Boston Dynamics and DeepMind form new AI partnership | Hacker News Reader

Boston Dynamics and DeepMind form new AI partnership

bostondynamics.com

95 pointsmfiguiere52 comments

I wonder if Google regret selling Boston Dynamics? It was a interesting proposition at the time - self driving delivery vans with a humanoid delivery robot (charging from the van's traction battery between deliveries) for the last few steps from van to door etc. Ahead of its time perhaps ... but these days perhaps that will become viable again "soon"?

At least they didn't kill it I guess and BD got to live on as an independent.

The problem with hardware is that it's always viable soon, but not quite yet. Hardware is multiple orders of magnitude more difficult than anticipated, the fact that the richest companies in the world can't quite make it work is a testament to that. In my mind, hardware moonshots are kind of like trying to embed Doom in a SharePoint Framework Extension while high on psilocybin — impressive if someone manages to pull it off, but not for the sane of mind.

The big time suck I see with robotic anything is that simulation for training will only take it so far...eventually it needs to be in the real world, making mistakes, and this comes with far more red tape and much higher risk, slowing down the process. I don't see hardware as the bottleneck, its software and hardware working together in an environment where the stakes are much higher than just in the lab.

I feel like this point is refuted by what we're seeing out of robotics in China? And to a pretty good extend the U.S. Of course there is a curve which plots dexterity (hardware) and resulting capability, but we don't need /that/ much dexterity for some jobs. We have tele-operated humanoids 'doing things' where it is self evident that the bottleneck is the ability of the robot to act autonomously, not its hardware.

>I feel like this point is refuted by what we're seeing out of robotics in China?

Humanoid robot Olympic Games in China:

https://www.youtube.com/watch?v=5Y-tElcmJVE

Also, reminded - Russian "localization" (typical Russian hi-tech today, especially when on government investment, is simple rebadge of a Chinese tech) - even good Chinese robots starts to fall like drunk:

https://youtu.be/WVKxw72vlmo?t=15

and for the GGP comment:

>self driving delivery vans with a humanoid delivery robot

why humanoid? Glorified Roomba-like robots would do such job just fine. Every time seeing how the Amazon driver parks his van and runs around in our complex placing packages in front of the doors and making photos of the placed packages i'm wondering why Amazon wouldn't use 5-10 such Roombas per van instead. (and every time i think that i have to make such a startup myself, and after that i immediately think that Amazon would easily beat me by developing it 100x faster - in a week where i'd spend 2 years - and so i don't do it, and Amazon apparently doesn't do it too)

The simple reason why they don’t do it: It has to be 100% reliable. Robots get stuck, need charge, software has bugs, … So your fleet of robots would need supervision. Probably for years to come. And a human driver only costs like 70-80k a year.

The point of machine learning based systems (imo) is that they aren't 100% reliable.

Idk where people are getting the idea that systems designed to mimic biological brains will have machinelike precision whilst also being flexible to adapt to new situations.

A human supervisor monitoring 10 or more vehicles and unstuck them in case will also not cost more per year.

Yeah, but this requires good teleoperator infrastructure. You can’t unstuck a robot that loses connection. There are just a lot of things that can go wrong. And an entire car being stuck and waiting for someone to come is also not cheap. I am pretty sure Amazon (one of the biggest robotics company on the planet btw) has done the math.

Many houses don't have flat approaches, our definitely doesn't. Whatever delivers a package to our town home at least needs to navigate paver stones. Dog bots could probably do it though.

Deliveries are probably one of the few legitimate applications for humanoid robots, but even then 99% of the work is done on wheels and the robot is just there to ring a bell, open a door and climb stairs.

This is probably something people don't understand about humanoid robots. Nobody is dumb enough to replace their CNC machines with humanoid robots holding power tools and yet that is what you're being sold on when Elon Musk is teasing a trillion dollar valuation.

Instead, the vast majority of humanoids will be used for pretty boring FedEx or door dash style logistics work, not much different from wheeled robots.

> why humanoid? Glorified Roomba-like robots would do such job just fine.

