IAMDinosaur

Why we included this project

This project is a compact, self-contained demo of a neural network learning a task purely through trial and error. It feeds a small feed-forward network data read from the pixels of Chrome's offline dino game, then lets a genetic algorithm evolve better jump decisions over successive generations. If you want to understand how evolutionary strategies and neural networks fit together, this is one of the clearest miniature examples around: a handful of Node.js files that read the screen, evaluate a dozen candidate networks, and mutate the best performers. Students and hobbyists exploring reinforcement-style learning will probably get more from tracing this code than from an abstract tutorial, and the video walkthrough explains the mechanism step by step. It is not software you would deploy in production, but as an educational artifact for basic neuroevolution it is unusually approachable.