Neat AI does Asteroids using NEAT and a Genetic Algorithm
2021 ж. 4 Ақп.
6 469 Рет қаралды
In this video I use a NEAT neural net and a genetic algorithm to defeat the Asteroid game.
It uses elements from machine learning to evolve the machine gunner bot, which sits in the middle and blasts all around it.
Music:
www.bensound.com
Cool video, I used to experiment with the NEAT algorithm some 8 years ago, glad to see someone actually made it work :D btw I recommend lowering the music a bit since it is a bit difficult to understand you, another thing to try would be to cut out a bit of the low frequency of the music with an equalizer, since you have a deeper voice, the music bass is occupying the same frequency
Thanks for the info! I struggle with the audio on all my vids..
It would be interesting to see multiple pilots at once with a reward for not shooting each other.
Who needs movement when you can just pew pew?
Great animations! Would appreciate code walkthrough video
Thanks for the feedback.. No plans to release the code at this stage but am planning to port everything to unity and c# so will eventually release it all.
To challenge the AI, how about the "one bullet" implementation where you cannot shoot again until the bullet hit something? And add some information about the direction of the closest asteroids? That would probably give some more interesting networks.
Yup,, the machine gunner solution then never emerges and its forced to get moving...
Real players tried to minimize the number of small, fast asteroids in play at once. I wonder if this kind of behavior emerged.
No, I never told it about the type of asteroid, just distance and direction.. But thats a good point.. I'll have to give it a go
@@neatai6702 will be interesting to see how it turns out.
Is this made in PyGame/Tkinter?
Hey, nice video! Would you mind sharing your source code for this?
Thanks for the comment.. I'll be releasing it when I move it to C#. Will probably do a vid on how to code it up if there's any interest
@@neatai6702 Any follow up on that? I'm super interested. Beautiful work nonetheless, and thank you for sharing this.