Michael Booth's GDC 2004 talk on CZ bot...

[dead bwoy]

that would be me, although i have no experience codinig, I wrote the tutorial in the Turtle Rock forum on how to hand tune custom navs. The map learning process and analyzation of the danger/camping/sniping areas depends on the size of the map. A map like alleyrush.bsp took me +/-15 seconds. On the other hand, CS_Pacific and De_Alps took 45 minutes each just to be generated and are 1.2 MBs large. The generated nav areas seem to be based on a 2D grid, however you can create new nav areas between two areas at different heights; but they have to be directly N/S/E/W of each other... You can also use commands like bot_nav_jump/bot_nav_crouch/bot_nav_no_jump (the bots jump when they approach ledges because they cannot see whats ahead, Mike did this to try to be able to get the bots to make jumps), or bot_nav_precise which makes the bots pay closer attention to staying in the center of nav areas. I have not waypointed with any other bot, but I find it VERY usefull to be able to create a "zombie" bot and be able to select an area and use bot_goto_mark to test on-the-fly.They're having issues with ladders, I think that's why Militia has one sided ladders now, as a quick fix to the problem. The analyzation does not always detect the corect side of double sided ladders to climb, and there is no manual tuning commands for ladders yet.
Let me know if there are any non-coding things you would like to know about, although I sugest you join the forum: http://www.turtlerockstudios.com/forums/csbot/ Mike posts there at least weekly.

Sometimes, technical talk is only as good as seeing the finished product. I'd be curious to see how a Realbot or RACC would match the "Official" CS Bot...

[KickBot]

Cool dead bwoy! I'm not sure "Mike" would answer to a question like "In your commercial bot, do you use navmesh generated from bsp or do you use cells generated from a walking algorithm?"

Originally Posted by dead bwoy

The generated nav areas seem to be based on a 2D grid, however you can create new nav areas between two areas at different heights; but they have to be directly N/S/E/W of each other...

Mmmh. Do the generated areas all look like axis-aligned, or are there some that are diagonal and non-orthogonal to others?
For instance in cs1.6 Italy -dunno how cz one look like-, when you go at the CT spawn there are two alleys that start from there. How does the *generated* (not hand made) areas look in these alleys? For instance do they seem to follow an arbitrary axis and there are "holes" in the alleys with no areas, or do they follow the direction of the alleys and cover almost all the alleys?
A bird view screenie of the generated areas in the alleys + the CT spawn would be great

[dead bwoy]

CS_Italy_CZ CT Alley:

Here, you can see that in the diagonal alleys there are a bunch of non-diagonal nav areas forming little triangle holes against the walls. I feel that the bots navigate well despite this. Also, you cannot even create areas with diagonal borders. Only verticaly:

You CAN create diagonal areas between different heights, like the jump areas (with the X's) above. This shot also demonstrates the defined grid I speak of. Notice how the one jump area goes through the box. The other box is perfectly set on the grid. This "off-grid" box can be hand tuned to cover the entire top of the box, and re-creating the area between the box-top and the floor. Areas cannot be merged together if they are not exactly the same width, on the grid, on their shared border.

I hope this answers some of your questions.

[KickBot]

Nice job!! Thanks!
Well so after all it seems they use axis-aligned cells merged into rectangular areas, perhaps by some walking algo or just by sampling the bsp surfaces into a grid.

[dead bwoy]

I bet it's done by the surfaces! For example, cs_rio: Initial Analyzation creates a ring of nav areas around the outside, beneath the map. Areas where an alive player could not reach! Same as de_menace, the nav areas go through the fence and beyond at T Spawn, where no player (living) could go.

[[NvT]_KaszpiR_]

Originally Posted by Pierre-Marie Baty

*edit* out of curiousity, it seems that even the storage in the .nav files is very similar... it looks like a dump of the raw corner coordinates in Vector data I think I must be able to write a tool that will convert RACC nav files into CSbot nav files and back

would be amazing!

[[NvT]_KaszpiR_]

Originally Posted by Pierre-Marie Baty

That hull test thing must be a pretty damn CPU intensive algorithm, elseway. Can someone tell us how long it takes for the CZ bot to pre-process a new map ?

the de_aztec first pass takes about 1 minute on average maps
second pass is much longer - it eats cpu at max - around 10 minutes - the bigger map or more complicated the more time it takes

tests on p3 800

[[NvT]_KaszpiR_]

Originally Posted by Bill

Sometimes, technical talk is only as good as seeing the finished product. I'd be curious to see how a Realbot or RACC would match the "Official" CS Bot...

the turtlerocstudios bot is the best one i have seen ( i consider here podbots, plbot, realbot , sturmbt for dod, got no idea about racc (no time) )
plays smooth as pb26, i got lags on realbot unfortunately
takes different paths,
obey orders,
they can ambush sometimes nicely a player
then can define what last route was a "mistake" and try another way in achieving bs
auto generating waypoints on new map when adding bot

there are flaws (at least in version i got)
they run with gun as if they were running with knife (doh)
the aim on highest skill is really lame (well this can be easily fixed )
no linux version

[Pierre-Marie Baty]

KaszpiR, please consider editing your posts instead of triple-posting