Unitree G1 Humanoid Agent
unitree.com
unitree.com
Even worse, and likely illegal(!), the warranty has terms that are likely against FTC guidelines, like voiding the warranty if you disassemble the robot. As far as I know, these type of terms are not valid. They also include terms saying your warranty is void if third party service or parts are used - these are also not valid under the Magnuson-Moss Warranty Act.
https://unitree.com/terms/policy/
https://www.ftc.gov/business-guidance/blog/2022/07/ftc-annou...
https://consumer.ftc.gov/consumer-alerts/2022/07/ftc-says-co...
https://www.ftc.gov/business-guidance/resources/businesspers...
https://www.ftc.gov/legal-library/browse/statutes/magnuson-m...
https://en.wikipedia.org/wiki/Magnuson%E2%80%93Moss_Warranty...
Also I’m not familiar with USA laws in particular, but often domestic consumer protection laws surrounding warranties etc do not apply to overseas sellers.
shop.unitree.com has options for shipping to EU countries so yes, they sell directly to consumers in EU.
Generally, offering a shipping service that accepts an address in EU doesn't mean anything. Practically any shipping service can arrange delivery anywhere in the world today, even national posts. When I buy something on the internet (and I do this very frequently), the seller uses their local national post and my national post delivers it - but it goes through the international arrivals center where I have to pay duty and VAT.
They would have to incorporate a local branch in EU, pay import duty and take VAT from you for it to be considered a local sale according to EU consumer laws. You can check whether that's the case in Terms of Services - it should be clear who is the seller from that. If it's not an address in EU, you are self importing.
If the sale transaction is with the Chinese entity, and the goods are then shipped to the EU, EU protections won't apply (will fall under Chinese consumer protection laws instead).
They are also almost impossible to self-repair & service beyond changing motors & batteries. And spare parts are not sold, at least not through the web store: https://shop.unitree.com/collections/accessories
It could change in the future, but for now, it's best to think of Unitree products as very advanced toys & prototypes.
We are really in the early days. And this $16K humanoid robot is by far (like, 5x-10x) cheaper than other alternatives on the market. I might even buy one, despite it not being repairable or serviceable.
Yea that's the joke. They can just claim a tiny scratch is why the robot is broken and void the warranty that way. Standard practice these days.
What is this, an ASUS product? lol
Hopefully we get some open source designs as well, and open collaboration in the engineering community.
And hopefully humanoid robots don't turn out like most "smart" devices, where all the guts are closed-source binary blobs the user has no control over. (I'm sorry Dave, I'm afraid I can't do that). Not to mention the consequences of having your humanoid robot hacked remotely could be quite horrific...
With that said, it is hard to imagine the economics of humanoid robots. The trend seems to favor automation in controlled environments for certain products. Then the displaced factory workers flood remaining jobs driving down labor costs and shutting down any prospect of a humanoid machine.
The point where it all changes is when the humanoid robot can repair itself / make a copy of itself.
That's a physical compiler that can bootstrap itself.
That's a genie that lets you wish for more wishes.
What that does to the economy is anybody's guess.
When we'll make one and when the average joe will be able to get their hands on one are some interesting questions.
Once people know it's possible with off the shelf materials they'll soon seek to replicate it, so it can't remain controlled unless it's hidden, because they will succeed in that attempt to replicate it. Who wouldn't?
What would the average joe do with a humanoid robot that can do most domestic and industrial tasks as well as self replicate?
There's not really such a thing (even conceptually) as a humanoid robot that can "self replicate" without a broader robot factory with additional machinery. When you ask "what would the average joe do with a humanoid that can ... self replicate" this sort of misunderstands how robots will be built. Certainly the average joe could ask the robot to go build more if the average joe also had access to a robot factory with CNC machines, metal 3D printers, plastic injection molding machines, PCB fabrication equipment, etc etc. But additional machinery will always be needed.
People in the 3D printer world got really excited about self replicating machines, but motors, cable harnesses, PCBs with lots of different chips on them, metal housings etc still cannot be 3D printed on such machines.
