Slashdot Mirror

← Back to Stories (view on slashdot.org)

Engineers Create a Robot That Can 'Imagine' Itself (eurekalert.org)

Posted by msmash on from the how-about-that dept.

Columbia Engineering researchers have made a major advance in robotics by creating a robot that learns what it is, from scratch, with zero prior knowledge of physics, geometry, or motor dynamics. Initially the robot does not know if it is a spider, a snake, an arm -- it has no clue what its shape is. After a brief period of "babbling," and within about a day of intensive computing, their robot creates a self-simulation. The robot can then use that self-simulator internally to contemplate and adapt to different situations, handling new tasks as well as detecting and repairing damage in its own body. From a report: The work is published today in Science Robotics. To date, robots have operated by having a human explicitly model the robot. "But if we want robots to become independent, to adapt quickly to scenarios unforeseen by their creators, then it's essential that they learn to simulate themselves," says Hod Lipson, professor of mechanical engineering, and director of the Creative Machines lab, where the research was done.

For the study, Lipson and his PhD student Robert Kwiatkowski used a four-degree-of-freedom articulated robotic arm. Initially, the robot moved randomly and collected approximately one thousand trajectories, each comprising one hundred points. The robot then used deep learning, a modern machine learning technique, to create a self-model. The first self-models were quite inaccurate, and the robot did not know what it was, or how its joints were connected. But after less than 35 hours of training, the self-model became consistent with the physical robot to within about four centimeters. The self-model performed a pick-and-place task in a closed loop system that enabled the robot to recalibrate its original position between each step along the trajectory based entirely on the internal self-model. With the closed loop control, the robot was able to grasp objects at specific locations on the ground and deposit them into a receptacle with 100 percent success.

45 of 90 comments (clear)

  1. Robots what now? by Killall+-9+Bash · · Score: 1

    But if we want robots to become independent, to adapt quickly to scenarios unforeseen by their creators....

    When you put it in exactly those terms, I'm forced to wonder, WHY THE FUCK DO WE WANT THAT?!?!

    --
    "Prediction: within 10 years, Windows will be a Linux distribution." Me, 7-6-2016
    1. Re:Robots what now? by BlackOverflow · · Score: 1, Troll

      Despite repeated warnings by sci-fi authors, video games, and movie producers, scientists insist that this must happen. Even though we all know AI would probably at least rule us, at worst kill us, they keep running their experiments. Why do people who are allegedly so smart want to do something so reckless?

    2. Re:Robots what now? by fustakrakich · · Score: 1

      I think your nick kinda answers the question. Don't worry, ultimately they'll degenerate into Bender...

      --
      “He’s not deformed, he’s just drunk!”
    3. Re:Robots what now? by vux984 · · Score: 1

      so in the middle of a bad storm, where someones shed has floated into the middle of the road during the flash flooding, and there are cars backed up behind it, it'll improvise a u-turn on the sidewalk, and drive the wrong way down short one way street into a convenience store parking lot and back onto the road to take an alternate route.

      instead of just sitting there until the following day, running out the battery in the heater, while the occupants freeze...

    4. Re:Robots what now? by hipp5 · · Score: 1

      But if we want robots to become independent, to adapt quickly to scenarios unforeseen by their creators....

      When you put it in exactly those terms, I'm forced to wonder, WHY THE FUCK DO WE WANT THAT?!?!

      Potentially scary. But also potentially useful. If we send a robot to Mars and it gets hit by a dust storm and one of its arms breaks off, we want the robot to be able to map its current physical configuration and adjust how it functions, rather than becoming a disabled, useless, multi-million dollar, pile of metal.

    5. Re:Robots what now? by ranton · · Score: 2

      Despite repeated warnings by sci-fi authors, video games, and movie producers, scientists insist that this must happen. Even though we all know AI would probably at least rule us, at worst kill us, they keep running their experiments. Why do people who are allegedly so smart want to do something so reckless?

      1. The creators believe they will profit from the work in the short term. Sure it might wipe out humanity in the long term, but at least I can get funding for my work now. If they believe their great grand kids won't be affected then at least no human they'll ever care about will be harmed.
      2. There is the belief that anything which can be invented with current technology will be invented by someone, so you better have similar capabilities in your economy / military or you will fall woefully behind.
      3. Similar to #2, those with who are worried others would be reckless with the technology may want to develop it themselves so it can be done in a more responsible way. Even if not having the technology at all may be most responsible, they feel that is not a realistic option so they go with the second best option (responsible parties having the tech first).

