An electric new era for Atlas
bostondynamics.com
bostondynamics.com
Farewell to HD Atlas https://www.youtube.com/watch?v=-9EM5_VFlt8
Boston Dynamics retires its legendary humanoid robot https://spectrum.ieee.org/boston-dynamics-atlas-retires
All New Atlas https://www.youtube.com/watch?v=29ECwExc-_M
Right to repair will be even more important for this technology than autos or general computing.
There have been some efforts for vendor-agnostic robot software like RoboDK and other warehouse execution systems, but the default is proprietary vendor software.
It would be nice for society if this were true, but we'd need someone to exist whose complementary technology was robotics who found it worth commoditizing the entire ecosystem against their will. Or regulators who weren't entirely beholden to industry lobbyists.
Fanuc robots are straight forward to service, they make the parts very available to do it yourself if you want. We order them here and there no problem.
But they are beasts and it can take an entire day just to replace a part. Then you have to reassemble it in the right order. None of it is made difficult on purpose. It has tight tolerances, and fancy shit like harmonic drives for zero backlash and more.
It's going to be kneecapped far worse than phones or tractors. A general purpose humanoid robot is orders of magnitude more complex than a simple gps farming tractor or a cheap android phone.
Companies will absolutely NOT want to give up that moat after developing such tech for 10-20 years.
Right now no regular user has the technical ability to fiddle with a phone's laminated screen glued to a touch matrix paired with a fingerprint sensor and a camera, so we're alreay past the complexity threshold.
But we could still reuse a screen block from phone A on phone B, except that's been forbidden by technical measures specially added to prevent it.
The same way we could probably replace a whole leg with another from a robot from the same series, except it will be DRMed to death.
We'll have to eternally push for regulation I think, companies will always try their best to fuck with repairability.
That's the wrong way to say "recouping the cost of a large up-front R&D investment".
The second someone releases a general purpose humanoid robot that is capable of self replication but is locked out from doing so with DRM the race will be on to break that DRM.
The self replicating humanoid robot will be a supreme game changer. It's a genie in the bottle that lets you wish for more wishes.
> Under the law, companies that make cellphones and other consumer electronics are required to provide the tools and know-how to repair those devices.
1. Do you think the Oregon law fell short by not requiring industrial electronics to be repairable as well? 2. Will the proliferation of tools and know-how for repair be sufficient to meaningfully extend the life of most electronics? 3. Is legal mandate sufficient or necessary to motivate companies to open their chests to the public? Or is a voluntary movement possible that still rewards the stakeholders?
My hope is that projects like Atlas will be sustainable and prices eventually come down to commodity levels - say the price scale of cars. If people are empowered with tools to develop on these machines in a safe way, I think we could see a revolution similar to the cell phone or PC. My fear is that these machines will become just an extra inefficient automation step in an overpriced supply chain one-off application.
Humanoid robots have many, many challenges to deployment. Especially, creating a machine that people can safely operate near is extremely challenging. The amount of intelligence person uses to not bump another person is very under rated.
Worth calling out that Hyundai is a major investor in Boston Dynamics.
FTA: This journey will start with Hyundai—in addition to investing in us, the Hyundai team is building the next generation of automotive manufacturing capabilities, and it will serve as a perfect testing ground for new Atlas applications.
I believe robots are currently making car parts in abundance. The robots usually are like a box with a hydraulic arm or something equivalent.
The specially and especially hard part of humanoid robots is justifying the cost and complexity of the construction by having them by "walk-on replacements" for humans and so they have failed entirely at being that.
What can humanoid robots making car parts do, that the already-existing and already widely deployed robots making car parts can't?
* which definitely requires human level general AI at fairly low electrical power demand
Why would the super villains operating these armies of human-capable robots bother paying into an upside down pension system?
At the very least they can defend themselves from the torch wield masses with even more robots.
In my opinion, repair and maintenance is the most commonly overlooked aspect of an automated system deployment. Scaling is impossible without efficient tools to fix problems when they occur, especially if the number of authorized service people is limited.
The more serviceability can be automated and standardized, the greater the number of areas that will benefit from widespread robotics.
But I would have thought they’d rather not have us experience atlas as some kind of freakish terminator mixed with the girl from the ring.
But the animal part of our brain that screams danger when we see this is just a byproduct of that. Anybody giving it real credence rather than just laughing it off should stop driving a car, travelling by air, talking to people on a phone, etc.
It looks amazing in the video.. But of course Boston Dynamics chose the most disturbing way of demonstrating its movement capabilities, as usual.
