What a Google Exec Learned After 7 Years Trying to Give AI a Robot Body (axios.com) 33
Wired published some thoughts from Hans Peter Brondmo, the former head of "Google's seven-year mission to give AI a robot body".
An anonymous reader shared this report from Axios: Building AI-powered robots that can flexibly operate in the real world is going to take much longer than Silicon Valley believes and promises, according to the former head of Google's robotics moonshot project, writing in Wired...
Everyday Robotics spent seven years and a small Google fortune developing a one-armed robot on a wheeled platform. By the time Google pulled the plug on the project in February 2023, the robots were helping clean up researchers' desks and sorting trash during the daytime; in the evening, they were improvising dances. [Google hired a professional dancer as an artist-in-residence who teamed with "a few other engineers" to build an AI algorithm trained on the dancer's choreography preferences...]
Google founder Larry Page — favored moving directly to "end to end" (e2e) learning, where you'd hand robots a general task and they'd be able to figure out how to execute it. That, Page felt, was a goal worthy of a moonshot. But it also turned out to be out of reach. "I have come to believe," Brondmo writes, "it will take many, many thousands, maybe even millions of robots doing stuff in the real world to collect enough data to train e2e models that make the robots do anything other than fairly narrow, well-defined tasks...." ["Building robots that perform useful services — like cleaning up and wiping all the tables in a restaurant, or making the beds in a hotel — will require both AI and traditional programming for a long time to come. In other words, don't expect robots to go running off outside our control, doing something they weren't programmed to do, anytime soon."]
The bottom line: So far, robot hype is outpacing robot reality. Boston Dynamics' back-flipping humanoid and quadruped bots have wowed YouTube viewers — but you wouldn't want to let them anywhere near your office or home.
It's an interesting look back. "My job: help figure out what to do with the employees and technology left over from nine robot companies that Google had acquired," Brondmo writes: Andy "the father of Android" Rubin, who had previously been in charge, had suddenly left. Larry Page and Sergey Brin kept trying to offer guidance and direction during occasional flybys in their "spare time...." I knew from firsthand experience how hard it was to build a company that, in Steve Jobs' famous words, could put a dent in the universe, and I believed that Google was the right place to make certain big bets. AI-powered robots, the ones that will live and work alongside us one day, was one such audacious bet.
Eight and a half years later — and 18 months after Google decided to discontinue its largest bet in robotics and AI — it seems as if a new robotics startup pops up every week. I am more convinced than ever that the robots need to come. Yet I have concerns that Silicon Valley, with its focus on "minimum viable products" and VCs' general aversion to investing in hardware, will be patient enough to win the global race to give AI a robot body. And much of the money that is being invested is focusing on the wrong things...
When I arrived, the lab had already hatched Waymo, Google Glass, and other science-fiction-sounding projects like flying energy windmills and stratospheric balloons that would provide internet access to the underserved... [But] in January 2023, two months after OpenAI introduced ChatGPT, Google shut down Everyday Robots, citing overall cost concerns. The robots and a small number of people eventually landed at Google DeepMind to conduct research. In spite of the high cost and the long timeline, everyone involved was shocked.
They'd tackled the problem with earnestness. ("[S]even robots working for months to learn how to pick up a rubber duckling? That wasn't going to cut it... So we built a cloud-based simulator and, in 2021, created more than 240 million robot instances in the sim.ma")
Brondmo adds this his mother had advanced Parkinson's disease, and hoped that one day robots could support her. "Our frequent conversations toward the end of her life convinced me more than ever that a future version of what we started at Everyday Robots will be coming. In fact, it can't come soon enough.
"So the question we are left to ponder becomes: How does this kind of change and future happen? I remain curious, and concerned."
An anonymous reader shared this report from Axios: Building AI-powered robots that can flexibly operate in the real world is going to take much longer than Silicon Valley believes and promises, according to the former head of Google's robotics moonshot project, writing in Wired...
Everyday Robotics spent seven years and a small Google fortune developing a one-armed robot on a wheeled platform. By the time Google pulled the plug on the project in February 2023, the robots were helping clean up researchers' desks and sorting trash during the daytime; in the evening, they were improvising dances. [Google hired a professional dancer as an artist-in-residence who teamed with "a few other engineers" to build an AI algorithm trained on the dancer's choreography preferences...]
Google founder Larry Page — favored moving directly to "end to end" (e2e) learning, where you'd hand robots a general task and they'd be able to figure out how to execute it. That, Page felt, was a goal worthy of a moonshot. But it also turned out to be out of reach. "I have come to believe," Brondmo writes, "it will take many, many thousands, maybe even millions of robots doing stuff in the real world to collect enough data to train e2e models that make the robots do anything other than fairly narrow, well-defined tasks...." ["Building robots that perform useful services — like cleaning up and wiping all the tables in a restaurant, or making the beds in a hotel — will require both AI and traditional programming for a long time to come. In other words, don't expect robots to go running off outside our control, doing something they weren't programmed to do, anytime soon."]
