Navmesh cell size

Troubleshooting ai
8

I forgot to mention I think I found the cause for the original problem this thread started with.

There is not an actual code fix for it because what cause’s the

problem also happened in recast.

I think this method in jme3AI

private TriangleMesh buildNavMesh(float[] positions, int[] indices, IntermediateData intermediateData)

Is similar to this method in recast,

public RecastBuilderResult build(InputGeom geom, RecastBuilderConfig bcfg)

Both rely on a thread to build the mesh, recast uses these.

private RecastBuilderResult[][] buildSingleThread(InputGeom geom, RecastConfig cfg, float[] bmin, float[] bmax, int tw, int th)

or

private RecastBuilderResult[][] buildMultiThread(InputGeom geom, RecastConfig cfg, float[] bmin, float[] bmax, int tw, int th, int threads)

Both of these methods have a executor timeout of,

ec.awaitTermination(1000, TimeUnit.HOURS);

Heres whats going on. When these two variables

int tw 
int th

initialize they are tile width and tile height so a tile that is m_tileSize 64 for instance calculates to a 32x32 iteration in this for loop in either of the two methods,

for (int x = 0; x < tw; ++x) {
	for (int y = 0; y < th; ++y) {
		final int tx = x;
		final int ty = y;
		ec.submit((Runnable) () -> {
			result[tx][ty] = build(geom, new RecastBuilderConfig(cfg, bmin, bmax, tx, ty, true));
		});
	}
}

In this instance that’s 32 iterations tw and 32 iterations of th. The th part of this loop actually builds the mesh. This can take anywhere from 300ms to 1100ms per iteration.

On a rather small mesh that’s 32x32 = 1024 iterations x min of 300 ms or about 5 minutes. On large meshes with different tile sizes this can increase to 25k+ iterations at 1000 ms each or in laymen terms, 8+ hours to complete this loop.

So its not a problem that can be resolved by anything other than more computing power or just use smaller meshes.

2 Likes