Skip to content
SatarovMOCR
Systems · File dinomind-evolution

DinoMind Evolution

CompleteBuilt by me2024

NEAT neuroevolution learns to play a dino runner — neural networks evolve generation by generation to dodge obstacles.

Open repository

The browser arcade version on this site uses a deterministic demonstration policy inspired by this project — it is not the original trained NEAT model.

Flight data file · dinomind-evolution · 2024

01Problem

Can a population of simple neural networks learn a timing-critical game from scratch, with no gradient descent and no labeled data?

02Architecture

Pygame runner with randomly generated cacti and increasing difficulty; NEAT-Python evolves network topologies and weights against a fitness function based on survival; real-time score and generation stats.

03Lessons learned

  • Fitness shaping matters more than population size — small input tweaks changed evolution speed dramatically.