This Adorable Baby T-Rex AI Learned To Dribble

Articles, Blog

This Adorable Baby T-Rex AI Learned To Dribble

This Adorable Baby T-Rex AI Learned To Dribble

Dear Fellow Scholars, this is Two Minute Papers
with Károly Zsolnai-Fehér. About 350 episodes ago in this series, in
episode number 8, we talked about an amazing paper in which researchers built virtual characters
with a bunch of muscles and joints, and through the power of machine learning, taught them
to actuate them just the right way so that they would learn to walk. Well, some of them
anyway. Later, we’ve seen much more advanced variants where we could even teach them to
lift weights, jump really high, or even observe how their movements would change after they
undergo surgery. This paper is a huge step forward in this
area, and if you look at the title, it says that it proposes multiplicative composition
policies to control these characters. What this means is that these complex actions are
broken down into a sum of elementary movements. Intuitively you can imagine something similar
when you see a child use small, simple lego pieces to build a huge, breathtaking spaceship.
That sounds great, but what does this do for us? Well, the ability to properly combine these
lego pieces is where the learning part of the technique shines, and you can see on the
right that these individual lego pieces are as amusing as useless if they’re not combined
with others. To assemble efficient combinations that are actually useful, the characters are
first required to learn to perform reference motions using combinations of these lego pieces.
Here, on the right, the blue bars show which of these these lego pieces are used and when
in the current movement pattern. Now, we’ve heard enough of these legos,
what is this whole compositional thing good for? Well, a key advantage of using these is that
they are simple enough so that they can be transferred and reused for other types of
movement. As you see here, this footage demonstrates how we can teach a biped, or even a T-Rex
to carry and stack boxes or how to dribble, or, how to score a goal. Amusingly, according
to the paper, it seems that this T-Rex weighs only 55 kilograms or 121 pounds. An adorable
baby T-Rex, if you will. As a result of this transferability property, when we assemble
a new agent or wish to teach an already existing character some new moves, we don’t have
to train them from scratch as they already have access to these lego pieces. I love seeing
all these new papers in the intersection of computer graphics and machine learning. This is a similar topic to what I am working
on as a full-time research scientist at the Technical University of Vienna, and in these
projects, we train plenty of neural networks, which requires a lot of computational resources.
Sometimes when we have to spend time maintaining the machines running these networks, buying
new hardware or troubleshooting software issues, I wish we could use Linode. Linode is the
world’s largest independent cloud hosting and computing provider, and they have GPU
instances that are tailor-made for AI, scientific computing and computer graphics projects. If you feel inspired by these works and you
wish to run your experiments or deploy your already existing works through a simple and
reliable hosting service, make sure to join over 800,000 other happy customers and choose
Linode. To reserve your GPU instance and receive a $20 free credit, visit or
click the link in the description and use the promo code “papers20” during signup.
Give it a try today! Our thanks to Linode for supporting the series and helping us make
better videos for you. Thanks for watching and for your generous
support, and I’ll see you next time!

98 thoughts on This Adorable Baby T-Rex AI Learned To Dribble

  1. Please make a rocket league bot! It's 3d spacing, alot of decision making and the pro scene is still developing in very creative ways

  2. Awesome video! Neuroevolution is an incredibly exciting area of research!
    This is a great resource on evolutionary optimization which came out 2 days ago (September 5th):
    There is another great paper on "recovering from surgery / damage" called MAP-Elites from Jeff Clune and his team at Uber. Basically the algorithm keeps a history of solutions based on diversity to explore once the system has been damaged.

  3. God damnit, I didnt know you lived in Vienna. I literally moved back to America a week ago, I wouldve bought you lunch or something if I knew.

  4. I know I'm an idiot but this is great, hopefully the next decade in computer technology will be a huge leap. I think AMD already is waiting for special orders for CPUs that are made for specialised workflows. Even the next 5 years will be epic but I don't think I will live to see that
    Anyway thanks for the video ^_^

  5. Kroly, this video made me realize how many episodes of yours I've watched. I've been here since episode #1 and I've seen every video. I appreciate your work and I can't wait to see what you'll release in the future! Thank you so much for your work. You're an incredible content creator!!

  6. Your channel deserves more sponsors. Your content is too valuable to not be sponsored.
    Thank you for another great video update on the frontiers of AI.

