Greetings I'm doing some research into how to implement pathfinding and I'm now stuck on generating info that can be used by a pathfinder for large and complex meshes. Imagine having a large/huge mesh, a city, or a cave, something large and full of complex objects. How would you design a system to generate pathfinding information for something like that? Traditional node-based approaches are not very reliable, as the geometry is way too complex. A Navigation Mesh approach seems the way to go, but generating this mesh is pretty hard, as generating it by simplification of the map mesh produces a too complex nav-mesh. Any ideas? Thanks in advance, Rui Casais

One way to do this is to de-tessellate (is that a word?) the navigation mesh into largers areas that are navigateable by the character that is moving through the world (i.e. throw away all the little tiny triangles and quads that the character couldn't fit within and join them up to neighboring triangles or quads).

By "throwing out" all the tiny areas and joining them with larger areas, you wind up with fews surfaces that you have to deal with. Then create a method of connecting these navmesh surfaces together (usually using the common edge between surfaces).

Don't forget that you can toss out any triangle or quad that isn't going to walkable. No need to include walls and ceiling polygons if your character can't walk on these surfaces.

Look at the Game Programming Gems books or the AI Game Programming Wisdom books for some tips on creating navigation meshes for you path finding.

botman

Although not optimal, what I've done in the past is loop through all the triangles in the scene for a game, and after discarding faces that don't face "up" enough, and whatever other filtering mechanisms you want to use, it ends up with a list of triangles that are walkable. Then I centered navigation waypoints in the middle of each face, and then double for-looped through all points to do ray tests to determine visibility for connecting them.

It's not perfect by any means, and not even optimal, but that was a while ago in an experimental project, and it works and is pretty easy to implement. It's basically a way to auto-generate a waypoint graph, and isn't a proper navigation mesh.

An alternative would be to loop through all the triangles and generate a graph by using the adjacent triangles as the connections, giving each triangle a max of 3 connections. Once again you'd probably skip unwalkable faces or faces that are too small.

One of these days when I get more time I plan on having a go at a proper navigation mesh pathfinder.

I'm developing FPS bots, and my current pathfinding implementation is waypoint system with some extra parameters such as waypoint radius that can be tweaked to provide loose or tight following between certain areas. Waypoint systems are so simple and still pretty decent representations they are still used in 95% of commercial fps games.

While definitely not optimal, an intermediate step could be to manually create a nav mesh. Depending on the content, this can actually be done pretty quickly. This would allow you to get all of the AI runtime side working with a nav mesh, and once that is done, go back and look at ways of automating the construction of the mesh itself.

Rui,

if you're new to pathfinding, I recommend using (manually) placed waypoints/nodes; this allows you to focus on pathfinding itself, rather than on generating a high quality terrain representation (a big enough problem in itself).

Otherwise:
- read Paul Tozour's "Building a Near-Optimal Navigation Mesh" (ch 4.3 in AI Game Programming Wisdom)
- read Jan Paul van Waveren's MSc thesis "The Quake III Arena bot", p. 23-45
(available at http://www.kbs.twi.tudelft.nl/docs/MSc/2001/Waveren_Jean-Paul_van/thesis.pdf )
- buy Half-Life 2, install the SDK via Steam, select "Create a mod" and then inspect the source-code installed in your "mod" directory. It apparently contains the BSP parsing/mesh generation code by Mike Booth developed for his official Counter-Strike bots (as discussed at the 2004 GDC: http://www.gdconf.com/archives/2004/booth_michael.zip).
Mike's code is in src/dlls/nav*.*. Remember to first read Valve's license accompanying the SDK, and don't forget to play the game!

Have fun,

William

I would like to recommend a paper called Interactive Navigation in Complex Environments Using Path Planning. The chapter about roadmap preprocessing describes a good algorithm for automatic nav-mesh creation.