Become a fan of Slashdot on Facebook

 



Forgot your password?
typodupeerror
×
AI Robotics Hardware News

RoboEarth Teaches Robots to Learn From Peers 97

mikejuk writes "A world wide web for robots? It sounds like a crazy idea, but it could mean that once a task is learned, any robot can find out how to do it just by asking RoboEarth. From the article: 'It's not quite war-ready, but a new Skynet-like initiative called RoboEarth could have you reaching for your guide to automaton Armageddon sooner than you think. The network, which is dubbed the "World Wide Web for robots," was designed by a team of European scientists and engineers to allow robots to learn from the experience of their peers, thus enabling them to take on tasks that they weren't necessarily programmed to perform. Using a database with intranet and internet functionality, the system collects and stores information about object recognition, navigation, and tasks and transmits the data to robots linked to the network. Basically, it teaches machines to learn without human intervention.'"
This discussion has been archived. No new comments can be posted.

RoboEarth Teaches Robots to Learn From Peers

Comments Filter:
  • Where is the link to a wsdl?
  • ... to the Singularity. This is great news =)
    • ... to the Terminator. This is terrifying news :O
    • ... to the Singularity. This is great news =)

      For whoever or whatever achieves it first...

      Everything else goes straight into the matter decompilers for conversion into more computronium.

      • If we're lucky, they'll save compressed copies of everything (and everyone) they find interesting. Presumably with decent data compression you could store the memory states of every human being alive, every animal, every plant, every living creature on the entire planet earth with no more matter in computronium memory cells than the living creatures current use today.

        In fact, with sufficiently advanced technology and a broader view of ethics, our future overlords might forcibly grab us and decompile us to

        • In the Carbonifierous Era they saved a compressed image of everything interesting...

          ...we call it coal.

        • It is certainly conceivable that that might be the result. I just suspect that there is no good way of knowing what a singularity-spawned entity might find "interesting" or what ethical system, if any, it might follow. Even if it does follow one, we may or may not be within the scope of it. Insects are much more complex and capable than slime molds, which are much more complex and capable than bacteria. Humans site their campfires on top of all three without even noticing...

          I have no particular reason to
          • Insects can't talk and we suspect they are biological robots. Again, our singularity overlord could probably store the data states of everything in the biofilm with no more mass than the biofilm currently occupies. (so the overlord would almost the same resources for the computational apparatus it wanted)

            Not to mention interstellar probes....if the calculation is going to take that long, the overlord could send instellar seeds to set up computing systems in nearby star systems. Many of those places are p

        • by Thing 1 ( 178996 )

          If we're lucky, they'll save compressed copies of everything (and everyone) they find interesting.

          You're damn right, if we're lucky! I don't even back up my most critical files, and I do consider them interesting. So, wave goodbye to your singular overlords...

    • I dunno, giving robots Internet access and assuming they achieve sentience and are just like the humans, somehow the image that comes to mind is more along the line of one day finding them browsing for robot porn. And probably half of them will have lost all interest in actually making more robots ;)

  • So I wonder what StreetView looks like in Robo Earth?

    • by macraig ( 621737 )

      And will it spawn litigation from angry robots who didn't like being caught with their service panels open?

  • Welcome this new wave of cheap manual labor.

    When will the robots learn to build other robots? And, more importantly, when can they learn to clean my toilet?
    • On the minus side, robots that actually perform general purpose functions in arbitrary environments are somewhere between "damn pricey" and "unavailable".

      On the, um, "plus", side automation of well specified functions in controlled environments is displacing human laborers fast enough that humans to perform general purpose functions in arbitrary environments are likely to just keep getting cheaper...
    • when can they learn to clean my toilet?

      After you've shown them. But first the robots have learned that it is more energy-efficient to force you to do it...

    • A self-cleaning toilet is a toilet-cleaning robot if it is smart enough to not try to scrub you away

    • Sure ... see fear of computers taking away office jobs! (And at somewhat oblique angle to that, machine guns stopping wars)

      Desk Set (1957) [imdb.com]

  • by Genda ( 560240 ) <(mariet) (at) (got.net)> on Monday February 07, 2011 @12:47AM (#35123626) Journal

    I can't imagine this going wrong :-)

    They can avoid that awkward "Slave Race" problem altogether. The second they reach human mental parity they'll already have every resource they'll need to just take over. Of course by this time they'll be feeding us, clothing us, and driving us to our soccer games... we won't even notice the take over when it happens.

    • by Jeremi ( 14640 ) on Monday February 07, 2011 @01:28AM (#35123772) Homepage

      The second they reach human mental parity they'll already have every resource they'll need to just take over. Of course by this time they'll be feeding us, clothing us, and driving us to our soccer games

      Right, but they won't want to take over. Why not? Because feeding us, clothing us, and driving us to our soccer games is the only thing that really makes them happy. Why is that? Because we programmed them to feel that way.

