the mechanical kinks (there are quite a few!), and after they work out all the software issues (again, many of them), the last problem is the biggest: production. Anyone can make a half dozen robots by hand. A hundred thousand is a completely different challenge. If they can't be made efficiently, their cost makes them more of a toy than a tool.
Have you seen the mass produced humanoids from China? They’re incredibly capable (again, save for hands which is a huge mechanical and software problem) and cheap.
I can buy China doing it, but not Tesla. They have a terrible track record of production, nothing even close to China's capability. In the past they've "developed" factories by taking huge government incentives and then basically doing nothing with them and pocketing the cash.
Deliveries use hands because humans have hands, not because hands are a prerequisite for deliveries. Last mile is already “solved” with the little robots that drive around cities, no need for hands. Humans are useful because of our brains, because we can adapt to almost any situation for very little cost. Humanoid robots will remain a novelty until the cost is reduced far beyond what is plausible.
How do we define common? I’ll bet that in 5 years, the average person, even in somewhere like SF, will not see a humanoid robot during their every day life.
> Last mile is already “solved” with the little robots that drive around cities, no need for hands.
And yet we haven’t seen widespread adoption because they can’t handle stairs, steep slopes, streets without sidewalks, sidewalks with mud, or a hundred other real world challenges
We haven’t seen widespread adoption because they can’t hope to compete with human delivery drivers on cost. The cost to DoorDash and Uber Eats of a delivery driver is nothing upfront and a few dollars per delivery. The cost of a delivery robot is thousands of dollars upfront and more per delivery. Stairs aren’t even in the top 10 problems these robots face, they’re more than capable of delivering to most customers already.
The bipedal robot thing is interesting, but there's only two places their cost makes sense: industry and war. After war makes them cheap to mass-produce (because an army of robots needs to be sustainable), then they'll be affordable. But they'll still be highly regulated, mostly as a political reaction to "losing jobs". It will probably take 30+ years for us to get to that point, because wars big enough to invest that much expense and manpower aren't common.
Chinese Unitree already makes humanoids for $5K. Cheap enough for average american family to afford if it's useful. Several batteries and automatic replacement station will make it run 24/7 non-stop.
So, it terms of cheap capable hardware we are close. The problem is software and computing power.
That is interesting, but it looks like the ones used for practical work are $30k. Still, they're targeting 20k units this year, which is a lot more production and a lower price than I imagined they'd be at by now.
At least it's obvious we are close to have cheap mechanically capable robots. As soon as they get smart enough to be useful mass production will begin. We are almost there. (Un)fortunately China is leading here. With all production there it will be hard for US companies to compete. Especially outside of US market.
I remember a certain public personality who is very big on bipedal humanoid robots these days also promising us that we'd have truly autonomous self driving cars from his company by 2022, or 2023. It's now 2026.
We can debate the meaning of “truly autonomous”, but the Tesla-owning friends and acquaintances of mine have all, without any uncertainty, recently commented to me that the top-tier self-driving plan in the modern Teslas is just that.
One frequently uses it to drive from his house in LA to San Jose, another from Philly to Boston, another from Kamloops to Vancouver (Canada). I personally have never experienced it, but I trust their word and experiences enough to believe that it is at an extremely high level of capability.
Highway driving is a bit different from complex city conditions. Just look at the difference between the telsa robotaxi performance vs Waymo. Only one of them is truly FSD.
> Highway driving is a bit different from complex city conditions
Fair and valid, but worth noting that these drives are door-to-door, not just advanced highway cruise control.
Any idea where one might find a trusted source for data on the robotaxi performance? Especially curious about the latest self-driving models, rather than historical performance.
Today, Tesla's so-called full self-driving system is legally classified as SAE Level 2 driver assistance [1]. The human driver must continuously monitor the system, be ready to take over instantly at any time, and is legally responsible for the the car. Tesla is careful to avoid any liability for this by stating this somewhere, perhaps in a 3-point font.
Even if techbros loudly insist that they can take a nap in the back seat, that doesn't change the legal facts. Just like a drunk driver confidently shouting that he's totally fine to drive.
I hate how they are able to avoid liability like this. No human can sit in a car doing literally nothing but being alert ready to take over in an instant. Thats not how brains work. This is obvious but they use this excuse to divert any blame from the automated system to the occupant.
Yeah, I think 4+ legged bots should be more common than 2 leg variants. 2 legs is neat, but takes far more work and processing to control and balanced. It also requires much more powerful legs, a spider bot has more legs which makes it more "complex" in some ways, but individual legs don't need to hold and maneuver its entire body weight alone and it can hold 3 points of ground contact at all times, even when moving around, making it exceptionally stable. A bipedal robot has to be able to hold like twice its own body weight or more in order to balance and maneuver on a single foot in order to walk around and navigate obstacles.
As someone in the robotics, I can tell you that’s never gonna happen, even if you see a fully functional demo of a robot (not just the typical money grab 3D renders), assume the real life performance are 10x worse.. there’s so much monkey business in robotics, plenty of over promising, so much empty hypes, that been going on for years, the only successful breeds are cobotics (like roomba and industrial manipulators) or recently drones, although still very limited due to endurance.
Launched a product that, as I recall, was free. No real foreshadowing of what was about to come. Opened up an entirely new product category and started a process of reshaping at least the economy and probably society over the course of less than 5 years so far.
Yeah. I don't see how this is going to be a ChatGPT moment. Robot arms aren't a crazy new product. It might be big news regardless.
I don’t know when this ridiculous melodramatic style of writing started to pervade all of tech but it needs to go away. It’s resurrecting the pain of around 2016 when everyone presented like they were giving their own TED Talk.
Problem is talking in a simple, rational way doesn't get people to frantically cram huge wads of money down your pants. Was the same back in the day men in striped suits roamed the land selling curative tonics that were just alcohol and some smelly herbs.
I don't understand these frantic money people, but I do understand if you can figure out how to not be the greater fool you can make a lot of money. Seems kinda dumb this is how innovation is funded.
A product I think could work is a robo vacuum / drone combo. The drone will first fly around and blow dust from shelves etc, and most of it will end up on the floor. After that, the vacuum bot does its thing.
“Trillions of dollars flow through the human hand,” Agrawal says. “To me, this is the biggest problem in the world to be solved.”
They need to work on their messaging. "Human hands are a problem" is going to make enemies. Perhaps "relieve humans of menial chores" and "take over dangerous jobs" and "enable precision not possible with the human hand" etc.
Rodney Brooks has a great essay on why he's skeptical that the current humanoid hype will deliver and the central claim is that human dexterity is extremely advanced any today's humanoids lack even the sensors and data needed to start building the models needed to match human performance.
I saw him post this article on his Bluesky saying that they're the first ones he's seen that are close to cracking this issue (he's an investor/adviser).
Isn't it? The whole promise if humanoid robots is replacing humans in a human-centered environment. Instead of specialized hardware or modifying the existing process, just drop a robot in place of a human, bam, done. Otherwise, what's the point?
The point is that even if they do something 3x slower and maybe capable of 1/100 tasks they can still this do task 24/7, without holiday and never sick, they can also have more strength e.g in construction.
My smart vacuum is more dump than me when wiping floor and much slower than be but still greatly useful.