The bottom line: So far, robot hype is outpacing robot reality. Boston Dynamics' back-flipping humanoid and quadruped bots have wowed YouTube viewers — but you wouldn't want to let them anywhere near your office or home.
It's an interesting look back. "My job: help figure out what to do with the employees and technology left over from nine robot companies that Google had acquired," Brondmo writes: Andy "the father of Android" Rubin, who had previously been in charge, had suddenly left. Larry Page and Sergey Brin kept trying to offer guidance and direction during occasional flybys in their "spare time...." I knew from firsthand experience how hard it was to build a company that, in Steve Jobs' famous words, could put a dent in the universe, and I believed that Google was the right place to make certain big bets. AI-powered robots, the ones that will live and work alongside us one day, was one such audacious bet.
Eight and a half years later — and 18 months after Google decided to discontinue its largest bet in robotics and AI — it seems as if a new robotics startup pops up every week. I am more convinced than ever that the robots need to come. Yet I have concerns that Silicon Valley, with its focus on "minimum viable products" and VCs' general aversion to investing in hardware, will be patient enough to win the global race to give AI a robot body. And much of the money that is being invested is focusing on the wrong things...
When I arrived, the lab had already hatched Waymo, Google Glass, and other science-fiction-sounding projects like flying energy windmills and stratospheric balloons that would provide internet access to the underserved... [But] in January 2023, two months after OpenAI introduced ChatGPT, Google shut down Everyday Robots, citing overall cost concerns. The robots and a small number of people eventually landed at Google DeepMind to conduct research. In spite of the high cost and the long timeline, everyone involved was shocked.
They'd tackled the problem with earnestness. ("[S]even robots working for months to learn how to pick up a rubber duckling? That wasn't going to cut it... So we built a cloud-based simulator and, in 2021, created more than 240 million robot instances in the sim.ma")
Brondmo adds this his mother had advanced Parkinson's disease, and hoped that one day robots could support her. "Our frequent conversations toward the end of her life convinced me more than ever that a future version of what we started at Everyday Robots will be coming. In fact, it can't come soon enough.
"So the question we are left to ponder becomes: How does this kind of change and future happen? I remain curious, and concerned."
They should have mixed and matched (Score:2)
AI is just one tool in the dev toolkit.
I'd have started with a robot that's already on the market and functional, then added AI extensions. Have optical AIs for object recognition, facial recognition, and whatever else. Train them separately, connect them on the back end to a supervisor.
Do the same for speech.
Re: (Score:2)
That's a reasonable plan.
But it would take time and fail because Google's top management has no patience.
Re: They should have mixed and matched (Score:2)
Re: (Score:2)
Maybe it's an underpants thing ???
Color me unsurprised (Score:1)
Obviously, claiming things are easy for computers or robots or AI is a main ploy to get people to buy them. Also obviously, things are not easy for them and most things that humans consider "easy" (even the average, dumb human) are entirely out of reach.
Re: (Score:2)
claiming things are easy for computers or robots or AI is a main ploy to get people to buy them.
Some robots work on an "employee" model.
A business can "hire" them to work in a warehouse or factory and then pay the robot provider based on the work done, such as per order picked or per widget assembled.
If the robot doesn't work, the provider doesn't get paid until they fix the problem.
Re: (Score:2)
You have the right idea, but you also have to make the robot company pay a fine every time their robot is late or produces garbage outcomes. That will change the equation substantially and make it unlikely any robot provider will want to do business with half assed solutions.
Alphabet is the new Xerox (Score:3)
All this article does is tell me that Google did a fuckton of research and development, and then gave up right as the robot revolution started to take off. Along those lines, I've been waiting for Waymo to be cancelled, frankly.
It is kind of crazy the stuff that Google has developed and then failed to successfully commercialize, because they just threw in the towel before it could gain traction, or never let it out of the lab because they were being conservative, or just plain couldn't get their shit together. I realize that there's this inherent conflict between their main business model (search) and everything else (not-search) and that not-search had been the red-haired stepchild... up until the point where all the focus on AI started to become an existential threat to their search business.