Steps. Garden gates. Uneven surfaces. Communal entrances.

The real world is messy and certainly not flat.

Some sort of wheeled-legged-centaur type robot might work though.

Every single attempt at lifelike autonomous hardware over the last few decades has been a bottomless money pit, and despite all the recent advances the entire field is yet to be proven. A partnership with a robotics company is probably the best outcome for Google at this stage, because it will let them focus on what they do best - software

Yes even lifts, pickers, etc are in the realm of not having financial-econ impact for the groups that have invested or bought them.

I don't think that's true at all. I had warehouse experience before going into software and automated storage and retrieval systems (AS/RS) such as the mini-load type systems are vastly impactful. They are far superior to forklift based retrieval or worse still, human picking.

Humans are often really good at unstandardized tasks that require a lot of physical manipulation even when they're considered "unskilled" or at least blue collar. I think of so many things I pay people to do with respect to my house and the idea that autonomous hardware could handle pretty much any of them is pretty much laughable.

There are probably some exceptions. I can imagine robo lawnmowers getting good enough to handle my yard at some point if not adjacent cleanup. But doing a monthly housecleaning? Don't think so. (And a Roomba really wouldn't be very useful; it's barely useful at my brother's house that is practically built for a Roomba.)

I agree completely with this.

However, houses and things inside it were designed by humans for humans, for either aesthetic or 'easy for humans to do' reasons. I sometimes wonder then what if they were designed from the ground up with let's call it 'robot repairability'/automation in mind. I don't know exactly what that looks like, but I'd imagine a lot more modular, a lot more brutalist/uglier, and probably a lot more expensive. But it's an interesting thing to think about nonetheless.

I have a town home, so not ideal for a Roomba but...I have a Eufy now and having it (mostly) mop/(some) vacuum my kitchen/living room every night has been a huge improvement to keeping the floor clean (before something gets stuck on it between when I can do it on the weekend).

Chinese tech has come a long way in the last 5 years. Roomba just went bankrupt for a good reason.

I see a future where some humanoid robot mows the lawn, picks up trash, puts the bins outside weekly, vacuums the house, loads the dishwasher, does the clothes washing. While not actively doing those things, it will plug itself in and be passive security.

They will need help doing the finer control things, and they won't do everything. But definitely useful enough, especially for older folk.

Let’s see if this is the time for industrial robots,

Boston Dynamica: Majority Owner: Hyundai Motor Group (80%) Minority Owner: SoftBank (20%)

Factories already buy over $25 billion worth of industrial robots every year. Humanoid robots are neither going to displace specialized industrial robots, nor are they going to be particularly useful since industrial robots can lift entire car chassis, or perform precision work that would be impossible for a robot that can walk around. Surgical robots are neither humanoid, nor are they equipped with human like hands, for very good reasons. I don't know where people get the idea that humanoid robots are the dawn of the robot era. It's a blind alley, a dead end, impractical, un-competitive with specialized robots, and dangerous.

We've shaped the world to be useful to humans with tools used by and for humans. It's going to be very advantageous to have generally capable humanoid robotic platforms that can take advantage of all of the shortcuts and hacks and efficiencies we tailored to ourselves.

If we get there, and I think we are there now, then the worst case scenario is having to tediously implement the hundreds of thousands of little tasks and skills needed to be effective for a particular job.

The best case scenario is we run training videos for AI that gets cloned to fleets, and then you can deploy the equivalent of robotic Amish carpenters to build housing, or robotic warehouse operators, and you're paying a tenth of the cost with a hundredth of the hassle for the same work output as a human, and the efficiency and effectiveness only go up year over year, while human labor has more or less peaked.

I'd rather have a fleet of general purpose robots which I can put to any use within the human repertoire than technically more efficient and cheaper specialty robots that only perform singular tasks in an assembly line.

Humanoid robots have advantages industrial robots don’t: they fit where humans fit and can use tools humans use. They’ll fold your proverbial laundry with nothing more than their robot hands, then they’ll unpack your dishwasher and mow your lawn.