The problem - robots do break, they need constant maintenance, repair, and replacement (especially the smaller ones like the humanoids), and can go wrong in all sorts of situations. The costs for robot maintenance largely depend on the reliability of hardware and that should be included in the ROI calculation (which almost no one is doing right now)
that's the thing, what's the appeal of humanoid robots then? why not something more fit to the task? imagine if your roomba had legs because well that's what a human uses to move around when cleaning
Accessibility and a single chassis that does the vast majority of things. Even if they're never as fully dexterous as the average human (doubt it) they're still as dexterous as a somewhat handicapped human, which is already clearly enough to function decently in most of society and is far from useless.
If you want several bots all custom built to specific tasks, go for it. That will happen too. But a generalist has value of its own.
That would probably be an improvement. Floors are designed for people, and may have several levels. An ideal vacuum would probably look something like a centipede.
Anyway, the appeal would be that it can perform several tasks. It doesn't need to perform all the tasks a human can to fulfill that.
They don't need to be anywhere near as efficient or effective as me, they just need to be able to do it unattended. Roomba is a great example - it's great to be able to have floors cleaned every day while I'm working, even if it takes 10x longer to do so. If a robot could do laundry while I work, put away the dishes from the dishwasher in the morning, and mow my lawn while I'm BBQ'ing, it's a life changer.
I wonder how accurate joint positions and muscle activations can be from just a POV camera. Maybe it’s not crazy to think someone could get tens of millions of hours of well-labeled training data.
Yeah I’m going to completely disregard this because I feel like we are less than a year away from completely human feeling humanoids. This is based on nothing but obsessively watching and following humanoid progress on the internet.
What was eye-opening, or rather, sobering for me was when I read an interview with an engineer who explained how incredible difficult it is for a robot to orient itself when it is lying on the floor and wants to stand up.
Yes, it can do the required motions just fine, that’s not the point. But think about yourself when you are lying on the floor: it’s really easy to determine if this is safe, if you are lying underneath something and so on. You just feel that.
A robot cannot do that; all they can do is look around as good as possible and visually determine their situation.
Yeah but imagine yourself lying on the floor with your vision being your only sense, plus an info floating in your mind: „fyi, you are no longer upright“.
That’s all, you feel nothing else. Now your job is to move all parts of your body in just the right way.
The point about being aware of lying underneath some object was interesting. Sound might matter, like the frequency of background noise changes when you're in an enclosed space, and listening to your own shuffling noises helps you know when you've planted your feet right - or something. I have some really effective ear plugs and I notice they make it harder to move around.
Having said that, I've probably hit my head on the underside of an open cupboard door five or six times in my life, and I expect to do it again.
It is also things like "I can feel that my left knee is bearing a little too much weight, I should shift weight to my right hand and use that to push myself up" - things that come automatically to animals after learning the hard way in infancy (some of it is innate; baby animals are clumsy, but usually more mobile than human infants). Regardless of learned-vs-instinct, these abilities rely on sophisticated "sensors" and cognition. I suspect engineering the sensors is actually a bit harder, but I'm also not optimistic about a deep learning approach to the cognition.
A significant underappreciated advantage of animals over AI: lifeforms can "learn the hard way" more easily than 2020s robots because of cheap self-repair. AI labs are reluctant to damage their robots, but an essential part of humans learning to move safely is severely bonking your head and reckoning with the consequences - "hey, dummy, why did you trip and fall and bonk your head? Because you were running like an idiot."
I am learning the hard way to this day :) I have been practicing with work knives. A few months ago I got stupid and impatient, and sliced my thumb nastily. If I didn't block the cut with my thumbnail (still ruined) I might have chopped bone. It is hard to say precisely what I learned from this experience - "don't be stupid and impatient" is facile - but I know I learned a lot. I am actually optimistic about targeted surgical robotics. But for a general-use humanoid robot, I would not want to give it a knife if it's not capable of feeling pain. I never use big knives anywhere near my cats because I understand intuitively that they are nimble and unpredictable and easily stabbed by knives. I didn't need to be trained on this. A robot kind of does. Yikes.
I obsessively avoid any kind of "technology is going thataway" content. So I haven't seen anything that looks like humanoid progress in quite some time. About the only thing that has snuck around my barrier is Musk apparently claiming he'll have it by the end of the year, which is pretty conclusive evidence that they won't have it by the end of the year.
So if you're seeing anything that actually seems to merit attention, I'd love a few pointers. I could use some good news.
Well, as someone who has tried to build at least a couple small robot arms, I think we are probably closer to 20-50 years away. Both the power and dexterity are not there.
Right now, only a human can both push over a boulder and pick up a tiny speck from the floor using the same actuator.
"a ChatGPT moment" doesnt seem very momentous. ChatGPT was surprisingly good compared to previous smaller models. But since then the LLM scene has just been insane amounts of hype and bullshit and financial skulduggery. Their actual utility is pretty niche imo.
Just a few weeks ago at work we got a Universal Robots UR5 from another project in-house along with a Hand-E gripper.
I've never had so much fun programming and playing with a device ever. And it completely took me back to getting an Armatron 40 years ago and having so much fun - but also wishing I could somehow control it with software.
Many of the Chinese companies are doing very impressive open-loop sim2real. They make great demonstrations. They are not great at dealing with the real world and unpredictable environments.
(That's not true of all Chinese companies - some are doing really impressive work with closed loop systems in unpredictable environments. But many of the highly viewed ones with coordinated dance performances or martial arts are intended more as theater to government financial sponsors than useful function. The technically impressive performances do not look as visually impressive.)
those were impressive but were also RC. I think an important part of robotics is not just the mechanics of humanoid motion, but the independent control of those mechanics.
Back in the 90s, I developed a rule of thumb: if I saw it in Wired, it's because it was either already over, or it wasn't going to happen at all.
I was so disappointed when I saw BetterPlace (the car with replaceable batteries) on the cover of Wired. It seemed like such a good idea. Too bad the rule of thumb meant it wouldn't work.
Rules of thumb were made to be broken. Maybe this time it will be different.
Yeah, betterplace made it from 2008 (wired) to 2013 (bankruptcy.) Nio is trying again and it looks like they hit wired in 2018, again in 2023, and are still active today...
Just run in the console window=null and you are good. It is valuable service until the websites get their shit together and finally fix their payments model.
> archive.is is malicious -- as in, uses your browser to launch DDoS attacks, and other things.
I think the attack was itself a response to a doxxing attempt. Also, archive.is being a free service doesn't quite fit with claiming they are malicious. The overall picture seems still positive.
Works for me. I use only Tor so it is actually far more accessible. Archive.is uses Google's Recaptcha, which for some reason rejects valid solutions submitted via Tor.
You don’t need to MITM it, this was a common pattern for a long time (not sure it still works though). There was no origin verification so you could just use a different site ID and have people respond to captchas you encountered on that site.
just because it's an article about techie stuff doesn't mean all the photojournalism has to be color-graded like a Matrix movie.
.. but it's kind of funny to read the fluff PR about saving humanity while juxtaposing it against photos that look like they may as well be screencaps from Prometheus or Black Mirror.
I'm having some house painting done and the painter asked me what line of work I was in. When I said computer programming he said, "ooh, bet you're worried about AI! At least painters are safe!"
Unfortunately, the only robots available will be connected to the cloud, paid by subscription, and will gather a continuous feed of audio-video data from you and your home. And sometimes it will be teleoperated, and you might not know when.