For example, the current AI boom. Google should have owned this. The seminal paper on transformers ( Attention Is All You Need: https://proceedings.neurips.cc... [neurips.cc] ) came from Google (at time of publication 6 of the 8 authors were directly Google affiliated, the 2 that were not contributed their work while at either Google Brain or Google Research). They had their own custom designed TPU farms ( starting in 2015 - https://en.wikipedia.org/wiki/... [wikipedia.org] ) in order to make their search better (this was prior to decisions to cripple product functionality for better revenue metrics). And yet?
https://www.latent.space/p/ade... [www.latent.space]
"Itâ(TM)s often asked how Google had such a huge lead in 2017 with Vaswani et al creating the Transformer and Noam Shazeer predicting trillion-parameter models and yet it was Davidâ(TM)s team at OpenAI who ended up making GPT 1/2/3."
Going back further. Everybody laughed at Google's inept efforts to chase the social media trend. Apparently there were 5 different efforts:
https://www.androidpolice.com/... [androidpolice.com]
Google Wave, Google Buzz, Orkut, Google +, and something called Keen (which I've never heard of).
They shut down Google+ in 2019. During the same time period, Facebook, the leading social media platform, was suffering from people bailing because of the Cambridge Analytica scandal:
https://www.nytimes.com/2018/0... [nytimes.com]
It still boggles my mind that Google + wasn't able to make inroads when people were searching for an alternative.
And Stadia... man...
Re: (Score:1)
They are the modern the Bell Labs.
They have monopoly money.
However projects have to make monopoly money, not razor thin margins. Compared to their ad business, nothing makes much sense.
Re: (Score:3)
It still boggles my mind that Google + wasn't able to make inroads when people were searching for an alternative.
By 2019 Google had eroded any bit of trust or goodwill it ever had with most people. Anyone realizing that Facebook was a problem wasn't looking to jump into bed with Google. I'm not at all surprised that they failed with social media.
Re: (Score:2)
Their search is a cost center and an intelligence gathering API where they spy on what people want, so it can be monetized via advertising auctions (that's their profit center). Robots do not necessarily fit into this business model. LLMs probably fit into this if they can be made to spruik products during conversations with users, ie if the LLMs become "travelling salesmen" or "influencers".
Was this before or after LLMs? (Score:2)
Before LLMs and the success of those models, the world of robotics was different. The downturn right before ChatGPT and AI fueled funding probably killed so many projects.
Re: (Score:3)
From the article:
"We were, in other words, on the cusp of truly capitalizing on the biggest bet we had made: robots powered by AI. AI was giving them the ability to understand what they heard (spoken and written language) and translate it into actions, or understand what they saw (camera images) and translate that into scenes and objects that they could act on. And as Peterâ(TM)s team had demonstrated, robots had learned to pick up objects. After more than seven years we were deploying fleets of robots
Hmm ... (Score:2)
What a Google Exec Learned After 7 Years Trying to Give AI a Robot Body
Dating an actual person would be (a) harder or (b) easier? :-)
All i know (Score:2)
Re: (Score:2)
Re: All i know (Score:2)
I'm confused, and maybe need to RTFA... (Score:2)
"I have come to believe," Brondmo writes, "it will take many, many thousands, maybe even millions of robots doing stuff in the real world to collect enough data to train e2e models that make the robots do anything other than fairly narrow, well-defined tasks...."
Which made me think of the simulated world that NVIDIA offered/used to train AI networks for stuff like this, allowing greater parallelism and zero broken robot parts when they mess up. Doubt it'd be perfect or the only training needed, but then la
Re: (Score:2)
When you stick a robot trained from a simulation of a subway ride in a real NYC train, you get failure to meet design requirements.
This has been the story of simulation over the last 50 years in all fields where it has been tried. AKA "black swan".
this man (Score:2)
wants to fuck a robot
Re: this man (Score:2)
At least Google's Body-types were Creative (Score:2)
IIRC, at the 2023 AI For Good Conference , of the 9 robots attending, all were humanoid, only one was a guy, and half of the remainder presented as hawt babes in their mid 20s
Bullshit (Score:2)
First build a robot with the hand dexterity of a human, then discuss software failure.
Just give all robots a connection (Score:1)
To a single, central AI where they can upload all of the experiences they have in the real world.
Re: (Score:2)
Story illustrates the failure of so-called 'AI' (Score:2)
If you want real 'AI', you need general AI, but we lack the understanding of how our own brains do that.
The current 'approach' to that seems to be a version of the million-monkeys-with-typewriters method; spoiler alert: that ain't gonna work.
Re: (Score:2)
Re: (Score:2)
Hype outpaces relaity - who knew? (Score:2)
"The bottom line: So far, robot hype is outpacing robot reality".
Who knew? In our world hype always outpaces reality - that's where the money is.
No one becomes a billionaire by being strictly honest and realistic about their prospects. Just as no one ever gets elected on a platform that is honest and achievable. To be elected you MUST promise far more than could possibly be delivered.
sf (Score:2)
Turns out it's hard to implement science fiction.
Awww... (Score:2)