This is way more of what I thought robotics companies should be working on over their dumb 4 legged dog like robots.
Wheels were invented for a reason and we are awfully good at moving things around on them at speed. Combining multiple attributes into a single form will ultimately prove to be the winning formula.
Just like in a year or two you will start to see more appendages than just two arms on humanoids.
A hub motor for each leg adds some weight, but you get a rolling mode on flat terrain with great range.
At the same time you keep obstacle climbing mode with stationary wheels, and can use the leg actuators as shock absorbers and obstacle avoidance on a bumpy path. Best of both worlds, with a new software challenge.
Literally we have a very old saying that is "don't reinvent the wheel" and yet people persist in doing so, when an already near optimal design exists.
I'm not saying don't do anything new and novel but that if a large portion of the design has already been done for you and you can improve it, like this company has done (wheels on the end of legs!), why not try that first and see if it works before starting over.
There are so many places wheels can't go and legs can. For example, most rough terrain is right out of the question for anyone using a wheelchair. There are all-terrain wheelchairs that can go over some rough terrain but you just can't move across the same range of terrains with wheels that you can with legs. Even Mars rovers have to carefully avoid areas of e.g. deep sand where they can get stuck etc (and they still get stuck occasionally like Spirit†). Hence also all the specialised wheel designs like tank treads.
If it were possible to make legged robots with the same fine motor control and dexterity as human legs, let alone four-legged animals' legs, we'd be way ahead of where we are right now in mobile robotics. We can't.
1. We need the capability to build the servos, actuators, etc. Supply chain, factories, specialized vendors.
2. We need to dump the firmware, decompile it, extract the models, and liberate it all. AI will help immensely with that. We can literally hijack everything software that isn't delivered over cloud now - so this is the easy part. We can then fine tune, adapt to new tasks, etc. Grabbing the brain = easy.
So the hard part is the hardware itself. The brain will belong to us soon enough.
Robotics distillation will be a big thing. You won't even have to farm out billions of requests across residential gateways. You just need to buy one unit.
You can't stop the software from becoming open.
Startup idea: AI model to dump the firmware from DJI, Unitree, etc. etc. and turn it into human-readable code. Broad utility across every industry with tons of willing buyers.
You can create a bounty through a nonprofit foundation, and use donations that are tax deductible to fund reverse engineering and creating an open version under an open license that defends against commercial capture.
This is one of the superpowers of LLMs and generative AI I think, providing cheap reverse engineering superpowers to clean room reimplement proprietary and closed systems as open source systems. Fiat, tokens, and context goes in, technology sovereignty comes out. In the same way people talk about a software factory, think about a reverse engineering factory. You are laundering the knowledge and tradecraft embodied in artifacts, both digital and physical, through model training and inference.
Open hardware developed by the above process can be produced by contract hardware manufacturers or someone CrowdSupply facilitates. If an expert opinion is needed on this, I’d recommend Andrew Huang of bunnie studios.
E.g. Unitree R1 has attractive low price but missing the “secondary development” pack (access to Jetson Orin and all the firmware stuff to let you run custom software). It’s all just their software. Useless tool. Click button and it dances. Then what.
I've seen a similar setup with a different Chinese cutting edge product. App access is free, ssh is open, but API keys and the ssh password is priced like a cheap car.
I imagine it ought to be possible to use computer vision to augment the traction control system, all the way up to locking the wheels altogether and using them as feet wherever necessary.
Even if the AS2-W isn't capable of doing that today, it's a fairly obvious goal for future development given how generally useful wheels are otherwise.
I love 4 leg robots with wheels because it is a form of locomotion that evolution didn't create and seems to be wildly effective with the wheels allowing for very fast movement on smooth surfaces and the legs acting like a suspension with huge travel range on rough surfaces and then it can switch to walking mode with the wheels acting like ankles. And they can step over obstacles or step up to higher surfaces.
Here -and with every product promo video - you are watching curated shots. Non-flattering footage simply will get deleted during edit, except perhaps for the blooper-reel, if that's the company coldly calculates to present a goofy, non-threatening vibe.
If I understand your question correctly, I think you may be missing that it can lock the wheels so they don’t roll sometimes like when it’s climbing uphill.
Looks like right now. They had an image of it with three small Asian men standing on it. I believe I’ve seen a previous video of this with a guy actually riding it. It should still be on their YouTube channel.
