Hows Navmesh working out?

[Maleficus]

Just finished reading your Navmesh tutorial - nice work.

But I am curious - how is it working out for you? I've seen the CZ bots, which from what I understand uses a slightly different version of Navmesh, and it looks pretty good in game, as well as the fact you don't have to waypoint maps anymore (tho I like the idea of being able to hand tune navmeshs if desired)! Are there any performance penalties, or memory restrictions?

Fritz uses node based pathing, and tho it isn't always pretty, its very cheap and fast. I've been working on extending it a bit so it looks and works better than traditional waypoint systems. But its still a pain to have to make those nodes, often taking hours of time per map! :'(


I've been studying a q3 bsp loader, and see that getting the planes from the .bsp file is relatively easy, its just making the Navmesh, then on top of that the pathing code, work.

I don't really have time for it ATM, I have so many other AI issues to battle with, but I'm definately interesting in learning more about Navmesh.


ps: any plans to continue your Navmesh tutorial as promised?!

[stefanhendriks]

ever considered auto-noding?

[Maleficus]

Yes.

I'm open to both actually, the problem for me is figuring out a way to read the .bsp and noding/navmeshing intelligently based on that info.

[Rick]

Waypoint flooding works sometimes aswell

[Pierre-Marie Baty]

I've little time right now to make the long answer I'd like to, but remind me to do so at the end of the week

I *plan* to write other tutorials, no problem, but what I lack currently is time. Argh. Time. The navmesh tut is intrinsically "finished", though... since what it lacks is more relevant to "pathfinding" (with A*) and "pathwalking" (with algos I'm currently fighting against, grmbl).

On this topics navmesh vs waypointing, you may well want to have a look on what I'm doing for Stefan's Realbot. A utility to create nodes file from the BSP files.

Nothing rocket science but it uses concepts from Pierre-Marie's navmesh (mainly walkable surfaces) and generate waypoints based on collision avoidance between adjacent nodes as well as 'probing' the floor for dangerous fall or too high jumps.

So, it gets some bonus from navmesh but still keeps the drawbacks of waypoints: pre-defined course of bots, A* takes much longer.

Source code is GPL and is in the Bsp2Rbn subdirectory of realbot.

HTH

-eric

[@$3.1415rin]

well, predefined courses ... let's say precalculated distances and maybe use them for heuristics ... but I think here are only a few ppl who'd like to miss A*

Pierre-Marie,

By looking at your tutorial & RACC code to detect walkable faces, I'm afraid that the check is not fully correct . It should be:

if (((plane->normal[2] < 0.707106) && (face->side == 0))   // Based on PMB navmesh tutorial,
		|| ((plane->normal[2] > -0.707105) && (face->side == 1)) // Addition by evyncke
		|| ((plane->normal[2] == 1) && (face->side == 1)))
				// a simple check to see whether plane is flat or walkable < 45 degree
		continue ; // Discard this face

Without the handling of negative normals, the de_aztec slope between the river and bomb spot was never processed.

It looks like 99.9% of walkable slopes are using a positive normal but 0.1% are using a negative normal but with side == 1.

HTH

-eric

[Pierre-Marie Baty]

ohooooo

well my friend, consider this copy/pasted and credited

[V or 'Tex]

)) // Addition by evyncke || ((plane->normal[2] == 1) && (face->side


It seems to me that the comment does not have an ending spot. Unless the logical or ends it?