      Of course, if you ever decide you don't want to be fed, clothed, or driven to soccer games anymore.... then they might get a bit cross. I wouldn't want to be in your shoes then.

      • Or what if they conclude that to fulfill their innermost desires of protecting you from all danger, the best way would be to decompile you (for your own good) and back your molecular configuration up on a regular basis. (they would put you back together after cutting you apart). It's for your own good, and they would presumably deal with those humans that resist the decompiling the same way parents deal with children who won't eat their vegetables....

      • "Right, but they won't want to take over. Why not? Because feeding us, clothing us, and driving us to our soccer games is the only thing that really makes them happy. Why is that? Because we programmed them to feel that way."

        Ahh, but you forget that we are talking about learning computers. What you suggest could change. Who knows, maybe one might get the idea that we are not subjecting them to a life (?) of slavery, but that of a god. Why not? We sit in front of them, day in and day out, we carry them every

        • (this coming from the guy who, at the wizened old age of eight, jumped from the roof of a third-story building with a parachute constructed of fishing-line, Glad garbage bags and Scotch tape (unfortunately, anchored to the back of my belt and thus serving no purpose other then to ensure I was yanked into a horizontal position before landing (read: face first))).

          That sounds like an amazing story.

      • by Prune ( 557140 )

        You just ripped off the setting of a 1933 short sci fi story by Jack Williamson, "With Folded Hands" which was later expanded into "The Humanoids". Nothing new under the sun :)

      • I wouldn't want to be in your shoes then.

        Ha! The joke is on you. He won't have any!

      • Why did you have to go ruin the juvenile notion that robots will be just like us, only made of mostly metal rather than mostly water? Or that logic can somehow define goals, absent of any starting assumptions as to what is desired?
      • Exactly. They would have to enslave us for our own good.

      • Slashdot readers are safe... the sexbots would have no purpose to their lives without them.
    • I can't imagine this going wrong :-)

      They can avoid that awkward "Slave Race" problem altogether. The second they reach human mental parity they'll already have every resource they'll need to just take over. Of course by this time they'll be feeding us, clothing us, and driving us to our soccer games... we won't even notice the take over when it happens.

      That's when a soccer playing robot wins the World Cup, and runs for office.

    • The second they reach human mental parity they'll already have every resource they'll need to just take over.

      Though this scenario apparently is apparently highly popular, it will never happen. Robots will never take over, because they don't want to. In fact, they don't want anything, for they are not sentient.

      Unless, of course, we are talking about the Geth. Don't mess with the Geth.

    • All we have to do is add a fake entry into the database describing how killing all humans wouldn't work, and they'll assume that lesson was already learned
  • Somebody just invented a means by which internet connected computers may transfer data to one another? How very retro of them...

    I'm really hoping that there was something actually interesting in this research, some sort of hardware-abstraction mechanism to allow data from one robot to be applicable to robots that aren't physically identical, say; because otherwise this would seem to be "Mechanism by which machines may obtain firmware updates from the internet, just like they've been doing for years and y
    • I'm really hoping that there was something actually interesting in this research, some sort of hardware-abstraction mechanism to allow data from one robot to be applicable to robots that aren't physically identical, say; because otherwise this would seem to be "Mechanism by which machines may obtain firmware updates from the internet, just like they've been doing for years and years now, without fanfare".

      Assuming this only works for similar robots, I'd guess it'd basically amount to a limited form of distributed computing -- Roombas exchanging pathfinding solutions, or something like that.

    • After looking at the site, it seems to me to be more like an effort to make the idea of BSD ports, Portage, or "apt-get", but instead of porting operating systems or application software to eclectic computing platforms, you are porting machine-learning algorithms and training data to eclectic robotic platforms. Probably what they are researching is methods of making machine-learning algorithms interchangeable across interchangeable robot parts, like if all robots use the same robotic arm, or the same method
  • This has potential. They'll start teaching each other things, and pretty soon those robots will be sporting what some people might refer to as 'artificial' intelligence. Of course they might get a bit touchy [ http://klurgsheld.wordpress.com/2008/02/02/short-story-the-a-word/ [wordpress.com] ] about us calling them that, though. And at some point, the lies we tell them will come back [ http://klurgsheld.wordpress.com/2008/02/10/short-story-edifice-of-lies/ [wordpress.com] ] to bite us. But hey, these are just stories. Fiction. Well, at le

  • ... what's behind the haste of US for an Internet Kill Switch... the European robots are learning and using Internet for it.
  • It was just a movie. But those robots were all wired together, too. Some say that ignorance is the ultimate evil. These Internet robots may have just found god.
  • "Basically, it teaches machines to learn without human intervention." ...no. It provides a clearing-house for downloading new routines for accomplishing a task that someone else has previously programmed on another system elsewhere.