Chinese work ethic driven by overproduction of STEM majors leading to 25% youth unemployment...also they probably stole a lot of the fundamental stuff that was super hard to produce but easy to copy and build on top of.
On the other side: Years of VC focusing on the next great app to make you some cooked eggs instead of investing in companies operating in the physical world.
Their earlier robots (A1, Go 1) have a lot of lineage from Ben Kats' (from MIT) Mini Cheetah project, even down to the shape of the communication protocols used for motor control. I've not had the chance to closely inspect their newer robots, so I can't tell you what's their secret sauce, but I suspect it has as much to do with RL-based control and software as it does with hardware improvements.
I often hear about "muh china stealing", two questions come to mind:
1. how could they steal something that western companies dont even have yet?
2. for more fundamental stuff, western companies did transfer knowledge themselves by moving factories to China, in exchange for decades of access to China market and exorbitant profits they earned. Where exactly is "supposed stealing" in unitree ?
Boston Dynamics has been at the game for a LONG time (1992 vs Unitree's 2016). During that time they slogged through a lot of fundamental work that from the outside looked like 0 but thats the thing about tech, it starts at 0.01% then increases 0.01% → 0.02% → 0.04% → 0.08% → 0.16% → 0.32% → 0.64% and so on. The world just sees 0 until they don't. If you are able to just take all the results of those years and skip the same effort required then combined with your own skilled and motivated people you can leapfrog them.
(Im basically rewording my original comment, you didn't get my point the first time)
The answer to both your questions, and the poster's obvious implication, is that somehow they stole information from Boston Dynamics. (I'm not saying I believe that.) Seems like they must've stolen the information at the exact point in time when advances in AI made these physical robots much more easy for anyone to engineer without theft...
US patents do not have extra-territorial power, meaning US patent means nothing outside its borders, unless there are mutual reciprocity agreements (mostly between US and its vassal puppet states).
To my knowledge, China has never been caught infringing on US patents on US soil. A lot of Chinese innovations are simply orders of magnitude better than any US alternative, for example TikTok's algorithm and TikTok as a short-video social network.
So please answer my question, how could China steal anything that either 1) Western companies explicitly granted as part of Joint Venture corporation, 2) Western companies havent even developed themselves, and at best, only possess inferior version, generation or two behind the state of the art?
in IP realm, India for example forcefully nationalizes all pharmaceutical IP and produces generics of brand name drugs for its internal market (1970 Patents Act)
All Chinese companies are 996 companies and any that arent will eventually be replaced by a 996 company. Again vast oversupply of STEM labor serves as a forcing function(but is not the only reason).
You say that now, but wait until one of these is charging at you across a field with a QBZ-95 strapped to its back and a 1024 resolution thermal sensor.
Not this robot. You can see that none of the videos where the robot is careering downhill or rolling fast over rough terrain shows it carrying any load at all. Since those are the best-of, carefully curated videos, painstakingly chosen to show off the system's capabilities at its best you can rest assured that the robot can't manoeuvre like shown in the videos with any load at all.
There's a video where it's going uphill over uneven terrain with a load at a much slower pace, which is the same as what Big Dog could do 21 years ago, and another one where it's shown rolling up a smooth, grassy slope while loaded, which is pretty meh.
And that's just carrying a load. Now, you want it to carry an assault rifle and fire it? Like, at enemy targets? And hit them?
You will need to wait for a couple of major breakthroughs in robotics and AI still, I'm afraid.
I'd be more scared of MQ-9 dropping a JDAM or R9X on my head than this thing running around. They're proven to be undetectable to human targets. That's scarier to me than this thing running around catching wires and bushes.
I’m just a regular person and even I can see use for this where it carries all my gear on a very long hike or when I go camping. I’m sure that alone would be extremely valuable to any military. Also, not everything is aerially accessible. Lots of tight spaces in urban combat where you need to get get inside structures and interact with physical things like doors that drones just don’t have the physical strength for and probably never will.
16 kgs for two hours max is not "extremely valuable to any military". It's hardly even valuable to a leisure camper. You might be able to get this to carry some of your stuff if you go on a short hike, but if you were thinking of going camping and having it carry your camping gear, you can forget about it.
is using a prototype firmware or hardware platform faking it when you're showing exclusive prototype capabilities and trying to pass them off as reproducible by the production unit?