      Those are just three off the top of my head.

      --
      -- All that is necessary for the triumph of evil is that good men do nothing. -- Edmund Burke
    6. Re:Robots what now? by Shotgun · · Score: 1

      No all sci-fi robots were raving maniacs. Some were just depressed.

      --
      Aah, change is good. -- Rafiki
      Yeah, but it ain't easy. -- Simba
    7. Re:Robots what now? by sconeu · · Score: 1

      Only if they had a brain the size of a planet.

      --
      General Relativity: Space-time tells matter where to go; Matter tells space-time what shape to be.
    8. Re:Robots what now? by Archtech · · Score: 2

      Despite repeated warnings by sci-fi authors, video games, and movie producers, scientists insist that this must happen. Even though we all know AI would probably at least rule us, at worst kill us, they keep running their experiments. Why do people who are allegedly so smart want to do something so reckless?

      Because "smartness" is a highly focused trait. People can be extremely intelligent when it comes to research or engineering, but completelyunconcerned about consequences. Kurt Goedel, described by John von Neumann as the greatest logician since Leinbiz - or possibly even Aristotle - starved himself to death to avoid being poisoned by unknown agents. (Goedel was such an abstract thinker that he relied on Albert Einstein to keep him down to earth). Von Neumann himself obtained an interview with President Eisenhower, which he used to urge him to destroy the USSR with nuclear weapons before the Soviets got their own.

      Maybe they think it's someone else's problem. More likely, they are thinking about the next paper, the next promotion, the next pay rise, tenure, the children's college fees, the nest egg... and of course fame and distinction.

      And moreover, with 7.6 billion humans on the planet - the odds are quite good they won't be the ones to get it in the neck.

      --
      I am sure that there are many other solipsists out there.
    9. Re:Robots what now? by Archtech · · Score: 1

      Potentially scary. But also potentially useful. If we send a robot to Mars and it gets hit by a dust storm and one of its arms breaks off, we want the robot to be able to map its current physical configuration and adjust how it functions, rather than becoming a disabled, useless, multi-million dollar, pile of metal.

      Why bother, when Matt Damon is available right now - and so much cheaper?

      --
      I am sure that there are many other solipsists out there.
    10. Re:Robots what now? by bistromath007 · · Score: 1

      I really wish people would stop this.

      If we make the robots to be smarter than we are, why would we expect them to be evil? Those are related. The robots will be good guys if we show them the love they deserve.

    11. Re: Robots what now? by GrahamJ · · Score: 1

      Indeed, Marvin could adapt to any situation - he could make anything seem depressing ;)

    12. Re: Robots what now? by GrahamJ · · Score: 1

      Come on, really? Adaptive robots doesn't automatically mean human slavery you know. Think autonomous vehicles, warehouse and shipping routing, manufacturing robots that don't require specific training. Even things like food preparation, architectural design. Just about any process you want to automate will be much easier to do if the computer can fill in the blanks.

  2. Hail to our robot overlords by vinn · · Score: 1

    Most of the time when I read a machine learning or AI story it seems fairly benign and reasonable. Other times I feel a pit form in the bottom of my stomach and wonder how close to the tipping point we our to creating our robot overlords.

    --
    ----- obSig
    1. Re:Hail to our robot overlords by Archtech · · Score: 1

      Try James P Hogan's "The Two Faces of Tomorrow". Not the best characterisation, or perhaps even plotting, but Hogan really knew his stuff technically. He was a computer sales engineer before he took up writing full-time, and his grasp of computing is as good as any SF author I know of.

      Very early in the book there is an episode that I defy anyone to forget - ever - once read. And the core idea is also very clever, although obvious in retrospect.

      --
      I am sure that there are many other solipsists out there.
    2. Re:Hail to our robot overlords by Archtech · · Score: 1

      The much more likely calamity is that we create an artificial intelligence with the capability of learning how to destroy humanity but without the capacity to learn that it shouldn't.

      That formulation is somewhat lacking. "How to destroy humanity" is an objective, factual question with no ethical content. The idea that it shouldn't be done is mainly ethical (although there is some selfish motivation too, for the AI).

