EvoFly — Simulating Evolution with Neural Networks & Genetic Algorithms

Birds that teach themselves to eat food — through evolution, in your browser.

EvoFly is a Rust + WebAssembly simulation where a population of birds evolves over generations to navigate toward food. Each bird has a neural network brain wired to its field-of-view eye. The genetic algorithm selects the best-performing birds each generation, breeds them, and mutates their offspring — gradually producing smarter birds with no hand-coded rules.

Built by following the Learning to Fly series, and extended with a custom browser frontend (EvoFly).

---

Demo

Open http://localhost:8080 after following the setup guide.

┌──────────────────────────────────────────────────────┐
│               SIMULATION LOOP (per step)             │
│                                                      │
│  👁 Eye         sees nearby food → 9 numbers         │
│  🧠 Brain       decides speed & turn → 2 numbers     │
│  🐦 Animal      moves, eats food, ages               │
│                                                      │
│  Every 2500 steps → Evolution:                       │
│    1. Fitness = foods eaten per bird                 │
│    2. Roulette wheel selection of parents            │
│    3. Uniform crossover + Gaussian mutation          │
│    4. New generation of birds                        │
└──────────────────────────────────────────────────────┘

---

Workspace Structure

rust-genetic-algo/
├── Cargo.toml                        ← workspace root
├── README.md                         ← this file
│
├── genetic-algorithm/                ← lib-genetic-algorithm
│   └── src/lib.rs
│
└── libs/
    ├── neural-network/               ← lib-neural-network
    │   └── src/lib.rs
    ├── simulation/                   ← lib-simulation (core logic)
    │   └── src/
    │       ├── lib.rs                (Simulation, Statistics)
    │       ├── world.rs              (World)
    │       ├── animal.rs             (Animal)
    │       ├── eye.rs                (Eye, process_vision)
    │       ├── brain.rs              (Brain)
    │       ├── food.rs               (Food)
    │       └── animal_individual.rs  (GA bridge)
    ├── simulation-wasm/              ← lib-simulation-wasm (WASM bridge)
    │   └── src/lib.rs
    └── web/                          ← browser frontend (EvoFly)
        ├── index.html
        ├── index.js
        ├── package.json
        └── pkg/                      ← compiled WASM (generated)

---

How It Works

The Eye

Each bird has an eye with a 225° field of view divided into 9 cells. For each cell, the eye returns a float in [0.0, 1.0] — 0.0 means no food in that direction, 1.0 means food is right there.

The Brain

A 3-layer neural network: 9 inputs → 18 hidden → 2 outputs.

• Output 0 → speed change
• Output 1 → rotation change

The Genetic Algorithm

Every 2500 steps, the GA runs:

1. Selection — Roulette wheel: birds that ate more food are more likely to be picked as parents
2. Crossover — Each gene of the child is randomly taken from parent A or B
3. Mutation — Each gene has a 1% chance of being nudged by a random Gaussian value

The WASM Bridge

lib-simulation-wasm compiles to WebAssembly and exposes the simulation to JavaScript. The frontend reads position, rotation, and per-cell vision data every frame to render birds and their field of view.

---

Key Technologies

Technology	Use
Rust	All simulation logic, neural network, genetic algorithm
nalgebra	2D vector math (positions, rotations)
rand	Seeded RNG, weighted selection
wasm-bindgen	Rust ↔ JavaScript bridge
wasm-pack	Builds and packages the WASM module
Canvas 2D API	Renders simulation in the browser

---

Crates

Crate	Description
lib-neural-network	Feed-forward neural net with ReLU activation
lib-genetic-algorithm	Generic GA: selection, crossover, mutation
lib-simulation	World, animals, eye, brain, simulation loop
lib-simulation-wasm	WASM bindings for the browser

---

Quick Start

# Run all tests
cargo test --workspace

# Build WASM
wasm-pack build libs/simulation-wasm/ --target web

# Copy WASM into web/
cp -r libs/simulation-wasm/pkg web/pkg

# Serve frontend
cd web && npx serve . --cors -p 8080

Then open http://localhost:8080.

See /RUNNING.md for the full guide.

---

Terminal Commands (in the browser)

Command	Effect
t / train N	Skip N generations
animals N	Respawn with N birds
food N	Respawn with N food items
p / pause	Toggle pause / resume
r / reset	Restart with current settings
help	Show all commands

---

Tests

cargo test --workspace

lib-genetic-algorithm   9 tests  ✅
lib-neural-network      2 tests  ✅
lib-simulation          3 tests  ✅
Total: 14 / 14 pass

---

References

• Learning to Fly — Part 1
• Learning to Fly — Part 2
• Learning to Fly — Part 3
• Learning to Fly — Part 4
• wasm-bindgen Book
• wasm-pack