The 5-10k$ price is for their consumer-grade robots like the Go 2 or low-end humanoids. I won't be surprised if this monster costs 50,000$ or even more
This is way more of what I thought robotics companies should be working on over their dumb 4 legged dog like robots. Wheels were invented for a reason and we are awfully good at moving things around on them at speed. Combining multiple attributes into a single form will ultimately prove to be the winning formula.
Just like in a year or two you will start to see more appendages than just two arms on humanoids.
I think this is a clever idea.
A hub motor for each leg adds some weight, but you get a rolling mode on flat terrain with great range.
At the same time you keep obstacle climbing mode with stationary wheels, and can use the leg actuators as shock absorbers and obstacle avoidance on a bumpy path. Best of both worlds, with a new software challenge.
> Wheels were invented for a reason
Literally we have a very old saying that is "don't reinvent the wheel" and yet people persist in doing so, when an already near optimal design exists.
I'm not saying don't do anything new and novel but that if a large portion of the design has already been done for you and you can improve it, like this company has done (wheels on the end of legs!), why not try that first and see if it works before starting over.
There are so many places wheels can't go and legs can. For example, most rough terrain is right out of the question for anyone using a wheelchair. There are all-terrain wheelchairs that can go over some rough terrain but you just can't move across the same range of terrains with wheels that you can with legs. Even Mars rovers have to carefully avoid areas of e.g. deep sand where they can get stuck etc (and they still get stuck occasionally like Spirit†). Hence also all the specialised wheel designs like tank treads.
If it were possible to make legged robots with the same fine motor control and dexterity as human legs, let alone four-legged animals' legs, we'd be way ahead of where we are right now in mobile robotics. We can't.
https://www.youtube.com/watch?v=lA8WuXDXfcI
relevant link.
I remember watching this video with the takeaway: if you want it secure, build it yourself... which is what I am currently doing.
We need open source versions of these.
1. We need the capability to build the servos, actuators, etc. Supply chain, factories, specialized vendors.
2. We need to dump the firmware, decompile it, extract the models, and liberate it all. AI will help immensely with that. We can literally hijack everything software that isn't delivered over cloud now - so this is the easy part. We can then fine tune, adapt to new tasks, etc. Grabbing the brain = easy.
So the hard part is the hardware itself. The brain will belong to us soon enough.
Robotics distillation will be a big thing. You won't even have to farm out billions of requests across residential gateways. You just need to buy one unit.
You can't stop the software from becoming open.
Startup idea: AI model to dump the firmware from DJI, Unitree, etc. etc. and turn it into human-readable code. Broad utility across every industry with tons of willing buyers.
Hardware, though. We need that part.
You can’t raise money for theft / piracy. Just have your claw guide you through the hardware.
You can create a bounty through a nonprofit foundation, and use donations that are tax deductible to fund reverse engineering and creating an open version under an open license that defends against commercial capture.
This is one of the superpowers of LLMs and generative AI I think, providing cheap reverse engineering superpowers to clean room reimplement proprietary and closed systems as open source systems. Fiat, tokens, and context goes in, technology sovereignty comes out. In the same way people talk about a software factory, think about a reverse engineering factory. You are laundering the knowledge and tradecraft embodied in artifacts, both digital and physical, through model training and inference.
Open hardware developed by the above process can be produced by contract hardware manufacturers or someone CrowdSupply facilitates. If an expert opinion is needed on this, I’d recommend Andrew Huang of bunnie studios.
https://bounties.fulu.org/
https://www.fulu.org/
https://simonwillison.net/2026/Jul/20/cheap-reverse-engineer...
https://www.bunniestudios.com/
https://hn.algolia.com/?dateRange=all&page=0&prefix=false&qu...
(think in systems)
It's an engineering tool. I can think of dozens of completely legit use cases off the top of my head and there are probably hundreds more.
Security review, checking for bugs or spyware, repairing your own hardware you own, modifying your own hardware you own, measuring performance, ...
It's legitimate as hell.
They never give you dev access without paying like 10x the price.
Interesting, you can’t use their SDK or you can’t load whatever you want on it?
E.g. Unitree R1 has attractive low price but missing the “secondary development” pack (access to Jetson Orin and all the firmware stuff to let you run custom software). It’s all just their software. Useless tool. Click button and it dances. Then what.
howc hard is it to jailbreak the device?
I've seen a similar setup with a different Chinese cutting edge product. App access is free, ssh is open, but API keys and the ssh password is priced like a cheap car.
