AMP: Adversarial Motion Priors for Stylized Physics-Based Character Control (Paper Explained)

Поделиться
HTML-код
  • Опубликовано: 4 дек 2024

Комментарии • 18

  • @YannicKilcher
    @YannicKilcher  3 года назад +3

    OUTLINE:
    0:00 - Intro & Overview
    1:25 - Problem Statement
    6:10 - Reward Signals
    8:15 - Motion Prior from GAN
    14:10 - Algorithm Overview
    20:15 - Reward Engineering & Experimental Results
    30:40 - Conclusion & Comments

  • @CristianGarcia
    @CristianGarcia 3 года назад +10

    What a time to be alive!

  • @johnpope1473
    @johnpope1473 3 года назад +1

    Appreciate the pace and clarity. Thanks

  • @AkashMishra23
    @AkashMishra23 3 года назад +8

    Love the Intro

  • @Konstantin-qk6hv
    @Konstantin-qk6hv 3 года назад +1

    Amazing!

  • @dinoscheidt
    @dinoscheidt 3 года назад +4

    RUclips auto played the video... really loud someone in our apartment asked for money 😱

  • @G12GilbertProduction
    @G12GilbertProduction 3 года назад

    And this motion priorities evidently works without a VGA picture acceleration in the 1x56 variant multidimensional map?

  • @LouisChiaki
    @LouisChiaki 3 года назад

    The audio seems a bit not clear this time.

  • @CreativeBuilds
    @CreativeBuilds 3 года назад +4

    I'm sorry Yannic, I'll have your money by tomorrow. Please don't hurt my family 🙏

  • @improvementbeyond2974
    @improvementbeyond2974 3 года назад +2

    Is this the robot motion revolution we expect in the next few years?

    • @NeoKailthas
      @NeoKailthas 3 года назад +2

      Instead of a wall they should've put a human...

    • @chrisfreel
      @chrisfreel 3 года назад

      @@NeoKailthas Boston Dynamics has got you covered.

  • @herp_derpingson
    @herp_derpingson 3 года назад +1

    This is surprisingly obvious. I wonder why nobody tried this before.

    • @dawidlaszuk
      @dawidlaszuk 3 года назад +3

      Because coming up with an idea is super fast but actually doing the work takes time. In coming year from now we'll see plenty of things that are "obvious" now.

    • @YannicKilcher
      @YannicKilcher  3 года назад +2

      I also guess many hacks are necessary to actually get it to work

  • @jrohit1110
    @jrohit1110 3 года назад

    Got deep mimic working: ruclips.net/video/yluXxxZWv8c/видео.html&ab_channel=RohitKumarJ
    ... next is amp :D

  • @JTMoustache
    @JTMoustache 3 года назад +2

    reward shapping is for boomers