I swear they do it on purpose at this point. Good lord! Put some googly eyes on these things at least.
https://www.avforums.com/reviews/lost-in-space-season-1-tv-s...
However if you look closely the robot does have scuffs and scratches on it so I think it's real.
On the other hand, as a non-American, I admire that the USA is seemingly the only place where people get funding for wonky ideas that sometimes become very successful.
1- https://www.cnbc.com/2024/02/29/robot-startup-figure-valued-...
This "robots + AI" space is heating up just as fast as LLMs, and every country seems to have a dozen startups in the ring.
Here is just a sample:
https://www.1x.tech/androids/neo
https://rainbow-robotics.com/en_main?_l=en
https://www.youtube.com/shorts/CToL2qkCd8g (funny)
https://www.reddit.com/r/singularity/comments/1b10p2i/chines...
https://www.engadget.com/menteebot-is-a-human-sized-ai-robot...
https://www.reddit.com/r/Damnthatsinteresting/comments/15jyw... (NSFW)
...
Everyone is working on this.
https://james.darpinian.com/blog/you-havent-seen-these-real-...
Edit: Haha, case in point. I opened Twitter and sure enough there's a new announcement of a humanoid robot today, from Intel/Mobileye: https://twitter.com/AmnonShashua/status/1780611499133685889
You doubtlessly know more about life in South Korea than I do, but i found this video [0] and its sequel [1] very enlightening.
0: https://www.youtube.com/watch?v=-Im4YAMWK74&t=1050s 1: https://www.youtube.com/watch?v=woB0eecbf6A&t=589s
Murdoch family for example has huge influence in US, UK and Australia.
I wonder if there is research on the topic - I mean Adam Smith is translated to all languages so it’s not about the ideas or non-tacit knowledge. Must be something institutional or otherwise cultural.
A lot of people just seemed content and satisfied with their lot in life, without much ambition or drive to improve their position. Is that a good or bad thing? I honestly don’t know. In this point in my life, I don’t particularly like it. I love visiting Europe but would not want to live (and work) there. Maybe I’ll think differently when I’m older and or retired.
Versus the US and Asia where many people are trying to claw their way to the top. Obviously most will not get to the top, but many do end up improving their socioeconomic status to varying degrees.
Then again, it's hard to deny the progress and products these countries have made. So what gives? To be honest, I don't know.
There's a flip side to South Korea's chaebol-centric economy, however. South Korea's national security situation is extremely dangerous, so in fact one of the reasons for the industrial policy has been to maintain a domestic defense industrial base so that they aren't dependent on arms imports from Western countries. Accordingly, most of the South Korean chaebols have a significant presence in the arms industry. In recent years, this sector has expanded, with South Korea becoming one of the world's leading arms exporters.
Here is the definition I'm working from: "private ownership of capital and means of production meets market competition, driving resource efficiency, innovation, and maximizing profit while respond to consumer demand resulting in GDP growth."
If you take that definition then look at the last 100 years, only 4 names come up:
Japan, South Korea, Singapore and Taiwan. No countries in the world in the last 100 years have applied capitalism, then grown, the way those countries have.
I'd be curious how you define capitalism, and then the countries you think have applied it better than the ones I've mentioned in my posts on this subject.
China is not far behind, despite an authoritarian govt.
KR & JP, as well as CH, clearly learned well from Americans.
https://www.youtube.com/watch?v=09ekK2QgflM
Maybe we need life-size robot battles? Would love to see son-of-Atlas suplexing TeslaBot like that mini white one does!
You apparently disagree? Was there something in the video you think marks it out as CGI? Or do we just have differing gut instincts about it?
This is going to haunt my dreams.
I am betting that this one is less powerful, no backflip.
https://spectrum.ieee.org/atlas-humanoid-robot-ceo-interview
It clearly has a much larger range of motion and if it is also stronger as claimed then I can't wait for the acrobatics videos that are surely coming.
But I think the most exciting thing is that it has hands from the start. Atlas didn't have hands for most of its existence and so couldn't do much in the way of useful tasks. I think controlling hands is actually much harder than walking or doing backflips. Hopefully Boston Dynamics will be able to make this version useful.
Or just use wheels / a wheel. This whole humanoid thing strikes me as an addiction to old sci Fi stories.
Hydraulic systems have very little "give", unless you put a hydraulic accumulator (an air tank with a fluid/air barrier) in the system. Electric motors have plenty of "give". Forcing a motor to turn backwards won't hurt it. The gear train is usually the weak point. As motors and controllers have improved, robot gear reduction ratios have decreased, which reduces the load on the gear train and lets the motor absorb shock loads. Direct drive robots eliminate the gear train entirely. Here's a nice one.[1] "You cannot strip the teeth of a magnetic field" - General Electric electric locomotive rep, around 1900.