  7. I was watching the video like "wow this is so interesting, amazing progress and potential" then I got hit by the left T-rex at 2:45 and laughed so hard hauahsushsu

  8. I feel like watching a video about a drooling baby T-Rex now. Is it just me that thinks that would be cute; like when your cat starts to drool?

  9. I think I realise when AI has the potential to be an enemy. I've not heard anyone say this, But that could be semantics.
    Its when any kind of AI can perform a simulation like a human, A bit vague but basically a human knows things that can go together in ways not personally done before. Extreme unsupervised learning. The kind of thoughts that get people killed for postal stamps.

  10. If that T-Rex and Humanoid can share the same dribbling actions without training them from scratch, that's pretty awesome, but I'm guessing it's because they're both biped creatures. Does it also work on creatures that walk on 4 legs?

  11. can anyone point me to evidence supporting the idea that our brains control motor-skills in a similar manner to this model? specifically i mean that part about breaking down of the complex signal(s) into their constituent base signals. is this basically a Fourier transform applied to locomotion? great stuff.

  12. Anyone else find this whole thing deeply terrifying? Like the Trexes are tiny and cute now but this is what the military want to make with it

  13. I always wondered if the Long Jump technique at the olympics was optimum. Could this technique be used to discover the best technique to maximise jump distance?

  14. The researchers clearly state that this is simply a machine learning algorithm but once again you are calling it AI. True AI is not a thing yet. Deal with it.
    Whatever makes you happy though…

  15. hmm makes it a little less exciting for me since it restricts the parameter space to a pre-defined area, which means you will never see any crazy different solutions to your problems.

    but of course, it's great if you want it to be able to do a specific thing.

  16. 郋 郇邽訇迡!!! 苤迡迮郅訄邿迮 邽邾郅邽
    郕訄郕邽邾 郈郋郋訇郋邾 (郈邽 郈郋邾郋邽 邾迮郇郋邿 迣邽)
    迮郅郋赲迮郕 邾郋迠迮 赲郱郅迮迮 郕訄郕 郈邽訄!!!

  17. Ah nice to hear that you are a student at the TU in Austria (; greetings from JKU Linz!
    I am considering studying A.I. instead of Electronics which is what i am stuying right now. Your videos definitely contribute to switching but i am not that sure if i should risk it, it would be the first session at JKU overall so i would be a test taker somehow.. but the fields of application are huge!
    Maybe lets have a skyping conversation before October.

  18. cool, some optimization so the AI can learn how to kill humans faster, like learning to swing a blade or shoot a gun. JK

  19. I said it on the muscle activation papers video and I'll say it again, these researches should get into robotics as soon as possible

  20. Nice work. There are good reasons why humans walks in a specific ways. So is true for t-rex and other animals.
    Hopefully you have incorporated those ideas in your research, like energy saving and balance. Not just mimicking. It's not specified in the video.

  21. I was going to skip this episode (not interested in adorable t-rex dribbling) but holy crap – those lego pieces is something done super well. I wish this concept to continue.

  22. its increidible what IA can do
    cant wait to see an IA generating songs
    oh, wait, they already do that
    cant wait to see an IA generating text
    oh, wait, they already do that
    cant wait to see an IA generating images
    oh, wait, they already do that
    cant wait to see an IA generating videos
    oh, wait, they already do that
    cant wait to see an IA seing images
    oh, wait, they already do that
    cant wait to see an IA playing videogames
    oh, wait, they already do that
    cant wait to see an IA playing chess or go
    oh, wait, they already do that
    cant wait to see an IA that talks to you
    oh, wait, they already do that
    cant wait to see an IA driving a car
    oh, wait, they already do that

    IA is awsome

  23. I took a physically based modelling class in the early 90s in which this stuff was already being done. They were teaching luxo lamps how to get the best distance in a ski jump, among other things.

  24. I was thinking, were you to place two soccer scoring AIs across from one another, competing to put the ball in the opposing goal (with a slight input lag for each AI to simulate reaction time), would they eventaully optimize for more efficient and faster movements? Or would this kind of dynamic and complex addition to the enviroment of either AI make this so computationally expensive that it wouldn't be feasible?

  25. fucking wow!…. It really helps to understand how extinct animals walked and will change the way we animate videogames and movies

Leave a Reply

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