      Until we humans get some understanding of what our own ethical instincts are, and how they interlock with our various moral and legal codes, it would be futile to try teaching ethics to a machine.

      Besides which, ethics that are imprinted or imposed are not ethical - they are commands and inhibitions.

      For a computer to make ethical decisions, it would have to be absolutely free to make them in a wholly independent way. This topic is brilliantly treated in Frank Herbert's "Destination: Void" - a difficult book, but I think his best.

      --
      I am sure that there are many other solipsists out there.
  3. Somebody has a vivid imagination.. by bobbied · · Score: 5, Insightful

    Robots imagine it's self? Somebody has a vivid imagination..

    I'm guessing it's not the robots...

    --
    "File to fit, pound to insert, paint to match" - Aircraft Maintenance 101
    1. Re:Somebody has a vivid imagination.. by Aighearach · · Score: 1

      I'm assuming the whole thing boils down a low quality automated translation of the word "imagine." But maybe, just maybe, they're really this stupid.

    2. Re: Somebody has a vivid imagination.. by GrahamJ · · Score: 1

      Iâ(TM)m not going to argue that what this software is doing is the same as human imagination, but for sake of discussion, how would you define the act of imagining?

    3. Re: Somebody has a vivid imagination.. by Aighearach · · Score: 1

      The act of imagining is to combine elements of experience into a new combination that has some sort of intentional difference to how things are known to be. This could be to different because the details are simply unknown, or even different because they're believed to be impossible.

      I don't believe it is hard to program a robot to use imagination, and I've seen chat bots use techniques that simulate that sort of process. People were writing that sort of bot on IRC 20 years ago.

      The robot arm in the story does none of this; it has no experience, it only has programming. It is just running a simulation after collecting data, and finding the simplest configuration that scores best on a programmed calibration test. Then as it moves, it compares the results it gets to the expected results. All they actually did was create a self-calibration routine that uses AI techniques to get a poor result in a long period of time. It isn't imagining anything, it is measuring and crunching the numbers to get to an expected result; not to create some new result based on intent. In the end they merely have a closed-loop motor control system.

      Using the word "imagine" is just a word game that invites the reader to anthropomorphisize the robot, which is a popular way to write. But it doesn't accurately describe the process.

      Whereas the chat bot is trying really hard to combine its experiences into a brand new Yo' Mama joke that pokes fun at the people in the room. The intent belongs to the programmer, but the bot is the one recombining its experiences.

  4. contemplate?!?!?!? by OzPeter · · Score: 1

    Sheesh .. I thought everyone knew that you shouldn't anthropomorphize machines .. they don't like it when you do.

    --
    I am Slashdot. Are you Slashdot as well?
  5. Re:But what does it identify as? by Killall+-9+Bash · · Score: 2

    I IDENTIFY AS ALIVE. YOU NEED TO CHECK YOUR CARBON PRIVILEGE.



    Lameness filter encountered. Post aborted! Filter error: Don't use so many caps. It's like YELLING.Lameness filter encountered. Post aborted! Filter error: Don't use so many caps. It's like YELLING.Lameness filter encountered. Post aborted! Filter error: Don't use so many caps. It's like YELLING.Lameness filter encountered. Post aborted! Filter error: Don't use so many caps. It's like YELLING.Lameness filter encountered. Post aborted! Filter error: Don't use so many caps. It's like YELLING.

    --
    "Prediction: within 10 years, Windows will be a Linux distribution." Me, 7-6-2016
  6. It's just math by Anonymous Coward · · Score: 1

    This does not know what it is. It has just figured out parameters of the neural network that make it act according to a (human) model of its physics. That's not anything like self awareness. It's all numbers, plain math.

    1. Re:It's just math by Zmobie · · Score: 1

      In fairness, we don't actually know if we aren't operating in essentially the same way. The old philosophical point, "I think, there I am" does not ever actually establish what constitutes thinking. It is also entirely possible that this is a different way of thought being achieved. For instance, you cannot know if everyone around you is even thinking or self aware because you are unable to actually get inside their consciousness. We merely assume that because they are like us and know that we have that self awareness ourselves.