Cloud connected (robot AI in cloud) home robots would be very unsafe, due to network slowdown/outages. Imagine it freezing/stopping right after it turns on water faucet or stovetop.
Why would consumers settle for that? Local models have scaled quite quickly. Just pair the bot with a LAN server as the brain that keeps all your data private.
Barring that, choose bots that use Zero Knowledge Proof architectures for all data so you know there's no in/out of personal data, only security proofs. This makes rental robots certifiably private too.
“Sure I’ll clean up the house, Mr. J. While I’m doing, so have you seen the new shoes from crocs? They’re sponsored by the Jenners and have great new designs with all of your favorite movie characters on them! Would you like me to order you a pair?”
“Hey Mr J for a low 7.99 a month I can unlock the Harley Quinn voice pack! For 39.99 we can upgrade you to unlock TikTok Rosie dance mode with special Harley Fortnite dance, a joker LCD breastplate for me and special “psycho partner” romance mode. What so you say, Mr J?”
I've spent ~$500 this month trying to get an LLM model to solve a Rubik's Cube. They can't. I'll post my Rubik's Cube MCP server next week if anyone wants to prove me wrong.
1. a human child learning 6 algorithms and a weekend can solve a Rubik's Cube
2. Reenforcement learning can solve a Rubik's Cube
3. The best LLM model using recursive tuning or not can't solve a Rubik's Cube.
Claude 4.6 got 60% of the way but couldn't figure out the last steps after running for 20+ minutes and hundreds of thousands of tokens.
People already pay shitton of money for silicon sex dolls and fantasize about robot sex online. Sex toys are connected to a network for remote control. As soon as a a humanoid robot becomes commercially accessible some will have sex with it.
We'll know this works when it starts replacing Amazon pickers in quantity. Amazon has been trying to automate that for years, with many demos and contests. So far, nothing can quickly and reliably take random products out of one bin and put them in another. Amazon's robotic systems move larger containers and shelves of bins around, but do not yet pick individual items.
You raise a great point. And the Amazon picking staff are onshore in wealthy countries. I guess the minimum wage paid by Amazon is around 15 USD per hour.
I wonder: Is the task of automating this work primaryly difficult in vision or dexterity (motion)? Or maybe they are equally difficult for different reasons.
Probably both vision and dexterity, and the first mistake we make as roboticists/engineers might be to distinguish the two like they're separate problems to solve or that a solution exists where the two live a separate life.
https://rodneybrooks.com/why-todays-humanoids-wont-learn-dex...
Agreed. The solution will likely be some vision foundation model that directly sends controls to the robot ("move here, grab, move there"), trained by Amazon with RL to integrate collision avoidance, object detection, grasping point detection, grasp verification etc.
If we're talking about picking objects at random from one bin and putting it in another, I don't need my eyes to do that. Proprioception (shape and location) and touch (texture) are enough to do that.
There's a lot more money being thrown at this than in previous years. Seems to be growing beyond corporate R&D labs and university research towards startups trying to productize it.
I've seen multiple articles about robotic claws. This one made the rounds previously https://www.firgelli.com/pages/humanoid-robot-actuators
>A humanoid robot takes roughly 5,000 steps per hour. Each step sends a shock of 2–3× body weight through the leg actuators—forces that would be fine occasionally, but become destructive when repeated thousands of times without pause.
As someone who comes from the world of running and knee problems, I feel this misses the issue. Normal walking should not produce these kinds of shocks unless your gait is really jumpy or otherwise screwed up. You only start to see these forces when running and that's where technique becomes important even for humans if you want to prevent damage to your joints over long distances. But at least for walking I suppose that a fully articulated humanoid with all the degrees of freedom of human gait should be mostly a control problem, not a mechanical engineering one.
The force an impulse generates on a contact depends on the speed of deceleration. It's just F=m*a
Slow deceleration leads to low forces. If you have a contact event with a hard substance, like a rigid metal for accurate kinematics, the deceleration to zero upon a contact has to happen instantly. Meaning the deceleration is incredibly high, resulting in extremely high forces for a few milliseconds.
Human bodies are made out of a flexible and impact resistant material: water. When a contact event happens, your body deforms, which means that the deceleration happens over a longer time frame with less force. Not just that, your muscles also have a certain amount of flexibility in them and basically zero internal inertia. All the inertia is in the limb as a whole, whereas for a robot there is a spinning motor and gearbox that needs to slow down as well.
You could solve this as a control problem by adding series elastic actuators, which means you need to change your mechanical design.
The human body goes further than that too, when you're out jogging - as your foot approaches the ground for a stride, you slow the velocity of your foot downwards towards the ground so there's less of a sudden deceleration.
Imagine when you throw a tennis ball high in the sky, and then you catch it on your racket without bouncing by matching it's velocity, your feet do the same thing with the ground on a smaller scale.
Maybe this workforce is useful not because of it's direct output, but for it's mere existence : look politian, I'm creating jobs !
As others say, not necessarily. The breakeven point for jobs like Amazon may be quite low (or high? I mean simply "not yet there").
I'd say that we'll know it works when robots with those hands start turning out on the Russo-Ukrainian frontline en masse, because it is there where the lack of manpower has the most pressing and brutal consequences, and cannot be mitigated by usual peacetime incentives (e.g. better benefits).
That frontline has already sucked in all the automatization innovations of the last decade, as long as they proved themselves in combat.
> We'll know this works when it starts replacing Amazon pickers in quantity.
That doesn’t follow. There are plenty of tasks that can be fully and reliably automated but aren’t, for the simple reason that human labor is dirt cheap compared to advanced robotics.
I disagreed, then re-read your post, then re-read the OP, and now I've come full circle to apologize; I think you make a fair point.
I work at a biotech. We spent who knows how much time and money trying to develop a 'lab technician bot' to automate one of our critical assays. Turns out, a 6-figure machine still isn't as economical as my coworker Y, one of the veteran lab-technicians. Sure she takes the occasional sick day but even at our volume (and we do industrial-level, multiple clients batched into a single assay pass) it won't be economical to replace her for a very long time (if we even reach that scale).
Absolutely. I worked at a gene sequencing company and I led the software side of making a robotic product[0] to automate the 20-30 minutes of sample preparation time. It's great for lots of uses, but for anything outside the exact thing it automates, it doesn't cover it. For that you need an expert human.
[0] https://web.archive.org/web/20250919140427/https://nanoporet...
But that does follow. The economics working is not some outside factor. If the robot “could do the task” but would cost more than paying a human to do the same task then the robot “does not work”. It is frequently because the robot would be too slow, or not reliable enough, or could only handle certain types of items. But ultimately all of these boil down to cost.
We have seen lab demoes of robotic manipulation for decades. The reason why they stay in the lab (when they do) and don’t become ubiquitous is because they are not good enough. In other words they don’t work. The economics and “does it work” is not two separate concerns but one and the same.
It's a continuum, not binary. The same robot that doesn't financially "work" for replacing a manual scavenger sorting garbage in an African slum might be quite cost-effective sorting recycling in Switzerland, and would likely have a niche regardless of price if used to (say) sort biohazardous or radioactive materials. And there are already millions of robots out there assembling cars etc.
What is the point of humanoid “general” robots then? We already have pretty reliable ways to make and train humans. Humans are cheaper and better than robots. I could imagine robots for some specialised tasks where you don’t want to use a human for eg security reasons, but you don’t need general purpose robots for that
Robots are good at things that are "simple" but where human precision is not good enough, or where people would get bored and make mistakes.
technology gets cheaper over time. If they were always going to cost the amount they do now, you might have a point. But they'll eventually get much cheaper.
