MarI/O – Machine Learning for Video Games



MarI/O is a program made of neural networks and genetic algorithms that kicks butt at Super Mario World.
Source Code:
“NEAT” Paper:
Some relevant Wikipedia links:

BizHawk Emulator:

SethBling Twitter:
SethBling Twitch:
SethBling Facebook:
SethBling Website:
SethBling Shirts:
Suggest Ideas:

Music at the end is Cipher by Kevin MacLeod

Nguồn: https://alancarlsonmd.com

Xem thêm bài viết khác: https://alancarlsonmd.com/game/

48 thoughts on “MarI/O – Machine Learning for Video Games

  1. i tried to download it, but it says it can't find the DP1.state even though i have it.
    i think i'm doing something wrong. the AI does move and stuff, but it doesn't restart to the saved state.
    Please help

  2. It looks like it doesn't carry state from one frame to the next? Or does it? If it can retain state could it track direction of movement, for example of the monsters? Without it it just knows "there's a monster here"?

  3. Interesting… What if you would start to learn this program using rom-hack that sounds like "Player 2 controls the enemy". Will be program smarter or dumber? Program can learn the level, but can't learn your control inputs… Very interesting thing if think about it…

  4. I was just trying to figure out what was driving the AI to be better but it seems it was just survivorship bias on my side, I didn't really think it was as simple as picking the good scoring ones as the final product

  5. How about those enemies that mario can’t kill just by jumping on them? Are they denoted by the black colour too?

  6. First we enjoy playing videogames, then we enjoy watching other people playing videogames, next we will enjoy watching machines playing videogames.

  7. Create a species – after it has evolved enough to match or even surpass real players, put it in a Cod MW session jusy so it can teabag them

  8. Recuerdo haber visto en su tiempo, cuando solo tenia 1000 views, en ese tiempo no lo entendía y ahora tampoco :'v
    PD:Nó se ingles :'v

  9. that 4:04 numbers is so beautiful. just to think and imagine how many millions of Input and Output (of the function) the original data sets of numbers has to go through for that 'neat' algorithm. Great project man.

Leave a Reply

Your email address will not be published. Required fields are marked *