It would be great to see an example of this that simply isn't a human wearing a VR headset.

Even if it's unglamarous, there's loads of economic value in tele-operated humanoid bots. They can improve productivity via one pilot driving N robots and issuing high-level commands. They can also allow factories to import labor from 3rd world countries without bothering with the visa stuff. Providing complete manufacturing synergy using world-class, follow-the-sun shifts, minimizing disruptions and ensuring you meet your re-onshored manufacturing goals.

I don't see these being used outside of high earning households in gated communities. The same humans being exploited for their labor, whose earnings are hoovered up by the ultra-wealthy, barely have the discretionary income for food and clothes.

It's that old tinfoil hat theory that the Jetsons and the Flintstones took place in the same point in history, the Jetsons were in the sky with their mind-bending technology, all their needs met, meanwhile the Flintstones are down on the planet, working menial jobs wearing and eating literal scraps.

The common man will never see a household robot, that is unless they cobble together enough components that have been discarded by the haves to be used by the have-nots.

To the point of your statement, humanoid robots will certainly fill lots of niches, it'll be fascinating to see what becomes prevalent first: menial labor, agentic-type household assistance, tutoring the kids, walking grandma across the busy intersection, sex tasks, etc.

> The common man will never see a household robot

– for small durations of never.

> It's that old tinfoil hat theory that the Jetsons and the Flintstones took place in the same point in history, the Jetsons were in the sky with their mind-bending technology, all their needs met, meanwhile the Flintstones are down on the planet, working menial jobs wearing and eating literal scraps.

That’s the current situation. Not tinfoil hat needed.

I recently watched a short clip (1) of the comedians who followed Joe Rogan to Austin lamenting how bad of an idea it was.

Notably Shane Gillis described to Rogan:

Gillis: Yeah you got a driver and a body guard and do Karate, it’s fun. I’m walking around thinking “I’m going to get fucked up”

Rogan: Don’t walk around, gotta secure the perimeter.

This is real life today and both of these guys are either millionaires or incredibly popular comedians with significant amounts of cash to throw around.

If the distinction between these two people is that broad, you’re well past conspiracy territory.

I can tell you for a fact in the trenches of Chicago and Miami where I have a lot of transiently homeless friends, they are living way worse than the Flintstones because they don’t even have a community to rely on.

1: https://youtube.com/shorts/shYkz-dlLQs?si=prN07elAoX-jWmNs

Cool! When can I buy one? The year 2100? They don't exist yet in a way that can do any of these things.

Yes, they will unload from the specialized robot(dishwasher). They can be the glue in certain situations that are not common enough to design a better solution for. But rapid prototyping, AI and other tech will also make it faster and easier than ever to produce custom solutions for niche applications. The "human robots will take over" bros are thinking one step ahead but not two.

There's still plenty of assembly work being done by humans in automotive factories. Maybe it's not humanoid robots, but quadruped robots or something with more human-like agility. [Microfactories](https://www.youtube.com/watch?v=VNqmvIuzbR4) are an interesting shift in automotive manufacturing that could take advantage of these more dexterous and mobile robot form factors.

Generally, you want specialized robots only for tasks that can keep the robots busy most of the time. If you have a variety of low-volume tasks, you can't fill your factory with specialized robots for all of them. You need more general-purpose workers that can do all those different things. Right now those workers are humans. That's where humanoid robots could fill a role.

Even if specialized robots were at parity with specialized people (they are not in dimensions like dexterity), the big thing missing is the flexibility. You can on the fly change the production process, humans will be able to accommodate the changes (not at perfect speed, but still). For specialized robots you need stop and reprogram them.

I don't know why people think that legs automatically make everything more flexible. It boggles the mind.

Look at real flexible manufacturing systems to see how much of a bullshit idea that is: https://youtu.be/gUvE2eFH6CY

Everything is transported via the central stacker crane that is directly connected to every machine. You don't need legs. This just leaves the arms and here is the thing, you can just have two robot arms in the same robot cell and call it a day. The humanoid form factor adds nothing.

