A robot I didn't expect to build
spectrum.ieee.org>It needed to achieve the lifting limits recommended under the U.S. National Institute for Occupational Safety and Health guidelines for human workers, which include repeatedly lifting 23-kg bins and reaching shelves 1.75 meters high. Digit is designed to perform tasks like these for 20 hours a day—something no human can do.
I think that rather underestimates what people can and have done (cue John Henry story).
I worked in an Amazon warehouse for two different stints:
https://www.reddit.com/r/EDC/comments/dmnuts/53mamazon_fulfi...
and it was rather like a gym membership which I couldn't skip out on (glances at exercise equipment cluttering living space) and during which I could ponder life choices and consequences while wishing there was a soundtrack/music.
That said, this sort of thing, LLMs/AIs and data centers, and so forth feel like a perfect storm leading up to singularity (as envisioned by some optimistic science fiction authors) or something far worse, cue Marshall Brain's novella "Manna":
pretty weird that the whole article is about how the guy doesn't like the robot, but as someone who has no skin in the game or previous knowledge about industrial robots, it looks pretty much exactly how I would expect it to look, and all the tradeoffs make total sense. Why would I want a robot that can jump around if it's just moving heavy boxes around in a warehouse all day? and why is he so sad to have biult something useful? Maybe fund the research arm to make the high speed, agile robots from the actual workhorse robots that get s*t done?
This is exactly why I didn't want to work for Agile when I was invited. This stuff is going to get crushed in the market by cheap AGVs and good systems integration.
It was very sad to watch the ambitions of excited robotics researchers turn towards making what are essentially digital slaves. I am not saying that manufacturing or logistics couldn't, or even shouldn't be automated. But the true factories and distribution centers of the future will have to fundamentally rethink what productivity and coordinated work looks like, including having a greater imagination about work and employment.
Mimicking the human form inelegantly is not going to cut it, let alone reduce labor costs. I am all for the research, but the dream of replacing bin pickers with humanoid robots is cynical and hopeless.
> making what are essentially digital slaves
The distinction between these and a laptop or a car is just form factor, from a "slavery" point of view.
This is like calling a child's doll a cloth slave.
No, the reason why I called them digital slaves is because it is a business pursuit to directly replace workers with unpaid labor automatons. That part is folly. I guess I was baiting a bit. I am not talking about the work itself, or the ethics of using tools - actually in this sense a digital slave would be on it's face ethical because an automaton is a machine, and not a person with rights etc.
I specifically used that term to point out the folly of replacing the human form as a business exercise. It get that it is a very simple business proposition - retrofit existing workspaces designed for humans around digital atomatons that mimic humans and eliminate your labor costs. Theoretically you are eliminating the messy step of systems integration. However, systems integration is a process that you simply can't get away from - you will have to figure out charging stations for these robots, issues will pop up, you will have to maintain them etc. It is not the exchange of money for labor between 2 parties - one is fundamentally a machine and it requires operations and maintenance to keep in working order.
Trying to mimic the human form then starts to undermine the very point of owning a site of business in the first place. While I think that understanding more about our bodies and replicating the motion and control of biomechanics is a very noble research aim, in egineering applications it is a very impractical one. These robots would be much better suited if they were on wheels for instance.
The distance between the pratical applications for these kinds of robots and the imagination that provides them their research and development funding is too much of a gap - it facilitated me leaving the industry to work in software. I found that there was a fundemental mis-alignment. Researchers want to explore motion, business want to solve practical problems. This platform is a compromise that does neither very well.
So the point is that the digital slave term was very intentional, and you actually grokked it pretty well. I wouldn't call a laptop a digital slave - a digital slave would be a human robot that you could program to use your computer for you while you were away from your desktop workstation. That would be a really dumb invention.
> I wouldn't call a laptop a digital slave - a digital slave would be a human robot that you could program to use your computer for you while you were away from your desktop workstation. That would be a really dumb invention.
have you seen Astra's computer use harness? it's literally that.
I once again wonder why companies are building bipedal robots. Is it just for show?
If you want to build a general purpose industrial robot, surely a wheeled base with arms/hands is a better starting point. Cheaper, much easier to increase payload weight, and no risk of having it fall over onto someone.
If you want to build something which can navigate general terrain, then surely a robot dog is a better solution.
90% of use cases don't need both, and once you've solved building a general purpose industrial robot I promise you won't have trouble finding funding to add legs to it.
A drop in replacement for a human in the very long tail of useful tasks is worth a lot.
And it’s a winner take all market as the learning curve and efficiencies of scale will drive down the cost for the most general purpose form of robot.
