MasterSkepticista / NEAT-Flappy-Bird

NEAT algorithm JS

NeuroEvolution of Augmenting Topologies

NEAT algorithm in p5.js

Run the Flappy bird demo in your browser! Flappy GA 🐦

Use the slider to speed up the learning. Birds will have a time-lapse.

Bonus point: If the simulation doesn't freeze when slider is pulled high, you have a good PC.

If not, buy a new one. 😐

Code

Sketch begins in game.js, contrary to nomenclature.

This is a Neuroevolution algorithm, NEAT (of Augmented Topologies), where optimizations are based on spawns of a single unit, each carrying its own Neural Network. In this bird game, a random population of birds is spawned, each with random 'brains'.

  for(let i = 0; i<TOTAL;i++){
      birds[i] = new Bird();
  }

The Bird class houses the characteristics of each bird, namely, location, speed, acceleration and brain.

Add and splice pipes in UI:

  Pipes.push(new pipe());

Derive fitness of each bird:

    for(let bird of savedBirds) {
      sum+=bird.score;
    }

    for(let bird of savedBirds) {
      bird.fitness = bird.score / sum;
    }

When all birds die, spawn new 250 of all previous generation winners

    calculateFitness();
      for(let i = 0; i < TOTAL; i++) {
        birds[i] = pickOne(savedBirds);
      }
      savedBirds = [];
      console.log('Built new Generation');

This process goes on and on. All mutation/copy logistics is handled by the NN class.