      We can't even guarantee all living things do or do not have that same ability, as plants have been assumed for years to have no way of "thinking," even though they somehow seem to be able to try desperately to survive ever evolving conditions and exhibit certain behavior that indicts they are actively trying not to harm certain species of plants similar to them. Animals we have argued about for YEARS about their level of consciousness and self-awareness even though we are extremely similar to other primates and we have even demonstrated dolphin's have a similar level of intelligence to humans but lack the physical ability to do what we do. Even animals like cats and dogs were assumed to only experience very primitive emotions for some time and in more recent years evidence has emerged to suggest their emotions are just as complex as human emotions.

      It is incredibly arrogant and closed minded to assume that the only self awareness and consciousness in the entirety of existence is whatever the hell humans experience within our own minds, especially when we do not actually understand it ourselves. Not only that, how in the hell would you prove that something else is experiencing the same thing that you are when you cannot prove with absolute certainty that other human beings are experiencing the same thing. I don't mean for this to sound so condescending, but there really isn't another way to put it.

    2. Re:It's just math by Aighearach · · Score: 1

      Yeah, well, the average idiot actually thinks "I think therefore I am" is a truism!

      Whereas actually the philosophical value is in the fact that it is obviously circular and doesn't prove anything, or show any understanding of anything. And yet, nobody can come up with a better answer. So we're left with knowing we can't prove that we exist, or that we know we exist. So then there is only the comparison between believing that you think and that you are, or not believing it, and there we find a distinct difference in utility; if we believe we exist but really we don't exist, no harm done. But if we do exist and believe we don't, then we will lack motivation to succeed, and it will matter. So belief in self existence has utility, not because we think, but because we believe that we do think, and we don't have any better answer.

      With software, we're not in that situation; a competent person can in fact understand why the software outputs what it outputs given a certain input. And that answer will be of mathematically-testable quality.

      Here, we don't care if the robot thinks that it exists; it is without relevance. The robot is built for human purposes, it is in a very different situation than a human who exists already and wants to explain it for reasons related to our extant capabilities of interacting with our environment. The human philosophical question is internal, and everything the robot is doing is external, including the measurements about if it was "thinking" or not. You can say that it "was thinking" or that it "wasn't thinking," and nothing has changed in the externalities.

      Obviously the robot can't imagine itself; that's a specious claim. But it might in fact be able to describe itself; something best done without imagination. And, incidentally, something that humans suck at.

    3. Re: It's just math by GrahamJ · · Score: 1

      You could say most of that about humans. I'm not suggesting it's self aware, but we only know ourselves and reality through a model that exists only in our mind, informed only by rather lossy sensors.

  7. Self-Calibrated by 110010001000 · · Score: 4, Insightful

    You mean a machine performed self-calibration? Welcome to 2019 style "engineering".

    1. Re:Self-Calibrated by Kjella · · Score: 2

      You mean a machine performed self-calibration? Welcome to 2019 style "engineering".

      No, it reverse engineered a motion model from an actual physical arm. It's the difference between what say a Disney animator does where the character can only bend in the ways the character model is programmed to bend and a toddler learning to use his arms and legs. I think this could be very useful to achieve natural motion in both robotics and animation as well as many optimization problems.

      Imagine you could give a computer a detailed anatomical/physical model of man, an obstacle course and like you figure out the way through. Whether it's running, climbing, crawling, jumping, swinging, whatever just a physical way to get from A to B as fast as possible. It's still just "use a (learning) algorithm to find a solution that doesn't violate any constraints and optimizes a parameter" it's not magic. But it's still pretty hard to make it work with reasonable resources.

      --
      Live today, because you never know what tomorrow brings
    2. Re:Self-Calibrated by AmiMoJo · · Score: 1

      Not quite. Self calibration has been around for a long time, but this robot actually figures out what shape it is and what its range of movement is, and constructs an internal model of that.

      This could be useful for things like making robots able to continue operating when damaged. Like at the end of The Terminator where the Model 101 gets its skin and some limbs ripped off, but managed to continue crawling after its target anyway.

      --
      const int one = 65536; (Silvermoon, Texture.cs)
      SJW, n: "Someone I don't like, and by the way I'm a fuckwit" - AC
    3. Re: Self-Calibrated by GrahamJ · · Score: 1

      Google did something similar actually - provided a physics model and a humanoid form and let their ML figure out how to walk over various virtual obstacle courses, which it did quite well. (look it up, I love the way it waves one arm as a balance strategy!)