I'd love to see an ebike-like vehicle in this spirit. Meant to be ridden, self-balancing, and all-terrain capable.
Despite being mocked by Henry Ford decades ago, we'll finally get our 'better horse'!
Build it in the shape of a domesticated Bactrian camel, put the seat in between the humps, use the humps for extra battery (or cargo compartments)
I think you've just nearly described the 'Chevaline' mechanical horses in The Diamond Age, but without the nanotech
just in time when yc is starting a primer call
Seems like that would significantly raise the center-of-gravity making it less stable?
I want to see it performing in sand or in very dry soil environment too. They didn't show that in these videos.
I imagine it ought to be possible to use computer vision to augment the traction control system, all the way up to locking the wheels altogether and using them as feet wherever necessary.
Even if the AS2-W isn't capable of doing that today, it's a fairly obvious goal for future development given how generally useful wheels are otherwise.
I love 4 leg robots with wheels because it is a form of locomotion that evolution didn't create and seems to be wildly effective with the wheels allowing for very fast movement on smooth surfaces and the legs acting like a suspension with huge travel range on rough surfaces and then it can switch to walking mode with the wheels acting like ankles. And they can step over obstacles or step up to higher surfaces.
This thing reminds me of something out of Death stranding, Hideo Kojima was way ahead of his time
You might also say that about Terry Pratchett… see Twoflower and The Luggage.
Man, this looks like a great read, Thanks!
how is this robot seems to have perfect traction ?
its either rolls perfectly without traction or walks uphill with perfect traction ?
is it AI generated video or did they make some groundbreaking success ?
or is it just promo video, showing future potential, not yet developed capabilities ?
Here -and with every product promo video - you are watching curated shots. Non-flattering footage simply will get deleted during edit, except perhaps for the blooper-reel, if that's the company coldly calculates to present a goofy, non-threatening vibe.
It doesn't have perfect traction. It stumbles in the videos. This is consistent with other wheeled robots that have come out the last few years.
If I understand your question correctly, I think you may be missing that it can lock the wheels so they don’t roll sometimes like when it’s climbing uphill.
when can i hop on its back and ride it like an all-terrain ultra-longboard?
Looks like right now. They had an image of it with three small Asian men standing on it. I believe I’ve seen a previous video of this with a guy actually riding it. It should still be on their YouTube channel.
Impressive … and scary. Why is it that this company is moving so fast when others like Boston Dynamics don’t seem to have made much progress?
Chinese work ethic driven by overproduction of STEM majors leading to 25% youth unemployment...also they probably stole a lot of the fundamental stuff that was super hard to produce but easy to copy and build on top of.
On the other side: Years of VC focusing on the next great app to make you some cooked eggs instead of investing in companies operating in the physical world.
Their earlier robots (A1, Go 1) have a lot of lineage from Ben Kats' (from MIT) Mini Cheetah project, even down to the shape of the communication protocols used for motor control. I've not had the chance to closely inspect their newer robots, so I can't tell you what's their secret sauce, but I suspect it has as much to do with RL-based control and software as it does with hardware improvements.
I often hear about "muh china stealing", two questions come to mind:
1. how could they steal something that western companies dont even have yet?
2. for more fundamental stuff, western companies did transfer knowledge themselves by moving factories to China, in exchange for decades of access to China market and exorbitant profits they earned. Where exactly is "supposed stealing" in unitree ?
Boston Dynamics has been at the game for a LONG time (1992 vs Unitree's 2016). During that time they slogged through a lot of fundamental work that from the outside looked like 0 but thats the thing about tech, it starts at 0.01% then increases 0.01% → 0.02% → 0.04% → 0.08% → 0.16% → 0.32% → 0.64% and so on. The world just sees 0 until they don't. If you are able to just take all the results of those years and skip the same effort required then combined with your own skilled and motivated people you can leapfrog them.
(Im basically rewording my original comment, you didn't get my point the first time)
The answer to both your questions, and the poster's obvious implication, is that somehow they stole information from Boston Dynamics. (I'm not saying I believe that.) Seems like they must've stolen the information at the exact point in time when advances in AI made these physical robots much more easy for anyone to engineer without theft...
Steal info from 10 different companies, and combine it to make a product better than any of the 10 individual products copied.
It's basically the patent problem.
US patents do not have extra-territorial power, meaning US patent means nothing outside its borders, unless there are mutual reciprocity agreements (mostly between US and its vassal puppet states).