If the average joe has his own robot factory, he's not the average joe. It is theoretically possible that all these different machines will be collapsed in to some sort of single machine - it's hard to say what will happen in 100 years or more - but that's not really anything I would get too worked up about right now.
If you had a pair of two of these robots and your run of the mill warehouse full of their spare parts (assume a tape of infinite length...) so that with careful planning between the two of them they could do the necessary preventative maintenance on each other so that they didn't break down at the same time, or could go rescue the other if it broke down unexpectedly in the field, how long would it take them to build the infrastructure necessary to produce new parts to replenish their stockpile and generate new robots so that they could experience exponential growth?
When I watch stuff like Primitive Technology[1] I'm in awe at how asingle well-fed and educated human being can with the right preparation and research can go into a jungle setting and speed run up to the iron age. It makes me wonder about the minimum viable number of people and education/knowledge/experience required to do the same thing for realsies to the computer age. It's just a pipe dream or two about when a robot equivalent to the average joe is a thing and when I see videos like the one we're talking about or the recent one from Boston Dynamics the gears in my mind start turning.
There was a time when computers didn't exist. Then they did. At first computers used to be the size of buildings, then they became the size of floors of buildings, then the size of rooms, then the size of appliances[2], and so on and so forth. we're seeing this sort of miniaturization across segments of industry driven in large part by the miniaturization of semiconductors. If you take a big picture view the economy as a whole it is a self replicating machine. As are the vertiable Adam and Eve that made that economy. In a complicated sort of orobourous way the entire economy is made up of humanoids (and let's not forget our quadroped hoven friends who serve as feedstock btw, or chickens and tuna and so on...) who replicate in some way or another.
There is this sort of impending collision between the organic and inorganic self replicating aspects of our economic system. Ribozyme and hominid, compiler and automobile assembly line, what are the difference exactly?
[0] https://www.gettyimages.ca/detail/illustration/doctor-workin... [1] https://www.youtube.com/channel/UCAL3JXZSzSm8AlZyD3nQdBA [2] https://media.cnn.com/api/v1/images/stellar/prod/11101401564...
High density data storage would probably also be an issue. And I'm not sure about batteries - I don't know how much proprietary tech is required to make the difference between a basic functional lithium battery versus one that actually stores enough energy to be practical.
A very very long time if they have access to all the mines and chemical sources needed, never if they don't.
And they will need space to build factories.
Primitive Technology can get you to the iron age, sure, but that's scrap quality. That's the INPUT to our current smelting chain. To make stainless steel you're going to need CO2 and Argon and Nickel and Chromium. That's just one of the structural materials needed, and it's one of the easiest! Polymers are going to be the real challenge...
A bigger problem is gonna be chips. Yes, chips are really small these days. Fabs are not. Fabs take billions of dollars and millions of man-hours to construct. Your robots which are accurate on the scale of millimeters (maybe), are going to need to produce and place optics to nanometer levels of precision (mm->nm is on the same scale as km->mm).
On top of all that, you need to confirm that the thing you're making is the thing you intended to, so you need to measure a bunch of stuff! You're going to need interferometers, CMMs, spectroscopes, electrical test equipment, etc. You can buy all that, sure, but now we're back in the billions of dollars and millions of man-hours territory.
There is a lot happening here. Really big jumps. Sure, humanoid robots with lots of spare parts could repair each other. You could also just imagine a robot that never broke down for this thought experiment. Still, going from nothing to "can build more complex humanoid robots", even with very persistent workers who never get tired, is an extreme endeavor.
I suppose some day the equipment to build a humanoid could be cheap and accessible to home fabricators. But that equipment still will not just be "a humanoid", it will be specialized machinery. You could certainly have humanoids operate that machine, but I still see this scenario as "imagine you have two humanoids. now imagine you have two humanoids and a humanoid manufacturing machine. they could make infinite humanoids". There is no getting around the fact that a machine that makes humanoids will be a separate machine.
