Navmesh with Blender

Yep, I’m aware of that :smiley: . The problem is that it isn’t doing the “spiral”. It is just doing the floor and not very well (as you can see, the gray fragment is overlapping the spiral. It is going on the 3D on a weird fashion).

I think I know where the error is…
Look at the huge distance to the outer edges of your gray quad.
Now look how wide your staircase is.
You need to configure a thinner walker (reduce that value to around 5 percent of what it is now).
EDIT: you can also configure the cell size for the voxels - it should be much smaller than the walker.

EDIT:
and that there is no nav mesh under the staircase may have two reasons:
a) Your walker is also very tall ( so the head bumps the underside of the stair ).
b) You know, these navmeshes are not full 3d, they are often 2d or 2.5 dimensional.

I’m not familiar with Blender’s Navmesh generator, but you probably want to be sure you’re not using ngons. Which is to say make sure your mesh has only quads and/or tris.

The problem couldn’t be the walker height, it wasn’t even creating a navmesh for the spiral.

Yeep, I know how the navmeshes work :stuck_out_tongue:

Good one, but wasn’t the problem.

The problem was that the spiral and the floor were the same object but disconnected, like two meshes. Two possible solutions?

  1. Separate them on two different objects.
  2. Connect them (ie: connect the two spiral rampage start vertex with the floor)

The resulting navmesh (a good one) is:

http://i.imgur.com/Izav5st.png

It is not optimized at all, there still some parameters play, but it’s a usable one.

2 Likes

Another thing talking about recast. I saw some time ago that @Tihomir started a project to use it on jme calling it jNavigation. The wiki talks about it usage but I think this was never finished, was it?. Is there any recast for jme project in the hoods?.

I don’t know. I did not have time to actually use much of the engine jME in the last 11 months. I was hacking new engine components to suit my basic needs because many features were poor or missing. Next time I touch a game engine things must be available and I will be a user, not a maker of engine stuff.

In your last image the distance to the border of the ground quad is zero - so something must have changed.

If Blender really uses recast, then this might not have been the problem. Recast uses voxels and doesn’t care about the connectivity of triangles in the mesh. If your action really solved the issue, maybe Blender doesn’t use Recast.

You might need to model an “underside” of your ramp, because the direction of the normals of the triangles might matter to detect where the outside is - but I could be wrong here.

I just don’t know enough at this point and every further thought is speculation.
Since you know how navmeshes work - maybe you know enough about Recast too.

I can’t remember what parameters were in that image but the point is that I tried much parameters before deciding to post one of the results (every of them were very likely).

Well, I can’t ensure it but it’s what they say at their wiki.

No, only some fast investigations about what recast is.

Okay, the wiki states that Recast/Detour is used as external libraries. I think that answers this uncertainty.

The quesion remains: What is the better workflow?
a) make your scene in Blender editor and then import the game level (the scene)
b) make your scene in jME (either SDK or via level editor or procedurally generated) + Recast

I think, if a) is possible and there is a good asset pipeline and a stable workflow for: everything Blender => into jME, then a) might be the preferred solution for all non-procedural games.

The problem is: Since I did not have time to make games and was stuck making engine stuff, I don’t know much about the best asset pipeline - only for text assets (fonts, Unicode, and some ideas about management of text assets and localization).

Well, I’m currently doing my own pipeline “engine” and is just the a) option ;).

But I must say that I would like to have the recast advantages for game runtime too. For example, the ability to update the navigation mesh based on scene objects addition/removal and many other functionalities (I think recast give this easily. I’ve been looking some information about critterai’s nmgen and recast but I can’t say I know enough to talk with real knowledge about them)

However, for static scenes, I’m going for a).

Yes, look at this video: Ogre Recast/Detour demo: dynamic navmeshes - YouTube
It’s the Ogre Engine (open source, C++) using Recast at runtime.
I think that is a must have for every game engine (except from RPG maker which is 2D and cell-based).

1 Like

Got a link to that CritterAI thing?

http://www.critterai.org/projects/nmgen_study/diffs.html

Totally agree but not in position to implement it :chimpanzee_sad:

The only other thing I have is this link: Navmesh generation in runtime

That guy called rickard is really great. I know him since jME 2 (2006). Also have his book about jME recipes.

Well, there him was talking about generating a simple mesh in runtime, not modifying it dynamically. Modifying it on a runtime-friendly way isn’t the same than recreating it on runtime (the first requires optimizations based on the already made navmesh and modifying it analyzing only the specific changed zone so it is much faster than recreating the full scene, or at least is what I think I know xD)

I’ve seen forum topics about separating the recast mesh into tiles.
So you only need to recreate a tile if something in the tile has changed.

But this is with recast xD, what isn’t on jme yet, we come back again to my question, is the recast library planed to be added on any fashion java way or jNavigation (which is currently stopped) is the only project?

There was a google summer of code project for adding it.
Don’t know if it ever got finished - @coredevs might know…

Some found info about it:

(This is why I think is stopped)

1 Like