Google’s origami-folding AI brain may power new wave of humanoid robots

duncansil

Ars Scholae Palatinae
883
The Three Laws, presented to be from the fictional "Handbook of Robotics, 56th Edition, 2058 A.D.", are:

  1. A robot may not injure a human being or, through inaction, allow a human being to come to harm.
  2. A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.
  3. A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.
I know a few stories where the above had some unforeseen outcomes :judge:
 
Upvote
9 (18 / -9)
The one thing that struck me with these demos is how unbelievably SLOW the movement and reaction times are, and that's probably with unlimited cloud inference power.

Which would mean they're probably several orders of magnitude of efficiency and speed away from doing local processing.

I mean, we're not actually going to have these things run from the cloud, right? Imagine one of these changing a diaper on a baby and Comcast farting out halfway through, leaving your baby suspended in the air (or worse)...
 
Upvote
39 (44 / -5)
Given the recent history of outright fraudlent presentations in this area, forgive me if i am skeptical until these tasks are verified by third parties with unfettered access.
It seems pretty legit (if perhaps cherry-picked and even then, not very impressive) -- I tend to trust this a lot more than Optimus because of the distinct lack of Elon Musk involvement.
 
Upvote
33 (34 / -1)
Post content hidden for low score. Show…

Frank Xavier07

Smack-Fu Master, in training
3
It would be interesting to have a comparative performance evaluation framework that provides meaningful insight into the evolving progress of robots comparing performance across a variety of tasks. One robotics company which appears to be routinely missed is Sanctuary.ai which entered into a major partnering arrangement with Microsoft in May 2024 and has been recognized as the global leader in Humanoid Patents. I have no financial interest in the company and have been following it since its inception.

 
Upvote
3 (3 / 0)

Fatesrider

Ars Legatus Legionis
22,987
Subscriptor
Tesla unveiled its Optimus Gen 3 robot last October, claiming the ability to complete many physical tasks, yet concerns persist over the authenticity of its autonomous AI capabilities after the company admitted that several robots in its splashy demo were controlled remotely by humans.
I wouldn't have a device made by a Musk-owned, or involved in company in my home to begin with.

But WRT in-home humanoid robots, it may be they'll arrive too late to be widely useful. They're really only good for personal care, and with the population of older people dying off faster than before (at least in the US - and three guesses why, all of them probably playing a role), they'd have to be affordable and hit the market within the next ten years as the Boomers continue dying off.

Demand-wise, they're the largest old-age demographic in the world. GenX'ers aren't nearly as numerous because a hell of a lot of Boomers (like me and my sisters) decided life was better without kids. And the trend thereafter was (for the most part) one or two, which further reduced the total population for each generation. We literally face a near future where there will be more people needing care and assistance than there are people available to do that.

Enter robots, which could perform the mechanics of the care, but not the personal interaction (unless mated with AI, which is a possibility), which is extremely important for the quality of life of the person being cared for.

Given that I'm a member of that aforementioned demographic, and don't especially want people fussing over me, I'd be more inclined to accept robotic aid if needed than human. But with my level of "wealth", I don't expect I'll ever be able to afford that.
 
Upvote
4 (9 / -5)
The one thing that struck me with these demos is how unbelievably SLOW the movement and reaction times are, and that's probably with unlimited cloud inference power.

Which would mean they're probably several orders of magnitude of efficiency and speed away from doing local processing.

I mean, we're not actually going to have these things run from the cloud, right? Imagine one of these changing a diaper on a baby and Comcast farting out halfway through, leaving your baby suspended in the air (or worse)...
For some industrial applications cloud would work just fine. Think, idk, warehouses.
 
Upvote
5 (5 / 0)

GatoEspacial

Smack-Fu Master, in training
25
Subscriptor
Microsoft released a similar thing 2 weeks ago: Magma.

I don't think it was reported here but it should be mentioned also, Google shouldn't be taking all the glory. Especially because Microsoft released the model for everyone to play with and Google isn't doing that. And I'm saying that as someone who is usually highly critical of Microsoft.
 
Last edited:
Upvote
2 (2 / 0)

alphathreethree

Ars Praetorian
526
Subscriptor++
Upvote
3 (3 / 0)
For some industrial applications cloud would work just fine. Think, idk, warehouses.
Sure, but then we're back to the hilariously low speed these things are working at. Look at the videos. Sure, this could, for example, tape a box closed -- but it would take a minute each to do it. That's about 50 times too slow for industrial applications. Even 10 times faster than this, you'd be laughed out of the room proposing this for industrial use.

Edit: and no, a lot of warehouses would not be OK with a cloud dependency either. "Oh, sorry, AWS had a little fart, you're down for a few hours" is not really acceptable for a manufacturing or distribution facility running on single-digit profit percentages. They'd want five 9s up and down, and you wouldn't get that from your ISP unless you pay out the ear... so what does this super-slow, super-expensive, cloud-dependent thing give me over a few purpose-built dumb robots with local control?
 
