Show HN: Teaching a Dinosaur to Jump: Rust, WebAssembly, and Neural Evolution

a year ago

The project started as a simple clone of the Chrome Dino game using Rust and WebAssembly but evolved into a complex simulation involving neuroevolution, physics, and AI.
Rust was used for its performance and memory safety, while WebAssembly enabled high-speed execution in the browser.
The dinosaur's behavior was controlled by a neural network with inputs like obstacle distance and speed, and an output for jumping.
An evolutionary algorithm was implemented to improve the neural network's performance over generations through mutation and selection.
Visual debugging tools were added to monitor the neural network's decision-making, including fitness graphs and weight heatmaps.
The neural network architecture was expanded to include a hidden layer, improving adaptability and decision-making.
The simulation was scaled to run thousands of agents in parallel, with optimizations to maintain performance.
The project was deployed as a static site on GitHub Pages, making it accessible for education and experimentation.
Key lessons included the importance of real-time visualization, the power of evolutionary strategies, and the efficiency of Rust + WebAssembly.
Future work could explore more complex actions, alternative activation functions, and user tools for inspecting neural networks.

Hasty Briefsbeta