Crash on complex scene

[GoogleCodeExporter]

What steps will reproduce the problem?
1. Load this .obj file into RecastDemo 1.4 ( 
http://dl.dropbox.com/u/3236655/navmesh.zip )

What is the expected output? What do you see instead?
Many of my other (equally complex) scenes work fine, but this one seems to 
cause some kind of memory error.

What version of the product are you using? On what operating system?
I am using Recast 1.4 on Windows Vista SP2

Please provide any additional information below.

This is for my game Overgrowth, and I think Recast and Detour will save me 
a lot of time. Thanks!

Original issue reported on code.google.com by david20...@gmail.com on 22 May 2010 at 12:50

[GoogleCodeExporter]

Please use the latest version from SVN. It is always the safest bet. 
Unfortunately I'm a bit lazy when it comes to 
making releases, instead I do my best to keep the latest svn version tidy. For 
major features or fixes, I usually 
detail them in my blog.

Recast uses voxelization process as intermediate step to calculate the navmesh. 
Huge worlds like your mesh 
will need to be split up into smaller chunks so that the process does not run 
out of memory.

In Recast this means tiles. Each tile is a square piece of navmesh and the 
tiles can be loaded and unloaded at 
runtime, Detour will stitch the mesh tiles together when a tile is added to 
create continuous mesh.

You can load your navmesh.obj to the (latest) Recast demo and choose "Tile 
Mesh" sample.

In your case I suggest using tile size around 128. You may need to use cell 
size around 0.15-0.20 to capture 
the finer details around the house. I tried with cell size 0.15 and agent 
radius 0.4, and it seemed to create ok 
results there.

The terrain is a bit tricky for Recast, but there are a couple of things that 
can be used to fix that.

Recast generates detail mesh for every navigation mesh polygon. This helps pick 
correct polygon when there 
are multiple layers of navmesh and it also can be used to approximate the NPC 
height location too.

The detail mesh generation is sort of geared toward smaller polygons (uses slow 
algorithm). From 2.5D 
walkability point of view your terrain is almost just a bumpy flat plane. The 
problem is that Recast tries to build 
a detail mesh for a large area with a lot of height differences and that is 
slow.

One way to remedy this is to adjust the detail mesh generation to create less 
accurate mesh. For example 
increasing Detail Mesh Sample Distance to 10 and max sample error to 2 speeds 
things up quite a bit (I got 
around twice the speed) without too much loss in accuracy. Another one is to 
keep the maximum area 
relatively small, in this case it means keeping the tile size moderate.

I'm big fan of Overgrowth, I have kept keen eye on your progress :) I hope 
Recast/Detour could be useful for 
you.

Original comment by memono...@gmail.com on 22 May 2010 at 1:41

[GoogleCodeExporter]

Thanks for the advice! I am using the tile mesh from the latest SVN version, 
and it no 
longer gives me this memory error. I am not so worried about mesh calculation 
time 
because I think the navigation meshes for each level will be cached in advance, 
though 
it's still good to keep it responsive for level creators. Maybe we should have 
a fast 
low detail pass for testing, and then allow for individual tiles to be refined 
as 
necessary (such as complicated buildings).

Original comment by david20...@gmail.com on 27 May 2010 at 4:25

[GoogleCodeExporter]

If you can keep track of the tiles that change during one edit session, you 
could
just update the ones that just changed. In your cases it should be just matters 
of
few seconds to do that. In my tests one 128x128 tile took ~150ms to calculate. 
In
case you change the whole terrain, or certain instanced asset changes 
drastically,
then you will need to rebuild all. That should keep the iteration times tight.

Complicated buildings may actually be processed faster than open terrain :) The
generation algorithm works better and faster when you have smaller walkable 
areas.

Original comment by memono...@gmail.com on 27 May 2010 at 6:20

[GoogleCodeExporter]

I'm closing this bug for now. If you have further question, maybe better place 
is to discuss them in the discussion groups at 
http://groups.google.com/group/recastnavigation

Original comment by memono...@gmail.com on 8 Jun 2010 at 5:21

• Changed state: Done