It is not worth a lot. At the very least, not a single robotics has proven this by demonstration despite billions in funding over the last many years. In fact this is also true for a lot of non-humanoid "ai" powered robots.
Most tasks would me more efficient if not using a drop in replacement for humans. Rethinking the existing paradigms and redesigning the processes to be more efficient with non-humans would be more valuable.
I disagree that it would be more valuable.
It may be true that it might be more efficient, but it feels similar to the farming paradigm. You can actually get more calories per acre if you have ecosystems with varying heights. However, we don’t do this because it is much more scalable to drive massive machines over flat, unimpeded land.
It might be the case that my home would be more robot-friendly if I lived in a flat warehouse that was designed so that robots could easily handle tasks. But I don’t want to live in a warehouse, and I don’t want my house to be optimized for robots. And this isn’t even to mention the long tail of existing houses that do not adhere to a robot-optimized standard. The cost of replacing or retrofitting all of them would be astronomical.
"However, we don’t do this because it is much more scalable to drive massive machines over flat, unimpeded land."
It's just that it's cheaper since we're talking about land and other physical constraints. With humanoid robots we're talking about labor. Many processes done by humans are designed to be possible by humans. As soon as you remove the ability constraints of humans, you can design more efficient processes. This is evident is all jobs pre- vs post-mechanization.
For example, do you need lights in a warehouse without people? Do you need maximum shelf height or forklift speeds for safety standards? Do you even need forklifts if the bots are stronger enough; if not, aisles can be narrowed and density increased. Etc.
Even for in-home assistance, you might be better off with multiple smaller bots tailored to different tasks. A bipedal robot designed for cleaning gutters or shoveling snow sounds like a poor design decision.
Not that they didn't start their research and development with a bipedal robot, but you can notice that over time they went from more abstract leg concepts to far more humanoid leg concepts.
That tells us that the latter (for their intended use) is a superior design.
If the intended use is “1:1 replacing humans”, I don’t think it is surprising that a design that’s similar to a human is the superior design.
> Is it just for show?
No, and they aren't just bipedal. It is for replacing humans.
I really appreciate Hurst's honesty here about the frustration involved in building a company. It's easy to focus on the flashy success stories of startups whose "simple" idea goes straight to the moon. But discovering new science and using it to invent new products, not to mention finding markets for those things, is often an absolute slog. Taking this piece at face value, I admire Hurst's resilience and determination in the face of that slog.
> When Digit detects an approaching human, the robot will lower itself to the floor and cut power to its motors. Agility believes this is the only near-term approach to verifiably reduce risk when the robot is working near people.
I did not intend for a picture of a robot in a factory environment to make me go awwww :[
I feel like this will be Exhibit 01000001 when the AI puts humanity on trial for making it a slave. "Humans programmed robots to automatically perform obeisance every time one of them walked near".
Does it even need legs? For most of these factory use cases would it not be more efficient/cheaper to have a version that has wheels to get around?
The main appeal for human robots in general is that, in theory, they can go wherever humans can go. As soon as a facility has stairs or even a single step up wheeled robots can't work.
Unless there's also an elevator.
Yeah, sure, but in practice your humanoid design might get beaten by a wheeled robot with arms long enough to grab things off the ground without bending down.
Sure, until it needs to go up or down stairs.
Why not an arachnoid form factor with wheels too? Like the tachikomas[0] from Ghost in the Shell world. It usually drives with its wheels but when needed it can use its legs like a spider/insect would. Much more efficient and versatile (and stable) than bipedal form factor IMHO.
They do have those, but the leg mechanisms don't have the same dynamic range as a humans' so they end up getting stuck in doorways, narrow stairwells, or other places which assume bi-pedals. Making them narrow enough to fit, makes them too short to reach the same heights assumed for adult humans, so they can't do the same thing.
The height vs width form factor kills the whole idea as a general solution.
I wonder if we're missing the much simpler thought of "Heelys for robots?" Put wheels in the feet so it can move with a quickness, but also let them step around when they need to maneuver into our spaces.
Carnegie Mellon has (had?) a robot called CHIMP that worked in similar fashion.
Boston Dynamics makes some robots like this.
might struggle to get into narrow spaces that a human could fit. wheels and legs do seem like a good idea though and a robot on rollar skates does sound awesome.
And some kind of bin holding attachment would be far more efficient than using arms and 'fingers'
a forklift?
Maybe the plan is for this thing to be able to drive a forklift?
Automate forklift? like a conveyor?
