Robots that can adapt like animals (Nature cover article)



The Intelligent Trial and Error Algorithm introduced in the paper ‘Robots that can adapt like animals’ (Nature, 2015): the video shows two different robots that can adapt to a wide variety of injuries in under two minutes.

A six-legged robot adapts to keep walking even if two of its legs are broken, and a robotic arm learns how to correctly place an object even with several broken motors.

Full citation: Cully A, Clune J, Tarapore DT, Mouret J-B. Robots that can adapt like animals. Nature, 2015. 521.7553, (cover article).

Source code: https://github.com/resibots/cully_2015_nature

Paper: http://www.nature.com/nature/journal/v521/n7553/abs/nature14422.html

Arxiv (free): http://arxiv.org/abs/1407.3501

source

Fahad Hameed

Fahad Hashmi is one of the known Software Engineer and blogger likes to blog about design resources. He is passionate about collecting the awe-inspiring design tools, to help designers.He blogs only for Designers & Photographers.

43 thoughts on “Robots that can adapt like animals (Nature cover article)

  • September 15, 2017 at 11:09 pm
    Permalink

    This is basically an evolutionary algorithm applied directly to the sequence of limb movements, with the fitness measurement being the ability to go in a given direction? I think I'll have to look up this bayesian optimization stuff though.

    Anyway, I think this would be a great addition to 3D printing arms; adaptation and self calibration all in one. Although, something in the overall production process might interfere initially, but that just means that things have to be re-jiggered to make it all work together.

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    Nice job, for the algorithm. Do you plan on making it open source? That would help me and I bet many of us for any kind of project. I could even donate

    I would appreciate if I would get an answer 🙂

    Thank you

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    wait. if the damaged robot is able to move faster than the undamaged one, doesn't that imply that the algorithm failed to find best possible walking pattern to begin with?

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    This is very good research. congratulations on this sucess. this is exactly what i want science to be doing.

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    生物が自分の体の動きを自己組織化する機能をシミュレーションできるロボットの登場は、搭乗型移動支援ロボットの未来を大きく変えると思います。

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    As a cognitive neuroscientist, I think really great research! One step closer to singularity!

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    can someone give me a push in the right direction – what should i look for in google to get acquainted with those algorithms?

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    This closely resembles the Q-Learning reinforcement learning algorithm.  I'd love to see someone build a service where you can specify the attributes of a machine (number of motors, dimensions, method of movement, etc) and then be able to purchase a pre-initialized Q-function map, giving you a "child" robot that can learn from an informed state.

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    If a robot makes a wrong call in a triage situation (e.g deciding which earthquake survivors to save first), whose fault would it be?

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    Beautiful!

    Shame "innovative" is so overused and just sounds like boring marketing jargon. THIS demonstrates real innovation! Can't wait to see how this helps the progress of robotics and AGI research.

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    I will load another, better option.
    To raise left middle leg and walk on four legs.
    No jumping around.

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    "trial and error" in cave of nuclear station ? )))

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    This woud b a perfect presentation: nice pleasant voice with perfect delivery n good camera work, if it wasn't muffled by the useless, distracting n 2o loud music track. whomever tells u it's necessary is lying. the fact that it's ubiquitous doesn't make it right.

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    Anyone else, for some reason, feel bad for the robots?

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    Do NOT worry. Our new robot overlords will find us dull and ignore us. BUT – 
    AMAZING! Machine learning and adaptation… a robot's algorithm adapts to post-damage behavior, testing performance over time, and learning to "limp" in a dynamic and almost animal like way…. limping away while compensating for the broken leg rotor.

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    if done right, robots can lead us into a utopia where all are taken care of and no one has to do menial labor to survive. not to say this is happening tomorrow. here at Galactic Public Archives we are optimistic about the future 🙂

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    I'm sorry.. I really love the scientific things we manage to do, but this and artificial intelligence would end up with what? Something like Terminator? Worse?

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    now they just have to figure out a way for the robot to run all those simulations internally and directly operate the most efficient process.

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    Or have a variety of onboard sensors to check for failed movements? A set of flex sensors and rotary encoders can identify limb failure, for which an alternative gait can instantly be implemented.

    But still, the artificial intelligence is still amazing for such a robot.

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    You guys are gonna make us all die, and I'm talking about human kind. A.I. and things like this could means the end of humanity. Once they take control of everything there's not way back. Think twice if you're thinking about improving this.

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    This is great, but it wouldn't really help in a real-time scenario.

    It would be much more suitable if the robot knew how to walk, after something was broken. As in, if it had pre-defined actions to take if Leg 1 was inactive. This would be a big advancement. However, I don't see much light in this.

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    Jean-Baptiste, Antoine, Danesh and Jeff – I really liked the Video. I think you did a great job showing how the algorithm builds on it's "experience" by "experimenting" with different gaits to see which combination of movements yields the  highest velocity in the desired direction. We sponsor two local F.I.R.S.T. Robotics Chapters in East Phoenix (Chandler HS and Hamilton HS). I am passing on your article on to the teams. Congratulations on your publication in JON, your families must be very proud!

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    Looks good but I was expecting a robot to go from A to B improving it's self as it goes. The reset to start to try the next walking style didn't show it well enough.

    Reply
  • September 15, 2017 at 11:09 pm
    Permalink

    Fantastic work.
    Through wouldn't it be better for the robot to completely recreate what it thinks it looks like using one sensor, and then running simulations on the result? Like in the "A Robot Teaches itself how to walk" by ForaTv

    Reply

Leave a Reply

Your email address will not be published. Required fields are marked *