With modern motors, you can get huge torque with light weight, and cooling becomes the limitation. Schaft used water-cooled motors in their direct-drive robot. Google bought Schaft, ran them into the ground and killed them.
[1] https://shop.directdrive.com/products/diablo-world-s-first-d...
I'm still mostly convinced that harvesting the energy and re-using it ala elastic tendons is a decently good idea. But probably far too complex.
One of the cheapest low-gear-ratios I found was the RMD series: (which is 1:6 or 1:10 gear ratio).
https://ozrobotics.com/shop/myactuator-rmd-x4-16-micro-motor...
If the gear ratio on these motors is high, then there can only be faked compliance in the tuned force-torque controllers you mentioned. MIT's little cheetah robot, on the other hand, deliberately used low-gear ratios to keep things naturally squishy if needed. This is the way to go; putting elastic tendons or spring elements seems like a good idea but then you can't actually model the non-linearity well (the 1st order motor becomes a 2nd or higher order system).
I don't have any particular problem with that, but its a little weird? I figured they were a more traditional industrial robotics company that just did the humanoid robots as a side line for publicity, but googling, I guess that's not the case.
The whole point of the article is speculating that they are specifically retiring their hydraulic robot because it was never going to be commercially viable. Which makes it look like they are finally ready to pivot from pure R&D to commercial production. Thus they want fully electronic robots instead of hydraulics that are messy and require more (almost constant?) maintenance.
I'm not an engineering guy but I assume the hydraulics were more useful for pushing the boundaries of possible motion with such a heavy, robust, and versatile design. Now that the AI systems controlling vision, motion, proprioception/spatial awareness, etc are more fully developed, they can create more specialized and scaled down versions of the robot for specific applications that are lighter and don't require hydraulics to perform their tasks reliably? Just guessing here, am happy to be corrected or given more a nuanced take.
I remember seeing prototypes from Toshiba when I was 10 (20 years ago), and every few months, there is a company releasing an "amazing video." its mother company then spins it off like there's no adequate progress, and so on.
In it she covers the latest and greatest robot news, with occasional commentary/perspectives.
However to more directly answer your question, you need to know/talk to someone in the industry at the moment. I am not aware of a single “spot” that gives an honest in depth appraisal of where we are.
From my experience there is a ton of new “hardware” coming out, not just in the humanoid space (Agility Robotics being imho the most “real”), but also in lower cost robot arms, end effectors, sensors, and compute.
Where things are harder to track is where we really are in the software realm. If you look at software driving this hardware, most of it is early stages. Perhaps TRL level 3 to 5 at best. The higher TRL is non-intelligent control software (that is based on decades of work). The newer, AI/Machine Learning/“Smart” software tends to only have limited roll out. At best it will be a startup at the relatively early stages, but more often then not it is still a researcher sitting at a University or a large corporations research lab. In either of those cases, you will see single to at most double digit examples of those systems actually doing work.
However, to your point, it is super easy to create a single (or even a series) of cool videos… it just takes one success in 100s of takes. It is harder to make something that will perform day in and day out and really change the industry/world.
In general robotics flies under the radar because it's rare to see a unicorn or anything really flashy and there is a big gap between big aspirations and fake demos and real world applications with polished use cases and diligent design, processes, etc.
source: I'm a skeptic roboticist working in the industry.
Like with every other market check if the product is available for sale and at what price point. And then look up what failure points people actually using the system are complaining about. (Because every system has problems and weaknesses. If you don't see reports about any then the system hasn't left the lab where the PR of it is controlled.)
worked examples: washing machine (that's a robot alright, has a computer, actuators, sensors). Readily available commercially for 200-500 GBP. Usually works reliably, occasional reports of flooding the room.
robotic vacuum: Readily available commercially for 300-1k GBP. Works okay, reports about it spreading pet's poop around rooms.
spot from Boston Dynamics. Not as readily available as the above, but can be purchased. Reported price 74,500 USD[1] Seems to trip over its own legs sometimes in a hard to explain way: [2][3] (not to count as a dig against spot, seeing these issues is actually a great thing. It means third party people in the real world use it.)
atlas from Boston Dynamics. You can't buy it. No price advertised. You can't see third party reports of it malfunctioning. Not because it is perfect, but because nobody has access to it.
