Superhuman AI - Deep Reinforcement Learning

Discussion in 'Knowledge Base' started by n01sname, Sep 19, 2020.

  1. n01sname

    n01sname Well-Known Member

    Joined:
    Feb 5, 2015
    Ratings:
    +956 / 0 / -0
    • Like Like x 1
  2. Maskerader

    Maskerader Well-Known Member

    Joined:
    Oct 6, 2019
    Ratings:
    +352 / 0 / -0
    I'm skeptical on that. Grid Autosport also used self-learning AI, for example. Result? The AI drives as a hotlapper and has massive problems with racing in a pack.

    For my tastes I expect a good AI to create an illusion of a good human driver:
    1) it has a bit if inconsistensy, imperfection in lines, braking, accelerating;
    2) it takes different corners with speeds similar to humans (i.e. doesn't take fast corners consistently faster than me and doesn't fall behind in slow turns);
    3) it can do clean wheel-to-wheel racing.

    Which of that can be achieved with self-learning AI doing hotlaps?
     
    • Agree Agree x 1