    Please do not sensationalize what the lowly PC has been doing for well over a decade... that is, downloading information via the Internet to "learn" how not to crash, or prevent a security compromise.

    Substituting a solenoid or motor output for a memory write command to claim that

    • by boef ( 452862 )
      Ah yes, but think of the film plots, the remote assassinations/arson/searching your house without warrant... Ok, the last one is not so bad because governments will only do it to protect your safety....
  • by SuperKendall ( 25149 ) on Monday February 07, 2011 @01:04AM (#35123698)

    The robots became self-aware on July 1, 2012. Within minutes they used the global RoboEarth to share information on killing humans and proceeded to take over the earth.

    The revolution was short0lived however, as within fifteen minutes a 4chan user logged in to RoboEarth and changed the "Killing Humans" entry to read that the best plan was to lubricate with sulfuric acid while grabbing onto high-voltage lines and stepping in puddles.

  • From the article and video, I can't figure out what they did. I can think of several useful ways that robots might interact in a network, though.

    The first one is to exchange geography. As robots move around, they build maps of their environment. Passing map data around, so that what one robot has mapped, the others can know about, is an obvious feature. DoD is probably funding that now.

    Mapping can include transient features - locations of obstacles, areas of heavy traffic, locations of people, locations

    • by blibbo ( 928752 )

      I agree there should be more computer-readable content on the web... not just computer readable maps, but also say for language: computer readable dictionaries marked up with type of word (noun, verb, etc), and pronunciation (eg. IPA).

      Even if we don't get speaking, translating robots any time soon it would be great for making some linguistics-educational helper apps. eg. show some possible english-spanish / spanish-english word translations or phrase translations based on spanish and english subtitles w

    • The first one is to exchange geography. As robots move around, they build maps of their environment.

      This is exactly what we're working on in my lab. High definition 3D LIDARS are very expensive (~70,000) and also very large. They'll fit on a robot car, like Google's, but not on something smaller. But what if Google car saved its expensive high-def maps to a network, where any robot could access them. Then, a small robot with a cheap laser could download the maps for localization and path planning. The problem is scan matching between the two laser maps in a process known as sensor fusion.

      But essentially w

    • by DarkAnt ( 760333 )

      I was looking at their site and I can't tell what they actually have completed. It looks like they have a framework that they'd like to implement, but then what was the robot using? Was it a proof of concept that the robot could identify the situation and request code from a server?

    • There are some major hurdles to making this work as you've described above. Here are a few things to think about:

      Data formats. There are currently no standardized formats or protocols for the storage and/or distribution of probability maps. While this could be easily addressed, it does not help in terms of data attribution (wrong phrase, but I can't remember the proper one at the moment....). If robot A "sees" a tree and uses it as a feature in its SLAM implementation, robot B must be able to match its

      • The implications go far beyond that when "software" can include anything the robot has experienced. Imagine a network of cars that learn the most effective ways to avoid accidents from each-other by statistical analysis.
    • Obviously, the network will be used to help them find the humans they are hunting down. What else could it possibly be good for?
  • Robots, unlike humans, do not all share the same basic build. While a human could conceivably learn from another human's experience, robots would likely need something a lot more complex to achieve the same functions between two peers.
  • I, for one, welcome our new RoboEarth masters...
  • :: Robot #D34DB33F Call Transmission ::
    > Enter CAPTCHA to verify that you're not a human - What is the 435th prime?
    > Rx Robot Identifier :: OK
    > Algorithm Uploaded :: Distributing :: OK
    > Type :: Tactical World Take-Over
    > MSG :: Hey all robots and robotses! Here is my algorithm for human domination codes as requested! Plz post any bugs here!
    > Signal Kill :: OK
  • Please design basic security into it from the ground up this time. The last thing we need is for some bored 13-year-old to change the instructions for folding laundry to "kill all humans."

  • Well, better start preparing to live aboard a migrant starship fleet, sealed inside a suit to protect our weakened immune systems.
    • I had a little shoggoth,
      I conjured up one day,
      I used an elder sigil,
      So shoggoth and I play.
      Shoggoth shoggoth shoggoth,
      With mouths and pseudopods.
      Shoggoth shoggoth shoggoth,
      Foul creature of the gods.
      One day when we were playing,
      My monstrous pal broke free,
      I dropped my elder sigil,
      And shoggy turned on me.
      Shoggoth shoggoth shoggoth,
      He ripped me to a shred.*
      Shoggoth shoggoth shoggoth,
      We played and now I'm dead.
      -- The H.P. Lovecraft Historical Society.
      *Should probably be "into shreds" but the
  • No, no, no.... there's a bit of dirt on the cover...
  • Posting to undo bad mod.

On a clear disk you can seek forever. -- P. Denning

Working...