1: https://spectrum.ieee.org/boston-dynamics-spot-robot-dog-now... 2: https://www.youtube.com/shorts/8bTo9Q3FWzE 3: https://www.youtube.com/watch?v=nJHAJm3uMEI
With FSD 12.3.3 released, it's clear FSD is getting smarter and smarter. How many of those releases left until people trust Optimus to fold their laundry? 1.0 Optimius will still be pretty dumb, but could still be worth the price (especially with continuous software upgrades!)
Such WBC then makes sure that the robot reaches both it's task goals (eg. grab something, with 1, 2 arms), as well as it's (dynamic) stability goals so it doesn't fall over. They are also capable of choreographing the robot pretty accurately as we say in earlier videos. But what is most very impressive to me is the robot using the mass and momentum of things it grabs to keep stable or move itself. In one of the videos it grabs a big piece of wood and uses it to turn itself around while jumping. Amazing! Controlling that in terms of dynamics is... wow!
https://www.youtube.com/watch?v=-9EM5_VFlt8
Looking forward to see some more robot parkour/dance
> https://www.youtube.com/watch?v=-9EM5_VFlt8
I wasn't expecting to see a robot bleed, several times.
Boston Dynamics Retires Its Legendary Humanoid Robot
Spoiler alert: dis-arm.
"Understood, now de-livering John."
But it was always going to be Skynet.
I bet the next version will have teeth.
Mostly wheels just seem like a better idea. For rough terrain, why not just fly ?
Weight
(You need way more energy to do anything)
Wonder if that includes weapon systems?
https://en.hyundai-wia.com/business/defense_business.asp
> With its cutting-edge unmanned and automated weapons systems, Hyundai WIA upgraded the level of defense industry system.
https://www.axios.com/2022/10/06/boston-dynamics-pledges-wea...
Ed: apparently the pledge came after Hyundai took control?
https://bostondynamics.com/news/hyundai-motor-group-complete...
https://m.youtube.com/watch?v=ZFRcle4Szo4 https://m.youtube.com/watch?v=a2Y52zjZYXo https://m.youtube.com/watch?v=9pZQ29RSz4I https://m.youtube.com/watch?v=XPOpnJSldUg https://m.youtube.com/watch?v=p535RRR5MqM
This one is actually pretty interesting cause handling big breakers is quite hazardous.
https://www.thestreet.com/automotive/boston-dynamics-robot-d...
so they do have non-military applications.
Is feels like they almost designed this video with that eventuality in mind. Like that wanted a second wave of organic 3rd party viral advertising.
Of course this is not directly comparable, but I think robotics is harder and more less open to brute force approaches.
Their hardware is second to none.
FML, that's dark.
There's also too much ambient occlusion in the rendering, but they were careful with the materials and the scene that it looks somewhat plausible.
Compare it with an actual video of their real robot: https://youtu.be/48qL8Jt39Vs?si=Kkb7oTnYf-GllRJv
The difference in the movements is clearly visible.
Will casting a net stop robot like this or do you have to somehow dismember it?
Geoffrey Hinton suggested that by 2030 the US military wants 50% robot
"Sorry we can't sell cigarettes to anyone born after 2009, or robots".
An elderly relative of mine has smoked his whole life. He admits it's bad but won't stop regardless of pressure from family. He enjoys the routine and the quiet moment. He would absolutely send his robot to buy cigarettes if possible. Nobody else in the family wants to buy them.
Boston Dynamics Retires Its Legendary Humanoid Robot
1. Amazing technical ability.
2. Feels scary, both the beyond-human movement, and the design of the 'face'.a logic of Atlases?
Boston Dynamics is not run by bozos, they have a pretty consistent track record of showing the real stuff.
For sure they have been working on this for a long time.
I predict that they will also move toward neural nets for all the vision, control and understanding of the world (like Tesla)
You sound confused.
> Billionaire-Fueled Lobbying Group Behind the State Bills to Ban UBI Experiments
You could offload the heavy processing to a larger computer in the back seat. Then even the robots get to suffer with backseat drivers :)
I bet it talks
The city I once knew as home is teetering on the edge of radioactive oblivion
A three-hundred thousand degree baptism by nuclear fire
I'm not sorry, we had it coming
A surge of white-hot atonement will be our wake-up call
Hope for our future is now a stillborn dream
The bombs begin to fall and I'm rushing to meet my love
Please, remember me
There is no more
I'm thinking more and more that that "Terminator" was the most accurate of all the sci-fi dystopias.
Progress in robotics is beginning to look non-linear and it will have only positive impact on the world.
Skynet won’t be humanoid terminators. Or even drones. The real threat of AI is if/when it will be applied to the field of virology.
