Karakanlud
Karakanlud
64 / 4
15th Mar 2017
19th Mar 2017
A remade version of Drone Generals, this game features a 1 on 1 combat against an AI-controlled drone. It uses AI cards, which you can paste in the reader to start the bot. Currently, AIs code random sequences and react to nearby shots.
neat electronics game robot smart intelligent brain

Comments

  • v325
    v325
    17th Mar 2017
    Cool, I shoot into hole for mooving and break AI-bot without detonating. +1
  • LimpDongGun
    LimpDongGun
    17th Mar 2017
    this is seriously cool
  • Karakanlud
    Karakanlud
    17th Mar 2017
    Also, I'll post the new version in ~7 h
  • Karakanlud
    Karakanlud
    17th Mar 2017
    I also find much more potential in this kind of AI. It's environment is much more defined, so you can actually program it to do something interesting.
  • Karakanlud
    Karakanlud
    17th Mar 2017
    In the case of bots, this is the furthest I got with using only neurons: id:1901407
  • Karakanlud
    Karakanlud
    17th Mar 2017
    So I tried to add digital parts to them, like the first of the decision gates I use now. id:1976460 These are the ones that actually show some patterns. You can add the huge feedback units from pure neural networks, and in theory, you can say that memorized input and output has some effect on the processing of current input.
  • Karakanlud
    Karakanlud
    17th Mar 2017
    connections, and the connection's strenght would determine the probability of that decision. In TPT you can't really make them do that, so deep learning this way is impossible. You can add some memory by making huge loops with precise temp settings, but it's just not worth it.
  • Karakanlud
    Karakanlud
    17th Mar 2017
    @QuanTech id:1966668 I've tried that, it won't happen for a number of reasons: 1 It's a real pain in the arse to code anything with them. You have to start it obviously with some basic approximative functions, but even the smallest tasks have to be routed with proper excitation/inhibition and delicate TSNS settings. This is also the 2nd problem. It's pretty rigid. It won't really do anything else other that the stuff you code in. In the real brain, experimenting would make the neurons form new
  • QuanTech
    QuanTech
    17th Mar 2017
    @Karakanlud could you try a pure neural network? That'd be incredible. It would take feedback and learn over time. If the neural net is advanced enough, it could get really good at this game. And if the neurons are really small enough, then maybe.... just maybe.... deep learning could happen
  • Karakanlud
    Karakanlud
    17th Mar 2017
    Yeah, that occurs sometimes. I rigged it blow up once the hull is melted through, but sometimes the firing mech. is spared.