SeerV7.1 vs. Smashy + Alemz

Поділитися
Вставка
  • Опубліковано 4 жов 2024
  • ► Playlist: • Seer: Reinforcement Le...
    .................................................................................
    Seer: Reinforcement Learning in Rocket League
    Socials:
    Twitch: / ultrawidegc
    Twitter: / ultrawidegc
    Instagram: / ultrawidegc
    Reddit: / ultrawidegc
    Tiktok: / ultrawidegc
    I created a discord for you guys to talk with me! / discord
    .................................................................................
    #UltrawideGC

КОМЕНТАРІ • 33

  • @shaun5657
    @shaun5657 4 місяці тому +12

    When is toxic seer coming out?

  • @Fauxmage
    @Fauxmage 4 місяці тому +3

    Games with Zen/Alpha/Rado vs Seer would have so much value. I feel like we are getting closer in finding the main lines of efficiency in RL. A bit like chess has "book" moves, theres a lot to learn from bot "decision making". Like why give away possession and shoot to the net when your positionning/speed allows you to steal both corner boosts. The boost delta is going to translate to increased pressure and will likely result in a better shot later on when the opponent is out of position . It's not a blunder to "sit in net" but rather an inefficiency if you give up too much ressources. I also think Smashy and Alemz played really well, they quickly identified that their speed/mechanics allowed them to have an edge as long as they kept great pressure or healty boost qty. Notably the Squishy saves seemed effective but also hard given the precision of the shots from the bot. The fact that the bot doesn't have "vision" to deal with gives it a tremendous avantage. Whats also crazy to me is that for the untrained eye, Seer looks not that great at the game. Althought thats going to change when/if it learns to be rewarded from dashes to regain momentum or learn other mechanics known by the bests. Amazing developper work Ultrawide and we also witnessed some amazing RL skills from the players. gg wp

    • @shaun5657
      @shaun5657 4 місяці тому

      I don’t think chess is completely solved as of yet, but we are not too far from that either. Same applies to RL, if you mean fundamentals of the game like shooting, dribbling and passing etc, i think that seer has already mastered it, but since its a bot, there will be ways to exploit it

    • @snekback.
      @snekback. 4 місяці тому

      @@shaun5657we are FAR from solving chess

    • @edgeribble
      @edgeribble 4 місяці тому

      Seer can only play 2s so Vitality can't play it as a trio but maybe Zen and Alpha could play Seer

  • @joakimheitmann1525
    @joakimheitmann1525 4 місяці тому +1

    Petition to make the bot be toxic after every goal

  • @Sodiumax
    @Sodiumax 4 місяці тому +1

    Could you try to make them do double taps? Maybe give a reward if a hit off the backboard shortly results in a goal?

  • @Aceofswade
    @Aceofswade 4 місяці тому +1

    Just how much "hard code" does Seer have? Other than the coded-in kickoffs and maybe the passing flip-resets, what else is put in by you compared to his own learning?

    • @IncineroarBestPokemon
      @IncineroarBestPokemon 4 місяці тому +1

      Not sure, but my guess would be not a lot of hard coding. Most likely neural network (reinforcement learning)

    • @snekback.
      @snekback. 4 місяці тому +1

      Only the kickoffs are hard coded

  • @Szeiyu
    @Szeiyu 4 місяці тому

    I think someone in the chat was talking about using the replays of the games seer played to help it learn. I have no idea how ai learning works, which is why im curious why that these games arent a situation in which seer can learn. Even if it could learn during these games it probably wouldn't matter that much because of how little gameplay time it is compared to all the other training its done.

  • @Idkcantthunkofone
    @Idkcantthunkofone 4 місяці тому +3

    Does seer work like a hive mind or do they think separately?

    • @sebseb6799
      @sebseb6799 4 місяці тому +1

      They think separately. Since this version of Seer has been only trained on 2s, it is pretty good playing with a teammate. It still sometimes bumps its teammate and double commits, though. So it's not perfect, yet.

    • @Idkcantthunkofone
      @Idkcantthunkofone 4 місяці тому

      @@sebseb6799ok thanks

    • @Idkcantthunkofone
      @Idkcantthunkofone 4 місяці тому

      @@sebseb6799I wonder how much better it could get is there a limit or will it eventually just get good enough and nothing will be able to beat it.

    • @maxoemusguenter2523
      @maxoemusguenter2523 4 місяці тому +1

      Propably separate but there fully aware of each other's positioning

    • @UltrawideGC
      @UltrawideGC  4 місяці тому +2

      Seperate

  • @sergemarlon
    @sergemarlon 4 місяці тому

    Would you be able to train a 1v2?

  • @Sodiumax
    @Sodiumax 4 місяці тому

    Also, what exactly is the number like 42400, like, games played?

  • @Ujerhudkjejbeje
    @Ujerhudkjejbeje 4 місяці тому

    I wonder if you could make SEER better at avoid being demoed

  • @flyingrooms
    @flyingrooms 4 місяці тому

    Where is the Bot? I mean is it on a pc or two pc's? How does it get into a game? Is it one bot controlling two cars? The bot can't sit in front of a screen I imagine, so what does it look at?

    • @UltrawideGC
      @UltrawideGC  4 місяці тому +2

      Seer lives in the matrix

  • @MintRL
    @MintRL 4 місяці тому

    On rl bot it’s seer version 0 how do I get the latest seer

    • @edgeribble
      @edgeribble 4 місяці тому

      u can't because Psyonix banned publishing bots because of Nexto in ranked

  • @phanepacc5065
    @phanepacc5065 4 місяці тому

    Could you program them to spam what a save after every goal they score