If robots ever do get cheaper than humans for it, though?
In natural ecosystems, nobody beats the apex predator directly, and nobody beats the hyperspecialized niche critter at their own game. The new species has some advantage that’s different than what is there.
If a humanoid robot is slower dumber human that is expensive, requires power, can’t get wet, falls over, and doesn’t understand stairs. Is not sleeping and being radiation tolerant enough of an advantage to be worth it?
The nature comparison doesn't work on a fundamental level because you're only getting a fraction of the human's power based on how much they're happy to sell.
You forgot a big one in your description of the hypothetical advantages:
No free will
They already are, the problem with humanoid robots is that people think that adding legs to the robot will somehow fundamentally make it more intelligent.
People see a robot arm attached to a stationary platform and understand it requires integration work to perform a single task.
But when those same people see a humanoid robot, they think they can just talk to it like a real human and it will do what you told it to do.
They don't think about the fact that the humanoid robot has to be programmed exactly the same way the stationary robot arm has to be programmed and that programming the legs in addition to the arms is a much more challenging problem.
Robots can be optimized for tasks and if they are, their benefits are greater. When cars replaced the horse, it was because they didn’t poop, and because a car designed only for transport would not suddenly have a heart attack and stop working.
Funnily enough, cars have their own way of pooping and dying of a heart attack.
Cars can stop working suddenly in many many ways, for many reasons.
A friend who works at Amazon made the same point: "We don't really need robots in the FCs urgently [other than the Kivas], because it turns out you can just pay people $17/hour"
I was thinking this week that AI token costs are probably going to get so expensive soon that bright spark CEOs are going to realise “why am I paying for such expensive coding agents when I can pay people from the third world to code!?!” and announce outsourcing like it’s some kind of stunning and innovative revelation.
I think you didn’t grok the hidden punchline - this is the stage after they’ve replaced all their third world coders with AI agents, until one day a C-suiter has the revelation that humans are cheaper and better, and the company then starts toting its humanistic credentials all over LinkedIn.
Mechanical picking has been too slow. It's not a problem with the robot mechanics. Here's 300 picks/minute from 2012.[1] The parts are all the same, so the vision problem is simple.
But picking arbitrary objects from fulfillment bins is still running at a few picks per minute.[2] As the speed picks up, humans become less necessary.
[1] https://www.youtube.com/watch?v=6RKXVefE98w
[2] https://www.youtube.com/watch?v=2X4CU3jmw-g
That's the point of the test condition. When running a robot becomes more economical than paying full-scale humans $17/h, something important about robot abilities will have changed.
I dunno, I worked in an Amazon Warehouse for a year part-time (and a couple of weeks full-time when in-between jobs) --- on one occasion, I pulled up to a bin full of non-descript cardboard boxes near where a group of trainees were working their way through, grabbed one box, spun it around for the six-sided box check, scanned it, confirming it was the right one, and before I could move on to my next pick, a trainee asked, "How did you know that was the right box?", which required a several minute explanation of how the item description and the slight differentiations of the boxes led to that conclusion.
The big win would be training the folks doing stowing to not create such situations and to put markedly different things in each rainbow bin.
This would be a more convincing take if reasoning LLMs didn't already exist. Given the growth in capability over the last few years alone nothing about your description "several minute explanation of how the item description and the slight differentiations of the boxes" seems beyond an artificial intelligence to solve by the time humanoid robots would be ready to physically traverse a warehouse.
Your last point is also interesting given perhaps a robot is more amenable to such instruction, thus creating cascading savings. Each human has to be trained, and could be individually a failure. Robot can essentially copy its "brain" to its others.
Or likely more accurately, download the latest brain trained from all the robot's aggregate experiences from the amazon hivemind hq
The "Markedly Different things" in each bin was a big Amazon Warehouse advance in warehousing. Traditionally - things that were "alike" were put on shelves/bins - but (according to Amazon) it was far more efficient for pickers (at least back in the day - may have changed since then) to have random things on shelves located near each other to allow for equal access to popular items by pickers.
“Eka, open claw!!!!”
“I’m sorry, OpenClaw is not approved for an account on your subscription tier.”
suffocates from being choked by robotic claw
Masterpiece.
This comment is not going unappreciated
I have high respect of Tuomas and his work around SAC for RL in robotics.
But this is slightly unconvincing, most because of the author
>They spend thousands of computer hours practicing movements inside simulated worlds and inventing their own solutions.
This is exactly what almost every other picking startups have been doing for the last couple of years.
I can think of at least a dozens, some even making their custom gripper hardware. It still relies on sim2real transfer and then there's a bottleneck of things such as representing deformable objects. And that's still just scratching the surface of it.
I can definitely see that they have the right team. But the claim made by this author is far removed from the actual demo he describes. I've seen same demo for years, last one was in CoRL by Google (Gemini) and even then you could see clever robotics guy (some Boston dynamics engineers) that came by and gave it a clever task it failed on.
> But the claim made by this author is far removed from the actual demo he describes. I've seen same demo for years
The article describes multiple demos. Are you referring to the chicken nuggets one? That sounded pretty impressive to me. Is there publicly available videos of this?
I'm describing all of them.
As for chicken nugget here's for example one (company) showing same capacity 4 years ago
https://youtu.be/6SbpfN5ed38?si=srtdZCdKOdPZ_wRn
They today have similar system that can quickly sort dumplings (more sensitive than chicken nuggets) ob conveyor belt.
No sim2real even needed. That haptics sensor is dirtcheap; camera based haptics sensor are today even available as open-source hardware that you can assemble for cheap.
If we don't limit to company demos we can dig up demos from I think almost a decade ago, and at least ~5 years ago for company demos.
Universal Robots ( https://www.universal-robots.com/ ) force sensing collaborative platforms were very advanced years ago, but like most bot firms small market demand made retail consumer pricing unsustainable.
>I can think of at least a dozens, some even making their custom gripper hardware.
The simplest solution sometimes is more robust in practice:
https://news.uchicago.edu/story/balloon-filled-ground-coffee...
Too many edge-case failure modes in an uncontrolled setting. Building platforms that could seriously harm people by just falling over is an inherent design risk. =3
All my life I've loved robotics, so I was very eager to get things in the house, but my primary problem with humanoid robots is that they're very different from my Roomba-successor Dreame vacuum in a crucial way: they can fall. The Dreame can occupy the same space as my toddler, but the more industrial grade robotics machines cannot. The Unitree Sun Wukong is unbelievably impressive and I could completely imagine a world where it replaces humans in existing dangerous spaces without requiring the spaces themselves retooled. But in my house, perhaps the future will be like what these guys say and I'll have an Eka Claw on my kitchen counter and another by my washing machine, and so on.
In the classic example of old-guy-gets-surprised-by-new-tech, I bet people will find a way around the problem: but the thing has to be powerful to be fast, and if it's powerful it can hurt.
Who can tell. It was just prior to the pandemic when I was showing my wife talktotransformer.com and thinking about how much needs to be solved before it's useful. More fool am I HAHA!
All I want is a machine that I can drop ingredients in it and it can give me a delicious meal
And another that I can just drop all my clothes in, and have them washed and ironed for me.
Doesn't have to be a humanoid.
