Did you know that WFC algorithm can generate ~unbiased (not-necessarily-shortest) paths between A and B? Each tile here has an even number of outgoing edges, therefore all paths split into cycles, except A and B, which become connected by a path! Tiles by @beast_pixels.
Thread.
Thread.
Take a tileset with Lines, Turns and DeadEnds, but ban DeadEnds everywhere except 2 endpoints. The key is to NOT include T-shaped roads, as they have odd number of outgoing edges, and the result then doesn't have to split into cycles. X-shaped intersections are fine though.
WFC may be an overkill for simple 2d paths, but it's very handy for more complex geometries like 3d stairs or heavily constrained tilesets.
The most common method I've seen to generate not-necessarily-shortest paths:
1. Generate a random metric (as noise, for example)
2. Find the shortest path in that metric.
reddit.com/r/proceduralge… By controlling noise frequencies (octaves) you affect the resulting path frequencies.
1. Generate a random metric (as noise, for example)
2. Find the shortest path in that metric.
reddit.com/r/proceduralge… By controlling noise frequencies (octaves) you affect the resulting path frequencies.
And you can go for models more interesting than Perlin noise, your random metric doesn't even have to be induced from a heightfield. Examples from Scott Sheffield's website: straight lines in random geometries
math.mit.edu/~sheffield/spo…
math.mit.edu/~sheffield/geo…
math.mit.edu/~sheffield/spo…
math.mit.edu/~sheffield/geo…
Another fun method for generating ~unbiased paths: @boris_brave's chiseling algorithm
1. Start with everything covered in path tiles.
2. Remove tiles one by one until only a thin path remains.
boristhebrave.com/2018/04/28/ran…
It can connect >2 points or be stopped early for thicker paths.
1. Start with everything covered in path tiles.
2. Remove tiles one by one until only a thin path remains.
boristhebrave.com/2018/04/28/ran…
It can connect >2 points or be stopped early for thicker paths.