Also what makes you think you don't have to program the humanoid robots? Again, everyone seems to think that if you build a human shaped robot, human level intelligence will automatically come as a result of the shape of the robot. The moment you remove the head, the intelligence vanishes.

> It's a blind alley, a dead end, impractical, un-competitive with specialized robots, and dangerous.

What a shocking lack of imagination. Do you seriously think in a hundred years you'll still hold this opinion?

Wait, let me ask my Humane AI Pin... yeah, I'll think the same.

This is you. https://www.astralcodexten.com/p/heuristics-that-almost-alwa...

Unironic comparisons to Humane AI shows quite how uncalibrated you are. Not to mention you'd also likely be wrong about that on a 100 year time scale. Undoubtedly you'd have the same opinions for the Internet. Try to reason better, you can do it.

>Let’s see if this is the time for industrial robots

This is a bit unclear to me. Is this implying that it hasn't been the time for industrial robots?

Few things that go overlooked from someone that works in industrial robotics:

- note that google already has investment in industrial robotics with intrinsic ai (that is the confluence of OSRF, bot and dolly, and a few other robotics/ai companies)

- this partnership specifically seems to be focused on humanoid robotics, which is not taken seriously by industrial manufacturing folks

granted, I find it somewhat amusing that the head of Boston Dynamics got invited to give a keynote at NEURIPS 2016 or 2017 and admitted than Spot the Dog didn't really use any learning...

As a close follower of Boston Dynamics since my high-school years, I feel like in hindsight that was a good call, they've managed to progress everything else seemingly very well - which now gives them a head-start for an industry that surpasses any amount of money Boston Dynamics could ever get by many times. The data and capabilities around data would unlikely have brought similar scale of innovation, as other companies could now come up with.

Now instead, they're ahead of end2end integration of their section of robotics - especially with actual business customers that gave them a great head-start in figuring out how to integrate them as a product.

Makes sense. AFAIK BD were using a lot of traditional programming and outdated ml concepts and haven't really managed to hop on board the choo choo transformer train a la VLMs/VLAs.

Deepmind has its own robotics team, but doesn't get much done. Best folks leave

There was this one Black Mirror episode..

What could possibly go wrong? Lame, I know, but seriously can we have some good, not sinister news for a week or two? With no AI or evil tech-bro, or senile despot overtones if possible.

What’s with the negativity? Deepmind is one of the lesser evil companies. Remember Alphafold? If there’s someone that can do something useful for humanity, it’s Demis.

As I said, I was being lame. But I get real Terminator vibes when a very good robotics company joins forces with a company that might make those robots autonomous. Terminator and Asimov robot stories make me pessimistic.

It was a knee-jerk reaction as opposed to detailed analysis.

Oh great.

tbh I think Google should sell all the companies they bought because Google can't deliver any products anymore. There are product companies (hello OpenAI, Anthro) and then there are corporate companies. Google is the latter. Their attempts at making new products have been failure after failure recently:

Bard → Gemini → Jules → Antigravity (they can't even come up with a good product name)

Their GCP console is a mess. Wanna get a Gemini API? Good luck with that.

And where is the "internet through balloons"? Where's their quantum HPC? Where is half the stuff they demo every year to devs to prove that they're still relevant but never ship them? e.g., where is that smart glass they used in their demo last year that had Gemini in it and could analyze what you see? Where's their "calls any restaurant on your behalf" ML model they introduced many years ago?

Google has lost it, and to make things even worse, occasionally they poke at their successful products and googllify them too (YT likes disappearing, Gmail with Gemini, Google Search performing worse than Bing, etc.)

Without regard to the rest of your comment, with which I largely disagree, at least this one feature I can confirm still exists and works fine:

"Where's their "calls any restaurant on your behalf" ML model they introduced many years ago?"

It might not run on your LibreNut 9000, but it works on my Pixel.

This entire AI boom is happening because Google published their Attention Is All You Need paper and made transformers.

I thought antigravity was a great name, I assumed obliquely referencing xkcd

Reply on news.ycombinator.com