If you like smoothies, I think I've got something for you!
> All I want is a machine that I can drop ingredients in it and it can give me a delicious meal
https://www.thermomix.com/
> In 1971, the original Thermomix VM 2000 was launched on the market – first in France, later in Spain and Italy.
There are also robotic versions like Nosh One where you just need to insert the chopped ingredients and it sequences the cooking operations for you.
> in a crucial way: they can fall.
The question is do they fall and can't get back up
The main issue is how heavy duty they are, because they operate on lithium batteries you can't make them too heavy otherwise it burns battery. So these humanoid robots durability will be closely aligned with innovation in lithium battery tech, or having larger and expensive robots with lots of battery.
I think he meant it can fall onto his toddler, causing injury.
Could we not simply encase the weaker unit some kind of armored robotic shell?
Then we would need better baby-gates. But that might lead to escalating scenarios.
Tragedy of the commons toddler armor. Can you guys fucking not? It’s best to death. I’d rather have a Ford vs Chevy vs Ram conversation again.
Might as well put an AI chip in the toddler's exosuit and get another bot
That's not a robot problem, that's a toddler problem.
We don't leave our young toddlers to roam freely around the house for a reason. Our homes are full of hazards to these risk-seeking small people and a robot is just one more on the list.
Not that big of a problem, right? Just put a lot of power sockets throughout the workspace. Robot gets to its work station, can be tethered and recharge when it's operating there. Similarly in a household.
I think the focus has been on lowering mass so that they can move quickly with low kinetic energy.
Meanwhile the Roomba-successor robots that I've tried still get tangled up on our laptop charger cables and wedged under the coffee table.
I got given a Huawei one for free which I found useful while I had a housemate with a cat since the place needed a vacuum almost twice a day. But after he moved out I just went back to vacuuming manually since it’s easier than having to scan the floor for every cable or throw rug it might get jammed on.
I don’t want to say home robotics will never happen since it seems likely eventually it will. But I think the deployment will be much much slower than entirely software based products like ChatGPT.
I don't see a future with humanoid robots inside the house. We probably will have specialized robots for certain task like the roombas.
I fully expect to buy one within the next three years. Probably Optimus 4, depending on the price.
You won’t have an eka claw. You will have a humanoid. It’s a no brainer. You will get used to the “danger” just like we got used to the danger involved in driving a car or ceiling fans or propane home heating. Every year you’ll have a handful of injuries/deaths but eventually because of how useful they are no one will care and rightfully so.
Well, except we have 5 decades of cautionary tales in film that show plausible ways this goes sideways when everything is connected to the internet.
what's the danger involved with ceiling fans (unless you are Korean)?
now that I think about it I can only remember videos of people doing really stupid things with them, then being surprised by really bad results, but never heard about any of them endangering anyone during normal operation
In rooms with low ceilings people can sometimes get hit if they’re standing on their beds. Basically the point I’m trying to make is you have something with metal blades spinning really fast only a few feet above your head.
It seems silly to be talking about a “ChatGPT moment” for a piece of industrial hardware that no regular person will ever have any cause to consider buying.
The ChatGPT moment was when they launched a product that was generally useful to the average person. Something that isn’t a consumer product at all is very unlikely to achieve success in the consumer market.
In less than 10 years there’s going to be millions of bipedal robots everywhere, doing all sorts of chores for us. They’re going to need hands.
I'd consider hands to be more important than bipedal mobility.
I work in R&D, supporting a high-tech factory. The factory has already been laid out so that the entire place is accessible for materials being moved around on carts. The worker could be replaced by a cart with hands. If we could solve the hands problem right now, we'd be buying robots by the dozens.
Also, lots of things could be done right now by stationary robots. But at the present level of technology, what we really lack are programmers. Naturally what I'm saying could be overturned tomorrow by AI, so I'm talking in terms of how things work today. I'm actually one of the few people at the site with experience at industrial automation, but it's not part of my job at present.
In a sense, the hands we lack are hands on keyboards.
Yes, thats your current perspective. But by the time you get programmers, the whole company will be automated
While I agree we'll see millions of bipedal robots, it won't be because they're doing our chores. People will buy them for the same reason I'd want one today: They're fun toys.
It's been 10 years since Boston Dynamics released this impressive video of Atlas. https://www.youtube.com/watch?v=rVlhMGQgDkY
Even though today's robots are vastly more sophisticated, the progress of the last decade shows we shouldn't expect a sudden revolution in their abilities over the next ten years. As is often the case, solving those final few challenges that really make a difference always takes the longest.
That’s an incredibly optimistic timeline.
Slowly, then all at once.
Computers were nowhere for ever, then everyone had them. The internet was tiny, then everywhere. Smartphones were a teensy market, then everyone had them. GLP1s were for a small group of diabetics, now a significant portion of the population take them.
This is how things playout time and time again.
Does it mean the commentors 10 years is correct? No. But it also doesn't need to be incredibly optimistic. All it takes is getting the robots right, and there are multiple companies who seem very close.
It took almost 20 years from computers that nobody brought on electronics and photography stores to computers in everybody's desk.
Robots will probably be slower, because there is way less room for optimizing their cost.
We have had more than 10 years of robotic vacuums and yet they are still a fairly niche product.
robots have existed for more than 20 years though. Boston Dynamic's dog is 22 years old, Atlas 15-ish
What can I as a normal person use these robots for?
Nothing. They were designed for the military.
And what do the military use them for?
Supposedly as a sort of extremely expensive but basically tireless mule.
Mostly just the cost, yeah. It will be like buying a car. The economics will have to make sense for regular people, while it starts popping up in tons of places and become a status symbol.
Digital computers existed for ~10-20 years before hitting the consumer market. It took almost a half-century for the microprocessor to become a ubiquitous appliance.
We’re already seeing huge progress in humanoids coming from china. The big problem is software and world understanding, but the data collection from today’s humanoids and the rush to capitalize on their potential now that manufacturing their form is largely solved (save for hands) will see these problems overcome.
I expect it will be common to see them make deliveries in five years. Regular people don’t have to buy them for them to see widespread use.
So, after they work out all the mechanical kinks (there are quite a few!), and after they work out all the software issues (again, many of them), the last problem is the biggest: production. Anyone can make a half dozen robots by hand. A hundred thousand is a completely different challenge. If they can't be made efficiently, their cost makes them more of a toy than a tool.
Have you seen the mass produced humanoids from China? They’re incredibly capable (again, save for hands which is a huge mechanical and software problem) and cheap.
https://youtu.be/mUmlv814aJo https://youtu.be/GzX1qOIO1bE
But what do they do?
I’ve only ever seen them performing choreographed routines or running races.
I’ve yet to see one doing something useful, so if you know of an example, I’d love to see it.
One of the Tesla's factories is winding down car production in a plan to convert to producing humanoid robots.
I can buy China doing it, but not Tesla. They have a terrible track record of production, nothing even close to China's capability. In the past they've "developed" factories by taking huge government incentives and then basically doing nothing with them and pocketing the cash.
The line between moonshot and scam is thin and elon is known for doing jumping jacks on it.
Is that because they're close to building robots or because Chinese EVs are eating their lunch and Tesla is flatlining?
Deliveries use hands because humans have hands, not because hands are a prerequisite for deliveries. Last mile is already “solved” with the little robots that drive around cities, no need for hands. Humans are useful because of our brains, because we can adapt to almost any situation for very little cost. Humanoid robots will remain a novelty until the cost is reduced far beyond what is plausible.