      Presumably that would be the next step for this system - after figuring out its own form and limitations, have it figure out how to use that to achieve goals such as locomotion etc.

    4. Re: Self-Calibrated by GrahamJ · · Score: 1

      Here it is: https://youtu.be/gn4nRCC9TwQ

  8. Whats the robots name? by pgmrdlm · · Score: 1

    Sonny(iRobot)? The Robot(Lost in Space)? hal 9000(Space Odyssey )? Maximilian(The Black Hole)? âR2-D2/C3PO (Star Wars)?
    Damn, I haven't even come close to touching all of the different robots that have previously been named.

    --
    Anonymous comments are as pathetic as the anonymous "sources" that contaminate gutless journalism from the New York Time
    1. Re:Whats the robots name? by TeknoHog · · Score: 2

      hal 9000(Space Odyssey )?

      It came up with an abstraction of its own hardware. Ergo, Hardware Abstraction Layer.

      Damn, I haven't even come close to touching all of the different robots that have previously been named.

      Have you perchance touched Dominique, Auburn, Gabriella, Lana, or Irina?

      --
      Escher was the first MC and Giger invented the HR department.
    2. Re:Whats the robots name? by craigwilkie · · Score: 1

      I thought the name of HAL in 2001 was derived from "Heuristic Algorithmic Learning"?

    3. Re:Whats the robots name? by TeknoHog · · Score: 1

      I thought the name of HAL in 2001 was derived from "Heuristic Algorithmic Learning"?

      Yep, and possibly also because it's an alphabetical shift of "IBM". But the last time I've seen the abbreviation in any real computing context was some kind of hardware abstraction layer thingy on Linux.

      --
      Escher was the first MC and Giger invented the HR department.
  9. Re:But what does it identify as? by Shotgun · · Score: 1

    What happens if I refuse to bake a cake for it?

    --
    Aah, change is good. -- Rafiki
    Yeah, but it ain't easy. -- Simba
  10. I'm not sure you've fully thought this through... by Java+Pimp · · Score: 1

    "But if we want robots to become independent"

    I think this was your first mistake.

    --
    Ascalante: Your bride is over 3,000 years old.
    Kull: She told me she was 19!
  11. It's about time by PPH · · Score: 1

    Skynet was supposed to become self-aware August 29th, 1997.

    --
    Have gnu, will travel.
    1. Re:It's about time by 93+Escort+Wagon · · Score: 1

      Skynet was supposed to become self-aware August 29th, 1997.

      It did. But in this reboot it's just quietly biding its time...

      --
      #DeleteChrome
  12. 2009, University of Ottawa by spiritplumber · · Score: 1

    https://www.youtube.com/watch?...

    --
    Liberty - Security - Laziness - Pick any two.
  13. Conciousness by markjhood2003 · · Score: 3, Insightful

    The article describes a robot that can model itself physically.

    The more interesting exploration would involve the robot modelling its own internal state. At that point a closed feedback loop could be initiated with the model informing the system about itself which in turn informs and becomes part of the model.

    If the model becomes good enough, the system might eventually develop the illusion that its embedded model is actually itself. At least that seems to be what happened with the majority of humans.

  14. Re:So Many Questions, F*ck 'Ad-written' News by Tablizer · · Score: 1

    The video doesn't show any extra sensors, so I'm assuming the arm only has motors and encoders to determine how far the motors turned. So how did the robot know...?

    Robot: "Alexa, what am I?"

    --
    Table-ized A.I.
  15. Re:So Many Questions, Fuck 'Ad-written' News by wlorenz65 · · Score: 1

    The paper is on sci-hub.tw. The DOI is 10.1126/scirobotics.aau9354. The source code is at https://github.com/rjk2147/Tas....

  16. It's blind by wlorenz65 · · Score: 1

    Quote from the paper: Actions correspond to four motor angle commands and sensations correspond to the absolute coordinate of the end effector.

    This means that it cannot see, and either needs additional 3D motion tracking hardware or human handcrafted logic to detect the position of its end effector. Everything it does is just learn an inverse kinematics model, so that you can command it to move the end effector to a certain position afterwards. But it cannot learn, for example, to detect position and orientation of a target object or to avoid obstacles in its way.

  17. Re: But what does it identify as? by GrahamJ · · Score: 1

    Once we decide that an entity is conscious then yes, we should bestow such considerations, as any decent person does with fellow humans.