Progress in robotics is beginning to look non-linear and it will have only positive impact on the world.
Skynet won’t be humanoid terminators. Or even drones. The real threat of AI is if/when it will be applied to the field of virology.
There are a sizable percentage of people out there that would love to use this for subjugation and control. Will we let them win?
Shock value PR stunt? Moving the Overton window for the general public’s aversion to what comes next?
Yes obviously there are limitations i.e. stairs and uneven terrain but there are wheeled/tracked solutions for those too
Most of these robots will be used in factories that have very nice flat concrete floors
One simple example: getting in and out of a car. Another thing to consider is that a legged robot can tilt itself for balance while carrying heavy objects. To carry a similar weight with a wheeled robot you'll need a much wider wheel base.
And then of course, if you want to build robots that can be useful inside a house, then they need to be able to cope with stairs. There's also construction... At some point, you don't have elevators... Or just circulating between buildings out on the street where the pavement isn't great.
You want to use the robot to inspect a tunnel in danger of collapse, or a factory that may be leaking a poisonous chemical out of a pipe.
And in such cases you very much want something that can navigate obstacles about as well as a human. You can't count on the area being devoid of rubble, and rebuilding a factory to make it wheeled robot friendly could be an enormously expensive and impractical proposition.
Now humanoids? We already designed everything for us. A good enough humanoid robot can go anywhere a person can, and manipulate anything a human was intended to touch.
I'd go out on a limb and say that we will NEVER have humanoid robots at home folding laundry, walking upstairs to put it away, or putting away the dishes in the kitchen. This is a 1960's sci-fi vision of the future, similar to that of flying cars. Any robot capable of fully navigating the human world will always be too expensive and unreliable as a home helper.
In a factory a stable wheeled robot is way more practical than a bipedal one. It doesn't need a humanoid head either - but I guess that makes for nice PR photos.
Right now the hardest jobs to replace will be those of plumbers, electricians, carpenters, etc where they need to operate with fine motor skills in unique and challenging locations - no two ever being the same.
What if it is knocked over and needs to get back up?
What about Steep inclines? Stairs?
What if it needs to climb on to a different platform? A conveyer belt? A vehicle? A beam?
Even in a factory or warehouse setting wheels are useless for anything but the most ideal cases. And there are already countless robots successfully operating in that space. A general purpose robot is the holy grail, and legs are a requirement for that.
But the long term prospects of robots would be in your home, maybe going to the store for you, whatever. We see the limitations of wheeled robots with robot vacuums. They do a decent job but are severely limited trying to do its job in a place that was designed for a human. (On the flip side it can also get some places easier than a human would, so it's a bit of a trade off).
By focusing on mimicking humans, we end up being in the best situation for both of these. Factories can try them out with minimal changes to how they operate.
Plus, it seems like the biggest hurdle isn't really walking. It seems like we have gotten that one down fairly well (not perfect obviously) and the bigger issues seem to be hands, object recognition, and just "general" AI. Can it actually do anything with the hardware it has on its own.
Seeing how this one moves, it is human-ish, being bipedal, but it isn't mimicking human movement range.
I’m gonna pass.
One of my pet peeves is the idea of asking the world to accommodate a situation rather than build solutions that adapt to the world.
Big example: the best we have for mobility nowadays is a wheelchair of some sort. That requires building special ramps and elevators everywhere.
If we had a four legged chair that could climb stairs, etc, like what BD is doing, it could transport people ANYWHERE. you could literally go for a stroll in the woods with it. People that are injured for 6 weeks in their home could go up and down steps, etc. The elderly could go for walks in a park.
So I for one fully support more research into smarter mobility that doesn’t require the world to accommodate it, but instead adjusts to its surroundings.
What changed my mind is thinking of humanoid robots as the “last mile” of robotics. All the thousands of use cases where there are no easy patterns and we need something that can fit into any human task without planning or modification.
The important part lagging is the brain. Understanding the world, reacting to it learning. Even an ant can navigate the world pick-up objects and do tasks.
But the fact is, the world has mostly been built by humans for humans. Pretty much any task you can think of can be accomplished by a human with their arms, legs and some tools.
A generalised robot would look like a human.
For one, there are many applications in dangerous environments that could benefit from the dexterity and ability of bipeds - rescue missions, mining, space walks, etc.
Are you sure? We had robots in factories for more than 50 years, and they don't usually move.
Too many responses: Oh, wow, it's so creepy, just like the book! Lol. Anyway, I'm pretty sure it won't turn out as bad as DCTTN. ;-) Best just get on with my day and mostly forget about it then...
(With apologies to Alex Blechman)