How do we define common? I’ll bet that in 5 years, the average person, even in somewhere like SF, will not see a humanoid robot during their every day life.
> Last mile is already “solved” with the little robots that drive around cities, no need for hands.
And yet we haven’t seen widespread adoption because they can’t handle stairs, steep slopes, streets without sidewalks, sidewalks with mud, or a hundred other real world challenges
We haven’t seen widespread adoption because they can’t hope to compete with human delivery drivers on cost. The cost to DoorDash and Uber Eats of a delivery driver is nothing upfront and a few dollars per delivery. The cost of a delivery robot is thousands of dollars upfront and more per delivery. Stairs aren’t even in the top 10 problems these robots face, they’re more than capable of delivering to most customers already.
I predict that we'll see limited combat use of the latest Chinese bipedal robots within two years. They'll sell to both Russia and Ukraine.
Ukraine already has UGVs and they're tracked or wheeled. I can't imagine bipedal robots being cost efficient for 99% of missions in the near future.
The bipedal robot thing is interesting, but there's only two places their cost makes sense: industry and war. After war makes them cheap to mass-produce (because an army of robots needs to be sustainable), then they'll be affordable. But they'll still be highly regulated, mostly as a political reaction to "losing jobs". It will probably take 30+ years for us to get to that point, because wars big enough to invest that much expense and manpower aren't common.
Also medical / elderly care. A large market.
Chinese Unitree already makes humanoids for $5K. Cheap enough for average american family to afford if it's useful. Several batteries and automatic replacement station will make it run 24/7 non-stop.
So, it terms of cheap capable hardware we are close. The problem is software and computing power.
That is interesting, but it looks like the ones used for practical work are $30k. Still, they're targeting 20k units this year, which is a lot more production and a lower price than I imagined they'd be at by now.
At least it's obvious we are close to have cheap mechanically capable robots. As soon as they get smart enough to be useful mass production will begin. We are almost there. (Un)fortunately China is leading here. With all production there it will be hard for US companies to compete. Especially outside of US market.
"Especially outside of US market."
A euphemism in polite company for: we'll ban them on national security grounds like we did with cars and phones.
I remember a certain public personality who is very big on bipedal humanoid robots these days also promising us that we'd have truly autonomous self driving cars from his company by 2022, or 2023. It's now 2026.
We can debate the meaning of “truly autonomous”, but the Tesla-owning friends and acquaintances of mine have all, without any uncertainty, recently commented to me that the top-tier self-driving plan in the modern Teslas is just that.
One frequently uses it to drive from his house in LA to San Jose, another from Philly to Boston, another from Kamloops to Vancouver (Canada). I personally have never experienced it, but I trust their word and experiences enough to believe that it is at an extremely high level of capability.
Highway driving is a bit different from complex city conditions. Just look at the difference between the telsa robotaxi performance vs Waymo. Only one of them is truly FSD.
> Highway driving is a bit different from complex city conditions
Fair and valid, but worth noting that these drives are door-to-door, not just advanced highway cruise control.
Any idea where one might find a trusted source for data on the robotaxi performance? Especially curious about the latest self-driving models, rather than historical performance.
Today, Tesla's so-called full self-driving system is legally classified as SAE Level 2 driver assistance [1]. The human driver must continuously monitor the system, be ready to take over instantly at any time, and is legally responsible for the the car. Tesla is careful to avoid any liability for this by stating this somewhere, perhaps in a 3-point font.
Even if techbros loudly insist that they can take a nap in the back seat, that doesn't change the legal facts. Just like a drunk driver confidently shouting that he's totally fine to drive.
[1] https://en.wikipedia.org/wiki/Tesla_Autopilot
I hate how they are able to avoid liability like this. No human can sit in a car doing literally nothing but being alert ready to take over in an instant. Thats not how brains work. This is obvious but they use this excuse to divert any blame from the automated system to the occupant.
To add context, his first predictions are from 2013. And some of those predictions had 2018 as a goal.
Why bipedal?
Because the human world is built for bipedal beings and everything else will encounter obstacles somewhere.
Dogs and cats don't complain. Gorillas and chimpanzee should be fine too.
Bipedal robots suck right now, but superhuman stability is achievable in near future.
Yeah, I think 4+ legged bots should be more common than 2 leg variants. 2 legs is neat, but takes far more work and processing to control and balanced. It also requires much more powerful legs, a spider bot has more legs which makes it more "complex" in some ways, but individual legs don't need to hold and maneuver its entire body weight alone and it can hold 3 points of ground contact at all times, even when moving around, making it exceptionally stable. A bipedal robot has to be able to hold like twice its own body weight or more in order to balance and maneuver on a single foot in order to walk around and navigate obstacles.
As someone in the robotics, I can tell you that’s never gonna happen, even if you see a fully functional demo of a robot (not just the typical money grab 3D renders), assume the real life performance are 10x worse.. there’s so much monkey business in robotics, plenty of over promising, so much empty hypes, that been going on for years, the only successful breeds are cobotics (like roomba and industrial manipulators) or recently drones, although still very limited due to endurance.
Launched a product that, as I recall, was free. No real foreshadowing of what was about to come. Opened up an entirely new product category and started a process of reshaping at least the economy and probably society over the course of less than 5 years so far.
Yeah. I don't see how this is going to be a ChatGPT moment. Robot arms aren't a crazy new product. It might be big news regardless.
This might age badly heh. Kinda like "we only need 2 MAYBE 3 computers for Sweden" (real thing people said back in the day).
I don’t know when this ridiculous melodramatic style of writing started to pervade all of tech but it needs to go away. It’s resurrecting the pain of around 2016 when everyone presented like they were giving their own TED Talk.
Problem is talking in a simple, rational way doesn't get people to frantically cram huge wads of money down your pants. Was the same back in the day men in striped suits roamed the land selling curative tonics that were just alcohol and some smelly herbs.
I don't understand these frantic money people, but I do understand if you can figure out how to not be the greater fool you can make a lot of money. Seems kinda dumb this is how innovation is funded.
I am waiting for a robot that can dust my shelves even when there are things on them. That would improve my life a lot.
A product I think could work is a robo vacuum / drone combo. The drone will first fly around and blow dust from shelves etc, and most of it will end up on the floor. After that, the vacuum bot does its thing.
IMHO the big winner is gonna be the toilet/tub/shower-bot.
I can never not think of Howard Wolowitz anymore with such news.
“Trillions of dollars flow through the human hand,” Agrawal says. “To me, this is the biggest problem in the world to be solved.”
They need to work on their messaging. "Human hands are a problem" is going to make enemies. Perhaps "relieve humans of menial chores" and "take over dangerous jobs" and "enable precision not possible with the human hand" etc.
Presumably the people they're talking to agree that that's the problem.
A couple of minutes of video (presumably by the author):
https://www.wired.com/video/watch/this-company-is-building-s...
I'm not sure I grasp the hype here. It's a few dynamixels screwed together. They're not even particularly good servos.
Call me when it can tie a shoelace.
The robotics Turing test: change the nappies of the designer's and company owners' baby daughters or grand-daughters.
I'd already be fine with some decent laundry folding in general.
I think the idea is that before they sell it to the public they should trust it with their own loved ones.