Will it be interesting when that happens? Yes, absolutely.
I think the market for consumer humanoid robots would be similar to that of the consumer automotive market. I’d certainly pay the cost of a midrange car to never have to do housework/yardwork again.
And that’s not including industrial use cases.
it is hard to imagine the economics of humanoid robots
What department of every municipality in the United States would love to get their hands on these?
Law enforcement will be chomping at the bit to get these on the budget. Learned repetitive movement, taking punches, crushing "walnuts", easily packed away in a trunk..
There is an enormous amount of work to be done, we aren’t running out.
We're already there with GPT, why would you expect anything else from a robotics company?
Hacked or not, I would be suspicious of anything closed source and connected to the Internet, especially if it has cameras, mikes, can roam around and possibly access to the local network too. This robot is insanely cool, but to me the chances that it contains code that phones home to do more than simply upgrading it are too high, like in every closed device.
I like that it is short. Humanoid robots should be short. Shorter is lighter, cheaper, less dangerous. If it needs to reach something high it can use a stepladder or whatever.
I doubt the basic model is what is shown in the demonstration, and the other model is priced at "contact sales" which usually means "do much that we don't want to say publicly as it'll put people off if we've not got them drunk, or otherwise stupified, somehow first".
We’re going to witness an explosion of affordable humanoid robots in every home & business, and as they become increasingly capable the use cases will be nearly limitless. Think deliveries, cleaning, even childcare (Bicentennial Man, anyone?). I think what will be shocking is just how fast this happens, and they’ll likely outnumber humans at some point.
https://mobile-aloha.github.io/
There was this video last week of a "westworld" factory in China too: https://newatlas.com/robotics/chinese-humanoid-robots-realis...
It looks like they have a direct-drive motor that can operate joints with no gears.[1] Is that right? If they use all direct drive motors, these robots are mechanically simple. No gears, no strings, no pulleys, maybe no separate bearings.
A production line for a custom motor is expensive to set up, but once production is running, each motor is cheap. An encoder is cheap if designed into the motor. This is the route to volume production.
Machine learning has made the control problem much easier. The classic approach was to work out the dynamics analytically. (I used to struggle with that in the 1990s). That's the old Boston Dynamics approach. Now that's obsolete. BD must be using some machine learning by now. The video of Atlas doing a flip showed the planning process briefly. It was very pre-planned. They must be beyond that by now.
With good temperature monitoring and over-speced MOSFETS, those direct drive motors can probably be way overdriven for a few hundred milliseconds. That's probably the future. You don't need that extra power all that often, but sometimes you do.
Good to know.
For the first year, after that society will collapse as the largely now unemployed low income and middle class arm up and start pillaging for survival.
$16k is what a wealthy person may spend on a vacation without a second thought. Amazing.
Now, you could argue in favor that Cloud makes some revenue beyond ads; but are a third tier offering that is only there because ads cash was tossed into a furnace. They are “profitable” now, but there’s no way they’ve made back the money they burned and they won’t for a while.
None of the robotics teams at Google with ever produce anything useful in the market and are pretty much on the chopping block as it is.
From the makers of "founder of #1 rocket company and #1 ev company just got lucky"
Robotics is the next frontier of war since our elites can't find enough idiots to kill/die for them. It's in a bit of equillibrium now since China has the upper hand in Hardware; US in Software. However, at this point, Robotics will rapidly become a natsec issue.
Pray for countries who can't afford R&D in this - they'll get genocided out like flies overnight, if they don't turn themselves over into slaves. We're looking at a the war-robot race by the end of this decade (legged robotics have made proven advances in the past 5y).
Governments will have to face the question whether they want the productivity of robots, while losing their monopoly on violence. I think it's inevitable that robots will win, and governments with their police and military forces will become irrelevant. Some governments might try to fight this revolution, and some will accept it quicker.
Also, there's really no need for governments anyway when robots are there to serve humans. It doesn't make sense to collect taxes from robots and then pay other robots to work with those taxes.
https://community.twistedfields.com/t/january-2023-update-ne...
