What it does

This is an HTML5 Flappy Bird clone where ten birds play simultaneously, each controlled by its own tiny neural network. When every bird has crashed, a genetic algorithm breeds the top performers into a new generation. Rinse and repeat until something survives longer than your patience.

The interesting bit

The neural network is almost aggressively simple: two input neurons (horizontal distance to gap, height difference to gap), six hidden neurons, one output neuron that triggers a flap above 0.5. The README walks through the full genetic pipeline—selection, crossover, mutation, fitness scoring—making it a rare codebase where the “boring” evolutionary plumbing is the actual educational payload.

Key highlights

• Built with Phaser and the Synaptic neural network library; no backend required
• Fitness function is explicit: total travelled distance - distance to closest gap
• Genetic operators are fully documented: top-4 winners survive, 6 offspring fill the rest via crossover and mutation
• Includes linked tutorial and video walkthrough for learning neuroevolution from scratch
• ~1,800 stars suggests it has served as a gateway drug for many developers

Caveats

• The Synaptic library and Phaser version used are from an earlier era; expect vintage JavaScript patterns
• Hardcoded population size (10 birds) and network architecture; not designed for experimentation without code changes

Verdict

Grab this if you want to see neuroevolution happen in a browser tab and finally understand why your GA homework mattered. Skip it if you need a modern, extensible reinforcement learning framework—this is a tutorial fossil, not a research tool.