Translation: Blender Shortest-Path Labyrinth

Source

Source file: https://download.blender.org/demo/geometry-nodes/labyrinth-shortest_path-simon_thommes.blend
Source group: Solvable Labyrinth Generator
Geometry Script recreation: examples/geometry_script/blender_labyrinth_shortest_path.py

Why This Graph

This graph is the first accepted pathfinding translation. It uses grid topology, randomized edge costs, shortest-edge paths, path selection, path-to-curve conversion, stored UV attributes, curve radius, and endpoint markers.

Graph Map

Build a square grid and store its UV map on the corner domain.
Mark vertex 0 as the path root.
Use edge-neighbor count and random booleans to create edge costs.
Run Shortest Edge Paths, convert the result to an edge selection, and capture that boolean on the edge domain.
Subdivide the mesh and delete selected points to carve the walkable graph.
Run shortest paths again on the carved graph, find the farthest X point on the path, and convert the edge path into curves.
Instance endpoint marker spheres on the path endpoints and capture their instance indices as Start/End.
Trim the solution curve by the exposed Solve value and capture a Path boolean on the curve domain.
Convert maze walls and solution curves into filleted radius curves, then convert them to mesh with a circular profile.
Join endpoint markers, path/wall geometry, and the base grid.

Metaphor: the grid is a street map, random edge costs close some roads, shortest path draws the route home, and the captured booleans are colored tape left on the roads so later carving and tube-building know what survived.

Verification

Run:

& 'C:\Program Files (x86)\Steam\steamapps\common\Blender\blender.exe' --background '.\experiments\source-blends\blender-labyrinth-shortest-path.blend' --python '.\tools\verify_labyrinth_translations.py'

Results from Blender 5.1.1:

Case	Source vertices	Translated vertices	Delta	Status
small unsolved	22730	22730	0.0	accepted
small solved	25484	25484	0.0	accepted

Edge and polygon counts also match for both cases.

Lessons

Shortest-path graphs can be translated as topology passes first, visual passes second.
Capture path booleans before subdivision/deletion. Once the graph is carved, the original edge selection is no longer obvious.
Anonymous Value nodes are easy for converter output to misassign. The source links showed 0.1 / Size driving radius and 3.0 driving marker scale; trusting the generated variable names produced the right topology at the wrong size.
Path outputs are fields. Geometry comparison catches the visible contract; scalar/boolean path outputs need a harness only when they become the target.

2.8 KiB Raw Blame History