You will run into energy issues very fast with this as soon as you do anything reasonable with it.
Like removing weeds, watering, fertilize, ...
It's been a while, I would like to post a new update. Other engineering work keeps taking priority, but I think I will start writing a new update later this week.
> You will run into energy issues very fast with this as soon as you do anything reasonable with it.
Our goal is specifically to research how to accomplish farming in our roughly 1kw work envelope. I personally believe this is achievable, it just takes a different approach to tools. We will build new custom tools for this. One thing to note is that since the vehicle is autonomous (there is no driver) it can move very slowly if this helps conserve energy. Tractors traditionally use a lot of power to save the driver's time, which is not a constraint for us.
> Like removing weeds, watering, fertilize, ...
I don't see why removing weeds would require more than 1kw. Watering is usually done with fixed pipes laid at the start of the season, rather than machines driving around transporting water. Fertilizing again doesn't require much energy. Our vehicle only draws 60-100 watts while driving on level ground. That leaves plenty of energy for distributing fertilizer (though we are specifically targeting regenerative organic, where fertilizer is not used).
That seems restrictive given the argument that liquified worm castings are organic: https://www.abc.net.au/news/2024-03-21/worm-healthy-benefits...
(and similarly that pig waste (urine+feces+time) from pig farms makes good starting soil .. although there are understandable objections to pig farms and to shit in dirt when growing some types of food).
Is "regenerative organic" that tightly insistant upon an immediate closed loop that not even an adjacent worm farm can be mixed in?
I'm curious when you will do those tests what the outcome is.
At least when I look at it it feels that the energy limit and tire size might be a problem but of course this depends highly on the size.of the field.
How much energy do you currently use when moving with things? Or while doing things?
My assumption when weeding is that you drag something heavy around or do it with laser.
Water I would not use dripping due to micro plastics and cost.
I would also fertilize to a certain degree. Even organic has to fertilize.
For weeding, the robot is heavy enough that if you were using a drag tool you wouldn't need extra weight. We might use a robot arm for weeding, not sure. However one of the tricks in regenerative organic is to prepare your soil in such a way as to avoid the need for much weeding during the crop growth cycle.
When I say "fertilizer is not used" I mean industrially-derived chemical fertilizer is not used. We would still need to spread compost and manure, which can be heavy! That would be one of the more energy intensive tasks. Who knows, maybe we will find we need a battery for that task. This would be easy to add. I think many tasks will not require a battery.
It sounds like you're interested in the farming process. Here is a quick overview of regenerative organic farming systems: https://www.youtube.com/watch?v=Ct3CL22RpTg
And this is a really nice detailed lecture series on a regenerative organic farming system: https://www.youtube.com/playlist?list=PLCeA6DzL9P4uRadXW0_hj...
In the second link, they use a two wheeled walk behind tractor that is I think 18 horsepower. We will have 1 to 1.5 horsepower effectively, so we will have to re-think some of the tools, but you can see how moving slower with no driver could get us in the ballpark.
Remember, farming can be accomplished by a single human with basic hand tools, and our robot has more power available to it than a single human can output continuously. Of course, humans are a lot smarter which certainly helps, but you can begin to see my reasoning for why I believe our approach is reasonable.
Also watering is usually done with impact sprinklers:
https://en.wikipedia.org/wiki/Impact_sprinkler
Cheers!
I will buy a small farm this year (if I'm lucky) and I do believe your general approach is very interesting.
Let see how it continues.
Regarding sprinkler etc. I saw a bigger drown spraying stuff. If the autonomy is very reliable you could refill it regularly.
Sprinkler I'm really not convinced but we don't have a lot of sprinklers in field in Germany.
Might be a little bit waist full from a water perspective
Hell, in 2008 the cheapest 3D printer was $25,000. Now a cheap 3D printer is $200.
Also, there would not have been any use for a humanoid robot in 2008. We didn't have the software to understand the world. Note for example this XKCD comic and the notion that "identify whether a photo is of a bird" was considered "virtually impossible".
