SHEET 14 · Simulation
Neural Maze Navigator
Feedforward neural networks evolve via genetic algorithms to solve procedural mazes — 11 presets, death heatmaps, live neuroevolution
no plate captured · maze.pjeide.com
2024liveSimulation
About
A neuroevolution sandbox where populations of feedforward neural networks learn to navigate procedurally generated mazes through genetic algorithm selection, crossover, and mutation. Each agent uses ray-cast sensors to perceive walls, with a custom physics engine handling ball-rolling mechanics. Features 11 difficulty presets, generational fitness graphs, death heatmaps showing where agents fail, and real-time visualization of the fittest network's topology and weights.
Key features
- 01Neuroevolution: feedforward neural networks trained via genetic algorithms
- 02Ray-cast sensor system for agent wall perception
- 0311 difficulty presets from trivial corridors to complex labyrinths
- 04Death heatmaps showing where agents consistently fail
- 05Live network topology visualization with weight rendering
- 06Generational fitness graphs tracking evolution progress
- 07Procedural maze generation with configurable complexity
- 08Multiple test modes (physics, neural, sensor, GA)
- 09Telemetry analysis tooling
Nearby
SHEET 12 · Simulation
Ant Farm Simulator
bitECS-powered colony simulation — 50K ants at 60fps with haplodiploidy genetics, multi-colony warfare, and a visual behavior tree editor
2024live
SHEET 13 · Simulation
no plate captured · live build
Particle Animation Demos
Hand-written WAT + WebGPU compute shaders push 100K particles at 60fps across galaxy, murmuration, and airflow simulations
2024live