There's a video of the founder of Figure robotics trusting it enough to let it do laundry next to his kids
https://x.com/adcock_brett/status/1950685253447913798
The first phase is likely don't let the kids go near it since it could easily hurt a human by accident.
Is it automated or is it like when they demoed that neo humanoid robot and it was actually just a dude driving it with vr goggles.
Notably it does not show the robot turning on the washing machine?
That requires the ProMax subscription at $2500/month.
Stockton Rush trusted his submarine with his own life.
> He criticized the Passenger Vessel Safety Act of 1993 as "needlessly prioritiz[ing] passenger safety over commercial innovation".
:-)))
Or to put it another way, before selling it for laundry folding, make sure it won't fold the baby that was left on the wrong table.
Other fun things. Living in apartment with only the robot doing any tasks or picking up any inputs like arriving packages.
Rodney Brooks has a great essay on why he's skeptical that the current humanoid hype will deliver and the central claim is that human dexterity is extremely advanced any today's humanoids lack even the sensors and data needed to start building the models needed to match human performance.
https://rodneybrooks.com/why-todays-humanoids-wont-learn-dex...
I saw him post this article on his Bluesky saying that they're the first ones he's seen that are close to cracking this issue (he's an investor/adviser).
> needed to match human performance.
This is not a remotely a real world requirement for them to be useful, and for them to sell like crazy.
Isn't it? The whole promise if humanoid robots is replacing humans in a human-centered environment. Instead of specialized hardware or modifying the existing process, just drop a robot in place of a human, bam, done. Otherwise, what's the point?
The point is that even if they do something 3x slower and maybe capable of 1/100 tasks they can still this do task 24/7, without holiday and never sick, they can also have more strength e.g in construction.
My smart vacuum is more dump than me when wiping floor and much slower than be but still greatly useful.
The problem - robots do break, they need constant maintenance, repair, and replacement (especially the smaller ones like the humanoids), and can go wrong in all sorts of situations. The costs for robot maintenance largely depend on the reliability of hardware and that should be included in the ROI calculation (which almost no one is doing right now)
that's the thing, what's the appeal of humanoid robots then? why not something more fit to the task? imagine if your roomba had legs because well that's what a human uses to move around when cleaning
Accessibility and a single chassis that does the vast majority of things. Even if they're never as fully dexterous as the average human (doubt it) they're still as dexterous as a somewhat handicapped human, which is already clearly enough to function decently in most of society and is far from useless.
If you want several bots all custom built to specific tasks, go for it. That will happen too. But a generalist has value of its own.
> imagine if your roomba had legs
That would probably be an improvement. Floors are designed for people, and may have several levels. An ideal vacuum would probably look something like a centipede.
Anyway, the appeal would be that it can perform several tasks. It doesn't need to perform all the tasks a human can to fulfill that.
They don't need to be anywhere near as efficient or effective as me, they just need to be able to do it unattended. Roomba is a great example - it's great to be able to have floors cleaned every day while I'm working, even if it takes 10x longer to do so. If a robot could do laundry while I work, put away the dishes from the dishwasher in the morning, and mow my lawn while I'm BBQ'ing, it's a life changer.
I wonder how accurate joint positions and muscle activations can be from just a POV camera. Maybe it’s not crazy to think someone could get tens of millions of hours of well-labeled training data.
Yeah I’m going to completely disregard this because I feel like we are less than a year away from completely human feeling humanoids. This is based on nothing but obsessively watching and following humanoid progress on the internet.
What was eye-opening, or rather, sobering for me was when I read an interview with an engineer who explained how incredible difficult it is for a robot to orient itself when it is lying on the floor and wants to stand up.
Yes, it can do the required motions just fine, that’s not the point. But think about yourself when you are lying on the floor: it’s really easy to determine if this is safe, if you are lying underneath something and so on. You just feel that.
A robot cannot do that; all they can do is look around as good as possible and visually determine their situation.
I naively assumed they have a gravity sensor, so will generally have an approximate up vector ?
Yeah but imagine yourself lying on the floor with your vision being your only sense, plus an info floating in your mind: „fyi, you are no longer upright“.
That’s all, you feel nothing else. Now your job is to move all parts of your body in just the right way.
Do I know which way my joints are bent (which a robot knows)? Then I can manage it.
And why don't I have any sense of pressure at all? We can put that into robots.
But I have more than that. I can definitely sense which way is up unless I'm underwater.
Or have an amusing inner ear infection. So OK, sure, it's vector, not a flag.
Theu have an IMU, what they don't generally have is the various aspects of touch.
The point about being aware of lying underneath some object was interesting. Sound might matter, like the frequency of background noise changes when you're in an enclosed space, and listening to your own shuffling noises helps you know when you've planted your feet right - or something. I have some really effective ear plugs and I notice they make it harder to move around.
Having said that, I've probably hit my head on the underside of an open cupboard door five or six times in my life, and I expect to do it again.
It is also things like "I can feel that my left knee is bearing a little too much weight, I should shift weight to my right hand and use that to push myself up" - things that come automatically to animals after learning the hard way in infancy (some of it is innate; baby animals are clumsy, but usually more mobile than human infants). Regardless of learned-vs-instinct, these abilities rely on sophisticated "sensors" and cognition. I suspect engineering the sensors is actually a bit harder, but I'm also not optimistic about a deep learning approach to the cognition.
A significant underappreciated advantage of animals over AI: lifeforms can "learn the hard way" more easily than 2020s robots because of cheap self-repair. AI labs are reluctant to damage their robots, but an essential part of humans learning to move safely is severely bonking your head and reckoning with the consequences - "hey, dummy, why did you trip and fall and bonk your head? Because you were running like an idiot."
I am learning the hard way to this day :) I have been practicing with work knives. A few months ago I got stupid and impatient, and sliced my thumb nastily. If I didn't block the cut with my thumbnail (still ruined) I might have chopped bone. It is hard to say precisely what I learned from this experience - "don't be stupid and impatient" is facile - but I know I learned a lot. I am actually optimistic about targeted surgical robotics. But for a general-use humanoid robot, I would not want to give it a knife if it's not capable of feeling pain. I never use big knives anywhere near my cats because I understand intuitively that they are nimble and unpredictable and easily stabbed by knives. I didn't need to be trained on this. A robot kind of does. Yikes.
I obsessively avoid any kind of "technology is going thataway" content. So I haven't seen anything that looks like humanoid progress in quite some time. About the only thing that has snuck around my barrier is Musk apparently claiming he'll have it by the end of the year, which is pretty conclusive evidence that they won't have it by the end of the year.
So if you're seeing anything that actually seems to merit attention, I'd love a few pointers. I could use some good news.
Well, as someone who has tried to build at least a couple small robot arms, I think we are probably closer to 20-50 years away. Both the power and dexterity are not there.
Right now, only a human can both push over a boulder and pick up a tiny speck from the floor using the same actuator.
Beware generalising from a carefully curated and presented set of demos to real life.
Mount 2 of these on a Segway and I can think of several tasks that could be automated where I work.
"a ChatGPT moment" doesnt seem very momentous. ChatGPT was surprisingly good compared to previous smaller models. But since then the LLM scene has just been insane amounts of hype and bullshit and financial skulduggery. Their actual utility is pretty niche imo.
If Figure acquires Eka they are so winning the humanoid race.
Anyone else here have happy memories of playing with Armatron? Circa 1984?