To my knowledge, China has never been caught infringing on US patents on US soil. A lot of Chinese innovations are simply orders of magnitude better than any US alternative, for example TikTok's algorithm and TikTok as a short-video social network.
So please answer my question, how could China steal anything that either 1) Western companies explicitly granted as part of Joint Venture corporation, 2) Western companies havent even developed themselves, and at best, only possess inferior version, generation or two behind the state of the art?
in IP realm, India for example forcefully nationalizes all pharmaceutical IP and produces generics of brand name drugs for its internal market (1970 Patents Act)
Idk about work ethic, but what are the chances it's a 996 company?
All Chinese companies are 996 companies and any that arent will eventually be replaced by a 996 company. Again vast oversupply of STEM labor serves as a forcing function(but is not the only reason).
There's no use for these robots(even militarily). It's very high tech, but that's it.
Veho[1] is experimenting[2] with similar kinds of robots for package delivery.
[1] https://www.shipveho.com/rivr
[2] https://www.youtube.com/shorts/d6jU_4XjkTU
You say that now, but wait until one of these is charging at you across a field with a QBZ-95 strapped to its back and a 1024 resolution thermal sensor.
Not this robot. You can see that none of the videos where the robot is careering downhill or rolling fast over rough terrain shows it carrying any load at all. Since those are the best-of, carefully curated videos, painstakingly chosen to show off the system's capabilities at its best you can rest assured that the robot can't manoeuvre like shown in the videos with any load at all.
There's a video where it's going uphill over uneven terrain with a load at a much slower pace, which is the same as what Big Dog could do 21 years ago, and another one where it's shown rolling up a smooth, grassy slope while loaded, which is pretty meh.
And that's just carrying a load. Now, you want it to carry an assault rifle and fire it? Like, at enemy targets? And hit them?
You will need to wait for a couple of major breakthroughs in robotics and AI still, I'm afraid.
I'd be more scared of MQ-9 dropping a JDAM or R9X on my head than this thing running around. They're proven to be undetectable to human targets. That's scarier to me than this thing running around catching wires and bushes.
mq9 ~$20m price tag will buy a LOT of robodogs with machine guns able to exterminate disproportionally more people per $ and time unit.
I’m just a regular person and even I can see use for this where it carries all my gear on a very long hike or when I go camping. I’m sure that alone would be extremely valuable to any military. Also, not everything is aerially accessible. Lots of tight spaces in urban combat where you need to get get inside structures and interact with physical things like doors that drones just don’t have the physical strength for and probably never will.
Guys, guys:
>> High Payload:
>> Static 150 kg;
>> Continuous walking payload approx. 16 kg
And on top of that:
>> Long Endurance:
>> Unloaded 3h+ (>33 km);
>> Loaded > 2h (>16 km)
16 kgs for two hours max is not "extremely valuable to any military". It's hardly even valuable to a leisure camper. You might be able to get this to carry some of your stuff if you go on a short hike, but if you were thinking of going camping and having it carry your camping gear, you can forget about it.
You wait until they mount them with a chainsaw.
Or five. One on each leg and one ON THE FACE!
And then a tiny little one on the bum like a cutesy tail <3
You are assuming this thing is real and not just a web page.
Not that you have much reason to believe a stranger on the internet, but I have seen this in person.
Not doing acrobatics like the marketing video of course, just doing mundane things like going from point A to point B.
Unitree is a huge well known company. You think they'd fake this?
The demo images look AI generated, so I can understand the impression that this is vaporware. The demo videos look very real to me however.
is using a prototype firmware or hardware platform faking it when you're showing exclusive prototype capabilities and trying to pass them off as reproducible by the production unit?
unitree has done this consistently.
What does unitree G1 do as a demo that it can't do once you bought one?
they're a chinese company and this is a VC funded western company; so that's half the answer.
The other half is yes, of course, any company will fake it till they make it.
But Unitree is definitely making robots.
I've always wanted something to carry my fishing gear on the jetty, can this do that?
Very cool tech, but don’t want one of these things coming after me
What's the price?
Based on the price of their other robots, probably $5-10k.
The 5-10k$ price is for their consumer-grade robots like the Go 2 or low-end humanoids. I won't be surprised if this monster costs 50,000$ or even more
I tried Unitree qugv and they were garbage, not so sure about this one
imagine that a an atv
Motorized mule essentially, but one that clears terrain like a mountain goat.
Maybe a Multiple Use Labor Element?