I made a try on a sphere based on the study below. I dont understand everything but found the pic 4 5 6 quite clear on how to proceed. It is too complex for a tutorial. but you can check the zip containing the model for the loop. the loop being the piece for spanning tree/maze creation.

https://www.sciencedirect.com/science/article/abs/pii/S0097849307001835

in particular fig 4 5 6 are easier to understand.

​​​​​​​

Maze algorithm to get the spanning tree is based on Aldous-Broder. which is not optimized at all but we get harmonious (uniform) results. I had to set 10000 max iterations without being sure to finish the maze... https://weblog.jamisbuck.org/2011/1/17/maze-generation-aldous-broder-algorithm

Buckblog: Maze Generation: Aldous-Broder algorithm

weblog.jamisbuck.org

to be investigated when time:

* for a UV type of surface, modeling the maze structure planarly in a matrix first and then mapped to the divisions of the surface could be faster than pure geometrical way (which I did)

* other algo than Aldous-Broder, but I am not good at algorithmic

ICD_HOTTOPICS ​​​​​​​Iloverd ​​​​​​​