Apparently yes: https://news.ycombinator.com/item?id=43718493
Oh hell yea!
Just a few weeks ago at work we got a Universal Robots UR5 from another project in-house along with a Hand-E gripper.
I've never had so much fun programming and playing with a device ever. And it completely took me back to getting an Armatron 40 years ago and having so much fun - but also wishing I could somehow control it with software.
I still have mine sitting on a shelf in my office.
Yes! The most amazing part about those things was they achieved all those axis' of motion with one or two motors.
And the associated grinding noises were kinda scary but damn if the thing didn't hold up.
This one is different? What about unitree? What about their demo at the Spring Festival Gala?
https://www.youtube.com/watch?v=Ykiuz1ZdGBc
That sure felt "different".
No doubt hands are important, but I think you've missed a lot here Wired.
Many of the Chinese companies are doing very impressive open-loop sim2real. They make great demonstrations. They are not great at dealing with the real world and unpredictable environments.
(That's not true of all Chinese companies - some are doing really impressive work with closed loop systems in unpredictable environments. But many of the highly viewed ones with coordinated dance performances or martial arts are intended more as theater to government financial sponsors than useful function. The technically impressive performances do not look as visually impressive.)
those were impressive but were also RC. I think an important part of robotics is not just the mechanics of humanoid motion, but the independent control of those mechanics.
Can you expand on what was RC? Was the compute off device?
Back in the 90s, I developed a rule of thumb: if I saw it in Wired, it's because it was either already over, or it wasn't going to happen at all.
I was so disappointed when I saw BetterPlace (the car with replaceable batteries) on the cover of Wired. It seemed like such a good idea. Too bad the rule of thumb meant it wouldn't work.
Rules of thumb were made to be broken. Maybe this time it will be different.
Yeah, betterplace made it from 2008 (wired) to 2013 (bankruptcy.) Nio is trying again and it looks like they hit wired in 2018, again in 2023, and are still active today...
https://paulgraham.com/submarine.html
Do they mean, the moment when everyone realizes it's not as useful as they at first thought?
> Companies pay people to spend hours doing routine tasks with their hands while wearing cameras and motion-capture gloves.
Dystopian. Which companies out of interest?
no paywall: https://archive.is/Wro1e
archive.is is malicious -- as in, uses your browser to launch DDoS attacks, and other things.
Stop using it.
https://arstechnica.com/tech-policy/2026/02/wikipedia-bans-a...
Is the person behind archive.today the same operator as archive.is?
Yes, they have a number of domain names, archive.is and archive.today are the most well known ones.
Just run in the console window=null and you are good. It is valuable service until the websites get their shit together and finally fix their payments model.
> archive.is is malicious -- as in, uses your browser to launch DDoS attacks, and other things.
I think the attack was itself a response to a doxxing attempt. Also, archive.is being a free service doesn't quite fit with claiming they are malicious. The overall picture seems still positive.
I don't care what the attack is responding to, nor do I care what services are being provided.
If, when I visit your site, your site causes my browser to participate in a DDoS attack without my knowledge, your site is malicious.
If you didn't care about the service you wouldn't visit their website in the first place, in which case there is no problem.
Is there an alternative?
https://removepaywalls.com/https://www.wired.com/story/when-...
Works for me. I use only Tor so it is actually far more accessible. Archive.is uses Google's Recaptcha, which for some reason rejects valid solutions submitted via Tor.
I’m not sure that is always a valid CAPTCHA and not one being proxied to you for solving it on behalf of some bot (presumably a crawler).
I don't know. I think people would notice if Google were being MITM'd on Tor.
You don’t need to MITM it, this was a common pattern for a long time (not sure it still works though). There was no origin verification so you could just use a different site ID and have people respond to captchas you encountered on that site.
just because it's an article about techie stuff doesn't mean all the photojournalism has to be color-graded like a Matrix movie.
.. but it's kind of funny to read the fluff PR about saving humanity while juxtaposing it against photos that look like they may as well be screencaps from Prometheus or Black Mirror.
see : two startled victims under a blue arctic sun - https://media.wired.com/photos/69f11cbf1b1015e12f65d23e/mast...
> a ChatGPT moment for the physical world.
That's not a good thing, WIRED.
I'm having some house painting done and the painter asked me what line of work I was in. When I said computer programming he said, "ooh, bet you're worried about AI! At least painters are safe!"
He "might" be but not any of his kids going into the business. The home maintenance bots will invade slowly, then all at once.
I want Rosie (fictional robot from the TV show "The Jetsons")
Basically, I want a robotic butler / maid that will do most of the cleanup around the house.
Unfortunately, the only robots available will be connected to the cloud, paid by subscription, and will gather a continuous feed of audio-video data from you and your home. And sometimes it will be teleoperated, and you might not know when.
I'd rather do my own cleanup, personally.
Cloud connected (robot AI in cloud) home robots would be very unsafe, due to network slowdown/outages. Imagine it freezing/stopping right after it turns on water faucet or stovetop.
I bet China will race to the bottom with cheap versions. 3D printers and LLMs, next home robots
Why would consumers settle for that? Local models have scaled quite quickly. Just pair the bot with a LAN server as the brain that keeps all your data private.
Barring that, choose bots that use Zero Knowledge Proof architectures for all data so you know there's no in/out of personal data, only security proofs. This makes rental robots certifiably private too.
They've settled for that in:
* Phones * Cars * Robotic Vacuums * Kitchen Appliances * Televisions * Home Lighting * Home security systems, doorbells, and locks * Web browsers * Operating Systems
So, uh, yeah, I'm pretty confident users will settle for that in robots too.
Some of them will be paid by subscription and have ads
Haha! Instead, you’ll get a robot that will make you art, music, and tell you stories and you get to toil away cleaning the house.
“Sure I’ll clean up the house, Mr. J. While I’m doing, so have you seen the new shoes from crocs? They’re sponsored by the Jenners and have great new designs with all of your favorite movie characters on them! Would you like me to order you a pair?”
The first voice accessed (in my brain) by this dialog was Harley Quinn, it took a moment for it to fall back to Rosie.
“Hey Mr J for a low 7.99 a month I can unlock the Harley Quinn voice pack! For 39.99 we can upgrade you to unlock TikTok Rosie dance mode with special Harley Fortnite dance, a joker LCD breastplate for me and special “psycho partner” romance mode. What so you say, Mr J?”
I've spent ~$500 this month trying to get an LLM model to solve a Rubik's Cube. They can't. I'll post my Rubik's Cube MCP server next week if anyone wants to prove me wrong.
1. a human child learning 6 algorithms and a weekend can solve a Rubik's Cube
2. Reenforcement learning can solve a Rubik's Cube
3. The best LLM model using recursive tuning or not can't solve a Rubik's Cube.
Claude 4.6 got 60% of the way but couldn't figure out the last steps after running for 20+ minutes and hundreds of thousands of tokens.
I am not sure how to say it exactly, but right now we are in situation in which we are complaining that a magical technology is not magical enough.
No, we're complaining that a technology that's hyped as an expert-level replacement for humans is completely inept.
Given how many people attempted to date their computer after ChatGPT launched, I don't even want to imagine what this technology has in store.
People already pay shitton of money for silicon sex dolls and fantasize about robot sex online. Sex toys are connected to a network for remote control. As soon as a a humanoid robot becomes commercially accessible some will have sex with it.
God we are a horrible species.