Tetris AI: from 0 to 100.000 in 100 generations 🤖
Ойын-сауық
This video shows the time-lapse of an AI learning to play Tetris.
From 0 to 100.000 pieces in 100 generations.
This video shows the time-lapse of an AI learning to play Tetris.
From 0 to 100.000 pieces in 100 generations.
Пікірлер: 19
AWESOME! And oh so great choice of soundtrack! Loving it!
That's so cool ! In the last generation it seems that most of the population died because the RNG didn't yield any straight piece for too long when all of them were ready for a couple 4-line combos
@AlanZucconi
Жыл бұрын
Life's so unfair sometimes 🤣
@nejsonsvejson9861
11 ай бұрын
Like when playing real tetris!
I was soo worried about you! I thought something happened to you, Love your Videos btw 🥰
@AlanZucconi
Жыл бұрын
Aww thanks! 🥰 WORK happened haha! With I had more time to publish more videos!
this is very fascinating, it would be awesome if you could do a video explaining how the AI works :) am I right in observing that the Tetris game is normalized, i.e. all instances will get the exact same pieces in the exact same order? and if so, what would then be the performance of a trained model on a randomized Tetris game?
@AlanZucconi
11 ай бұрын
You're right! All instances get the same sequence of tetrominoes, so it is easier to compare the performance. The AI is based on Utility Theory. Each AI lists all possible moves, and gives each one a score. Then it picks the move with the highest score. The score is calculated as a linear combination of several heuristics. The coefficients of this linear combination is what the evolution system is optimising for!
does this have SRS, 7-bag generation or other modern tetris features?
@AlanZucconi
11 ай бұрын
Not really! 😅 My version of Tetris doesn't support all of the features modern Tetris has. My focus was on a generic AI, not so-much on the game of Tetris itself!
@moontr3
11 ай бұрын
@@AlanZucconi well this is really an amazing project anyway!
Which software (library) did you use?
@AlanZucconi
Жыл бұрын
I made this in Unity!
@cfilo88
Жыл бұрын
@@AlanZucconi I assumed it. Fantastic work!
@AlanZucconi
Жыл бұрын
@@cfilo88 thanks! I used an evolutionary algorithms to optimise the weights of several heuristics!
@cfilo88
Жыл бұрын
@@AlanZucconi please tell me more 🙏
@AlanZucconi
Жыл бұрын
@@cfilo88 I hope I'll find the time to do a proper video on this!