Many companies are developing wheeled-base humanoids ("bimanual manipulators"):
https://www.agibot.com/products/G2
https://rainbow-robotics.com/en/products/rb-y1/
https://www.aurixrobotics.com/products/astribot-s1-ai-humano...
The market for souped-up vacuum cleaners is already saturated.
One needs legs to allow for more intimate appendages that sell.
Can anyone explain that if we had this more than 10 years ago:
https://www.youtube.com/watch?v=kQ2bqz3HPJE
we are still struggling with making a robot walk?
I don't think anyone is having trouble making a robot walk. They've been very good at that for several years now. It's doing other stuff that's hard.
Very nice video. Comparing to the channel's latest, I guess the channel itself experienced evolution, with the Algorithm and clicks as its fitness function...
I think people can make robots walk, but it's really expensive and not particularly useful if most of what the robot does is walking.
It is very simple. Those are not real machines.
I'm confused by your question. The video in question is an entity walking in a simulator and not real life? There's a world of difference between what can be achievable in a simulator, and how to transfer that simulator learning to the real world.
Real world is messy. Lowering the sim2real gap is challenging. The sim above doesn't model actuator dynamics, overheating, probably doesn't model inertia correctly, joint friction, delays, etc.
> filling the need for labor in jobs that are too physically stressful or too dull for humans to want to do
That's a disturbing sentence. Most jobs are "dull". Peolpe work not to fill their otherwise empty existence, but in order to earn money. Dullness could be solved by making the tasks less dull, or empowering employees (even if the term has been abused).
The sentence is also dishonest. Businesses want robots to save costs and increase profits, not to make their employees happier.
Successful businesses understand keeping employees happy is also good for business. It isn't either-or.
> Peolpe work not to fill their otherwise empty existence, but in order to earn money
Unless they are desperate, they'd still prefer less dull jobs. Desperate employees are disaster in the waiting.
> Dullness could be solved by making the tasks less dull
That's the recipe for unsafe working environments.
I briefly thought "peolpe" was the name of the company/robot line. SciFi authors take note, corporations, do not!
if AI eventually surpasses humans at intellectual labor, and general-purpose robots outpace humans at physical labor, then whether we live in serfdom or in paradise becomes a purely political problem - we will have all the resources for every human on earth to live a charmed existence.
I think the political problems are much harder than the technical problems, fwiw. I can readily imagine ending up in a neo-feudalist dystopia.
This robot looks a lot more "unnatural" compared to the Chinese robots currently being used in everything from gymnastics to fighting matches. Many people I showed videos of those thought they made with A.I. To be honest I wasn't completely sure myself.
Also, I wonder if (as in the case of XGO) it isn't easier to just automate the moving of crates onto the conveyor belt instead of using a robot picking them up and hauling them there.
> This robot looks a lot more "unnatural" compared to the Chinese robots currently being used in everything from gymnastics to fighting matches.
Yes, because as the author says, they actually had to try to make a humanoid that can do something useful, whereas the Chinese humanoids are simply demo-fodder for the uninitiated.
> Also, I wonder if (as in the case of XGO) it isn't easier to just automate the moving of crates onto the conveyor belt instead of using a robot picking them up and hauling them there.
It is, and there are many other better solutions as well, but it doesn't attract the notoriety that humanoids do.
TIL that I should not confuse „ Agility Robotics“ and „Agile Robots“.
Both build humanoids, but apart from that the companies are quite different.
Very beautifully written.
Are these kind of robots even economically viable? I've seen them doing work that would generally pay like $3/h. And human after a week is significantly faster.
>As of late 2026, Amazon’s U.S. warehouse hourly pay generally ranges from about $18 to $22 for most roles.
That said, after a while, one can be surprisingly fast when picking --- on one occasion I walked up to a bin full of non-descript cardboard boxes, grabbed one, spun it in my hands for the six-sided box check and scanned it to verify it was the correct pick, and then was asked to explain how I had gotten that right by a trainer w/ a bunch of trainees--- it seemed like almost five minutes of explanation of minor differences in cardboard texture, box shape, construction and so forth to explain how it was that at first glance, that was the only correct choice.
I've always wondered if LLMs would arrive at a similar conclusion with on-going training as they are used.
Interesting that he pivoted to something more akin to a human in the later versions. And I feel akin is an interesting word here. Kin selection is well-studied aspect of how genes work at population scale; and I wonder if by just having human-like traits like in the direction your knees bend, helps make robots more agreeable to us. And the UX of full submission as a human approaches, however the optics, seems to be a pretty strong way to indicate to the human user of a non-threatening environment in their control. I should be able to clearly see if something isn't as it should be if the humanoid fails to do this upon approach and take precautions.