General Electric built a walking truck in 1965, where each actuator was manually controlled by a human.
https://en.wikipedia.org/wiki/Walking_Truck
Robotics has generally been held back by software constraints. Deep neural nets didn't get their start until 2012. In 2008, there would have been no market for a humanoid. This one is being sold as an "agent", a term in this context used to describe something acting based on deep neural networks. That's a viable research area now, but was not in 2008.
Honda debuted Asimo in 2000, so the tech was possible. There were just limited uses and the software was really complex.
1: https://www.youtube.com/watch?v=FEXSqsW6rMM
2: https://www.youtube.com/watch?v=g0TaYhjpOfo
3: https://palm-e.github.io/videos/meta/planning_4x_compressed....
Looks quite a bit smaller in person (size of a 8-10y old).
Make sure to watch the video.
Given how much capital is being invested right now, and how quickly things are improving, it's hard not to imagine cheap robots taking on many jobs that until now have been done by human beings. Much like water always migrates to the lowest accessible point in a lake, jobs always migrate to the lowest-cost option in a market economy.
Like it or not, the robot future is almost here.
Having a humanoid in a factory doing a repetitive task it was trained for is one thing, but having one at home and telling it "go clean up the kids bedroom", or "go put the laundry away", is probably still 20 years in the future.
I seriously doubt we'll ever see humanoid form factor robots used in a domestic setting, and even in a factory setting the form factor is basically a gimmick. Why does a factory robot need (or want!) legs when an upper torso on a wheeled base would do the job better?!
All that came to be are small scale testing in couple US cities, and steering wheel stabilization mini computers for mid-to-premium range cars that aren't that better than craft-built implementations from back then.
Old old me sees this as something to be scared of.
2. This one is produced in China.
We can benefit much from robotics but humanoid for at current state of tech is honestly not much useful. We can benefit from cleaning robots, no need to be humanoid, we can benefit from autonomous wheelbarrows, 4-axis storage assistants and so on, sure they are far less fascinating but they are useful and they can have reasonable price tags TODAY.
The demo with the walnuts and coke sure looks like CGI.
So this is like a kickstarter where they promise the world using CGI/mockups and then maybe deliver?
Unitree already makes a larger humanoid called H1 that has shipped to customers, and they have manufactured cheap robot dogs in volume for many years. They are a legit company. The only thing questionable here is the price, it'll probably be more than quoted. But it'll still be cheap for a humanoid.
The videos of it moving that I have seen, and I did some Twitter searches, are all CGI.
I have no doubt they will sell a robot, but it isn't clear it will move like in robot the CGI videos or have those capabilities.
Last year, I had a chance to see Unitree H1 in action, in person. It's their previous model, a bigger one. Still, it was very real, very capable and very scary. I suppose, the choice to make a smaller G1 is in part to make it less dangerous, both perceptually and because it's less mass and energy.
https://www.unitree.com/images/Unitree%20G1%20EN%201080p%20%...
I do however find it amusing that it is also apparently a requirement for a humanoid robot to stand up in the most uncanny way possible.
It's much easier to make an 'animatronic' robot which has pre-programmed motions and can't withstand mechanical impacts.
[1] https://www.espn.com/mma/ufc/story/_/id/38827608/art-jimmers...
A) Real, straight-up, what-it-says-on-the-tin
or
B) Satire. I mean, who wouldn't want a robot that opens walnuts by smashing them with its fist? What a great party trick! It for sure won't accidentally classify your pet gerbil's head as a walnut, don't be silly.
A big problem is the lack of innovation in touch sensing. We mass planar (sometimes flexible) products with millions of little active devices (OLED). Cannot be this hard to assemble a decent touch sensing planar device. It will be like a camera for touch and then all the beautiful neural net techniques can be used.
1. No way am I linking that.
I love how fast they are dropping prices.
Sub $10k and if robot is useful/general, they’ll be doing millions of units.
1B+ humanoids by 2040 has a decent chance of happening.