Skip to content

Neural agents learn to play in an agario-like environment with Neataptic

License

Notifications You must be signed in to change notification settings

guzuligo/agario-ai

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Agar.io-AI

This repository shows how you can use Neataptic to succesfully teach neural networks to play Agar.io. You can see the genomes live in action here. These genomes have been trained for over 1000 generations, and definetely show sume human-like traits. Visualisation done with P5.js.

Read an article on this repo here.

Settings

If you manage to optimize the settings, please perform either a pull request or create an issue here.

Settings (contained in js/main.js):

  • WIDTH set the width of the playing field
  • HEIGHT set the height of the playing field
  • MAX_AREA set the maximal area a genome can acquire
  • MIN_AREA set the minimal area of a genome
  • RELATIVE_SIZE set how much migger a genome should be to eat a blob
  • DECREASE_SIZE set how much area is kept each round
  • DETECTION_RADIUS set how far a genome can see (pixels)
  • FOOD_DETECTION set the maximal amount of food blobs a genome can detect
  • PLAYER_DETECTION set the maximal amount of other genomes a genome can detect
  • MIN_SPEED set the minimal multiplier speed a genome can have (bigger genomes move slower)
  • MAX_SPEED set the maximal multiplier speed a genome can have (smaller genomes move faster)
  • FOOD_AREA set the area of food blobs
  • FOOD_AMOUNT set the amount of food blobs on the field
  • PLAYER_AMOUNT set the amount of genomes that play on the field (population size)
  • ITERATIONS set the amount of iterations/frames each generation is tested for
  • START_HIDDEN_SIZE set the amount of hidden nodes each genome starts witch
  • MUTATION_RATE set the mutation rate
  • ELITISM_PERCENT set the percentage of elitism

Most important setting:

  • USE_TRAINED_POP setting this to false will start the evolution from scratch (USE THIS WHEN OPTIMIZING THE SETTINGS), setting this to true will use the pre-trained population

Default setting values

var WIDTH            = $('#field').width();
var HEIGHT           = 800;

var MAX_AREA         = 10000;
var MIN_AREA         = 400;

var RELATIVE_SIZE    = 1.1;
var DECREASE_SIZE    = 0.998;

var DETECTION_RADIUS = 150;
var FOOD_DETECTION   = 3;
var PLAYER_DETECTION = 3;

var MIN_SPEED        = 0.6;
var SPEED            = 3;

var FOOD_AREA        = 80;
var FOOD_AMOUNT      = Math.round(WIDTH * HEIGHT * 4e-4);

// GA settings
var PLAYER_AMOUNT     = Math.round(WIDTH * HEIGHT * 8e-5);
var ITERATIONS        = 1000;
var START_HIDDEN_SIZE = 0;
var MUTATION_RATE     = 0.3;
var ELITISM_PERCENT   = 0.1;

// Trained population
var USE_TRAINED_POP = true;

About

Neural agents learn to play in an agario-like environment with Neataptic

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • JavaScript 99.9%
  • HTML 0.1%