Pathfinding through denoising
A diffusion model that solves mazes. It was never taught to search; it was taught what a solved maze looks like, and it recovers the route from pure noise. Draw a maze below, or let it finish one for you, then watch the path resolve.
Loading model
Drag either endpoint to move it. Hold ⇧ or use the right button to invert the brush.
Specifics
- Architecture
- U-Net, 5,218,561 parameters
- Board
- 23 × 23 pixels, an 11 × 11 cell maze
- Training set
- 20,000 mazes carved by recursive backtracker, solved by A*
- Solve rate
- 94.1% of 256 held-out mazes, against 94.5% using all 1000 steps
- Inference
- onnxruntime-web, WebGPU with a WebAssembly fallback. Nothing you draw leaves the tab.