Upvote
9 (11 / -2)

hizonner

Ars Scholae Palatinae
974
Subscriptor
Regardless of what they look like, how will society change when people begin to have emotional attachments to the machines?
I've been carping for years about how it should be illegal to build a robot that people will get attached to that way. Or at the very least you should have to do certain things to discourage it, and definitely not do anything to encourage it.

I am not, however, optimistic about anybody doing anything about it.
 
Upvote
-3 (1 / -4)

hizonner

Ars Scholae Palatinae
974
Subscriptor
Sure, but then we're back to the hilariously low speed these things are working at. Look at the videos. Sure, this could, for example, tape a box closed -- but it would take a minute each to do it. That's about 50 times too slow for industrial applications. Even 10 times faster than this, you'd be laughed out of the room proposing this for industrial use.
How much of that time is spent in understanding what the task is, though, and how easily can you feed back from past performances of very similar tasks to make it faster at any given one?

It took a while to figure out what to do with the fruit, but notice how fast it reacted when the guy moved the clear box.

Humans are slow the first time they do something, too.
 
Upvote
2 (3 / -1)
Microsoft released a similar thing 2 weeks ago: Magma.

I don't think it was reported here but it should be mentioned also, Google shouldn't be taking all the glory. Especially because Microsoft released the model for everyone to play with and Google isn't doing that. And I'm saying that as someone who is usually highly critical of Microsoft.
MSR does so many things nobody has ever heard of... "hey, guys, we really want an OS that is fundamentally secure" -- welp, MS did that 22 years ago and nobody cared :-(

https://en.wikipedia.org/wiki/Singularity_(operating_system)
 
Upvote
5 (5 / 0)
How much of that time is spent in understanding what the task is, though, and how easily can you feed back from past performances of very similar tasks to make it faster at any given one?
If it's going to do the same thing over and over, you don't need multi-billion dollar AI to figure it out, it can just be programmed to do it. The flexibility is the entire point of this, and there's just no point in that kind of flexibility in any industrial application I can think of. There is a very good reason this is marketed as a home assistant and not a generic industrial robot.
 
Upvote
9 (9 / 0)

phik

Ars Centurion
203
Subscriptor
Sure, but then we're back to the hilariously low speed these things are working at. Look at the videos. Sure, this could, for example, tape a box closed -- but it would take a minute each to do it. That's about 50 times too slow for industrial applications. Even 10 times faster than this, you'd be laughed out of the room proposing this for industrial use.

Edit: and no, a lot of warehouses would not be OK with a cloud dependency either. "Oh, sorry, AWS had a little fart, you're down for a few hours" is not really acceptable for a manufacturing or distribution facility running on single-digit profit percentages. They'd want five 9s up and down, and you wouldn't get that from your ISP unless you pay out the ear... so what does this super-slow, super-expensive, cloud-dependent thing give me over a few purpose-built dumb robots with local control?
1. They're not replacing industrial robots. They're going to replace people by working slowly all night, every night, forever.
2. They're not going to remain this slow.
 
Upvote
13 (14 / -1)

TheShark

Ars Tribunus Militum
2,945
Subscriptor
I still think the ultimate benchmark for robotics will be changing a diaper.

While changing a diaper isn't really a pleasant experience, I suspect a fair amount of emotional bonding happens during it. As a parent, there are a lot of chores I would put the robot to work on before that one. Kinda like plopping your kid down with headphones and an audiobook is not anywhere near the same thing as reading to them at bedtime.
 
Upvote
13 (13 / 0)

bugsbony

Ars Scholae Palatinae
910
Stan Lee and Steve Ditko: In our comic books we invented Doctor Octopus's robot tentacles as a cautionary tale.

Google: At long last, we have created the robot tentacles from the 2004 classic movie Spider-Man 2, whose main plot could be basically summed up as Don't Create The Robot Tentacles.
I don't much about comic super heroes. But I really didn't get the impression that these were cautionary tales rather than "watch these super strong heroes fight super strong villains, that's so cool".

Anyway, all stories need some drama, things that go wrong, otherwise they are boring and nobody cares. And even (literal) cautionary tales are only the caution of one guy that probably doesn't know much about the technology in question.
 
Upvote
2 (2 / 0)

SiberX

Ars Scholae Palatinae
1,212
Subscriptor++
I'll put money into one when you can give it a hamper of dirty laundry and it'll hand back a stack of folded.
My personal benchmark scenario is blindly reaching into a cluttered pocket of items and pulling out a specific one, without dumping any other items or upending the pocket.

It's a seemingly simple task for most humans, but combines a variety of skills that are currently well out of reach of the state of the art in robotics.

Once somebody develops a reliable RummageRobot, we'll know that general purpose robotics (and the future) have really arrived.
 
Upvote
15 (15 / 0)