Humanoid AI using ML-Agents in Unity - Preview

Поделиться
HTML-код
  • Опубликовано: 18 дек 2024

Комментарии •

  • @NPCUniverse
    @NPCUniverse 3 года назад +3

    Thank you for showing this experiment it was interesting, looking forward to future updates.

  • @TheCunningStunt
    @TheCunningStunt 3 года назад +3

    Look forward to seeing how this works out.

    • @ClockworksGames
      @ClockworksGames  3 года назад

      Thanks for the comment! More videos coming soon.

  • @jcookev
    @jcookev 3 года назад +3

    So glad I discovered this channel, you make such good quality videos and explain the topic so well. Looking forward to the next episode of 'Understanding Unity ML-Agents'. Hope you are doing well, all the best!

    • @ClockworksGames
      @ClockworksGames  3 года назад +1

      Thanks very much for this comment! I really appreciate the feedback. I am working on the next video now...

  • @parikshitsaikia5355
    @parikshitsaikia5355 Год назад +1

    What will be your approach to solve the issue of jerking movement of the humanoid

    • @ClockworksGames
      @ClockworksGames  Год назад

      Please see the videos I made after this one, especially: ruclips.net/video/nIozvpOsb0Q/видео.html.

  • @erobusblack4856
    @erobusblack4856 3 года назад +1

    Multi modal ai additions to the neural network, id say link a perceiver, GPT-3 and apply unitys computer vision, like if there were basic npcs around that environment with scripts or GPT-3, for the ml one to learn helpful info from in catching the dog, and putting different ways to catch the dog

  • @erobusblack4856
    @erobusblack4856 3 года назад +1

    🤩 cool

  • @erobusblack4856
    @erobusblack4856 3 года назад +2

    Training something like this will take time and patience btw, at this point all u need is a language modal that reacts with the npc, and you have created a baby virtual human 😁
    The clockwork mage strikes again

  • @erobusblack4856
    @erobusblack4856 3 года назад

    Idk, its alot as u know about ai, but i loved the progress, any item interactions?

    • @ClockworksGames
      @ClockworksGames  3 года назад +1

      No item interactions yet, other than "finding" (actually touching) the dog.

    • @erobusblack4856
      @erobusblack4856 3 года назад +1

      @@ClockworksGames thats easy fix, small proximity trigger

  • @VishnuVardhan-vy1ve
    @VishnuVardhan-vy1ve 2 года назад

    Sir I had a doubt .. that can I know the name of algrothim used in this video

    • @ClockworksGames
      @ClockworksGames  2 года назад

      The video uses Deep Reinforcement Learning implemented by the ML-Agents toolkit. Does that answer your question?

    • @VishnuVardhan-vy1ve
      @VishnuVardhan-vy1ve 2 года назад

      @@ClockworksGames sir I know that you used deep reinforcement learning only but I am asking what type of deep reinforcement learning algorithm like DQN,PPO ,A2C,A3C...

    • @ClockworksGames
      @ClockworksGames  2 года назад

      @@VishnuVardhan-vy1ve Ok, thanks for the clarification on what you are asking. The ML-Agents toolkit uses PPO by default and also supports SAC. I was using PPO. See the docs for more information: github.com/Unity-Technologies/ml-agents/blob/release_17/docs/ML-Agents-Overview.md

    • @VishnuVardhan-vy1ve
      @VishnuVardhan-vy1ve 2 года назад +1

      @@ClockworksGames thanks sir

  • @erobusblack4856
    @erobusblack4856 3 года назад +1

    Also sorry i blast comments i get excited about this stuff, lol

  • @erobusblack4856
    @erobusblack4856 3 года назад

    Um, ya, but i got an idea 💡
    Do a needs based state machine 🤔and have it designed for the ml agent to use as little input as necessary to function in an everyday living environment, everyday events, items to interact with, it would be alot, but a base model would be worth $ if u got that working, basically a sim with a GPT-3 to talk with, a perceiver ai modal linking GPT-3 and ml agents,

    • @ClockworksGames
      @ClockworksGames  3 года назад

      Interesting and ambitious ideas...

    • @erobusblack4856
      @erobusblack4856 3 года назад

      @@ClockworksGames lol, my study of humanoid ai might be useful in concept design 😄

  • @erobusblack4856
    @erobusblack4856 3 года назад

    Make the most humanoid npc possible before applying ml agents, but let the hard coded stuff be choosing which the ai would do

    • @ClockworksGames
      @ClockworksGames  3 года назад

      It does seem that ML-Agents is best for one specific goal, if that is what you mean, and that some logic at a higher level is needed if there are multiple goals.

    • @erobusblack4856
      @erobusblack4856 3 года назад +1

      @@ClockworksGames ya basically, but it can handle multiple, but less for the "mind" to still get a clear path to the optimal solution for the task
      But lets say 1
      Interact with objects would be the action that would then inform effect the hard coded state machine stuff i told u about