A question nobody has ever asked [-ne 10^(-N), -de 10^(-N)]


(mena666) #1

Hi!
This is a question about so called epsilon.
There are two of them, distance epsilon and normal epsilon.
I played around with them in quake3 and found out the following facts.

  1. distance epsilon has almost no effect
  2. normal epsilon, when reduced, makes auto clipping secure at the expense of increase of bsp size :rock2:

So my question is what is the best value for q3 aas creation?
Should I disable -meta?
Because it speeds up aas creation.
Still and all bots do nothing at all, though. :roll:


(Shallow) #2

Some advice: When you have a problem and want people to help you, it’s best to clearly state what the problem is, rather than spouting psuedo-technical stuff that 99% of the people reading this forum won’t understand, and probably has absolutely nothing to do with your problem anyway. Just like in your lightgrid thread, you’re making people guess what you need!

So.

If I filter through your post to find the relevant stuff:
-You are using autoclip.
-Your bots are inert.

So I’m going to guess most of your map is models? Big stuff, like caves or terrain?

The bad news is that even if you get a valid AAS out of BSPC, sometimes it can be too complex for bots to use it to navigate. If your map has a lot of autoclipped stuff in it there’s probably lots of untidy clusters with funny angled surfaces in them. Bots really don’t like that.

General bot optimisation techniques will need to be applied. Cover complex geometry with greatly simplified botclip brushes. Make sure any empty spaces that are outside the map’s playable space, but inside the hull, are filled with clip or botclip too.

To answer the -meta query: Always use -meta, it makes bigger surfaces with more shared vertices which is much better for modern video cards.

As you are probably aware, you must use the -forcesidesvisible switch on BSPC when you compile your BSP with -meta.


(mena666) #3

Thanks Shallow for your advice.

O.K.
Directly speaking, I just wanted to know the reason why those switches “-ne, -de, -snap” were implemented.

As for autoclipping, I tested with simple axial meshes and -ne 0.0001(x10 of the default value) to see if bots can navigate through the arena.
And they could.
This means, as far as meshes are rather simple, bots can tolerate with autoclipping, I guess.