POD-bot back into shape.

Hi people

I've been using the re-released version of POD for 1.6 for sometime now with no problems. Recently though after updating to the latest version ive noticed that when using the PODBOT menu to fill a team i can only fill the terrorist side. If i try and fill the ct side some sort of auto balance occurs. I have of course got autobalance off and mp_limitteams 0 so im lost as to whats going wrong. I even tried pb fillserver 2, believeing it could be a problem with the menu system, but unfortunately it occured again.

Any ideas?

Thanks

Hi People

I've been using the re-release of PODbot for 1.6 for sometime with no problems. However, i recently updated to the latest version and now using the PODbot menu i can only seem to fill the terrorist side. When i fill the ct side it appears to auto balance in some way or atleast the bots join both teams. I have of course got autobalance off and mp_limitteams 0 so im lost as to the problem.

Any ideas?

Thanks

[sPlOrYgOn]

i agree with everything that soul father has said..
except "bots can do everything a human can do"
i say this because they can't see throught the glass in cs_siege..
and the slow motion effect is while the computer is doing the vis tables i think.. but mine never stops slow motion because my computer is so slow it doesn't finish caculating all the stuff before the next map comes around which is 30mins..

[Austin]

Originally Posted by SoUlFaThEr

I figured out why vents are so difficult for a bot to navigate:
...
if he runs out of patience before getting to this next waypoint(stuck halfway then) he turns around......whether the waypoints are one-way or not......THIS is causing them to get stuck in cs_assault the same way

AWSOME detective work SF!

This is certainly VERY useful information.

[>BKA< T Wrecks]

Hey SF, thank you for that hint concerning vents etc. - I'll use that to improve some of the WPs I already made.
Concerning as_tundra: Well, some regular CT bots at least take the right side now, but the VIP still doesn't (my WP)

The autopath on/off feature is sooo great - I just finished cs_waterside (brand new CS 1.6 map, see new thead) and had to use it a lot to have only the desired connections made instead of getting lost in a maze of connection, 90% of which I didn't even want. And that brown line also helps
a lot!

The reload issue is a point that still needs to be addressed. Besides, while checking my cs_tibet WP yesterday, I saw a CT bot with hostages happily ignore the FL_NOHOSTAGE of a camp waypoint and... well, camp.
And the translucent player models at spawn points are a great feature as well, because it's very important to see where the start positions of each team are - this way you can place important waypoints much better. However, I only see those models in noclip mode - is that normal?

But there's one point that still keeps bots from reaching human-like performance: Buttons!! I'm not talking about doorbells, I'm talking about buttons that trigger doors or elevators. Here's some thoughts on that. I don't know the first thing about mapping, ok? So maybe this is bullshit... anyway, let's begin:
First a curiosity: When I place a WP close to a button, in the console it says something like "Button detected, flag set" - but I can't see any flag with "wp showflags". An "internal" flag? ???
Now to the point: Doors and elevators are entities, right? And the buttons provide an external trigger for those entities. You could say that an entity has an idle (0) and an active state (1). The idle state is the position where the mapper put it into the map, whereas the active state is the position it moves to when triggered. Ok so far? Now I wonder if it would be possible to determine in which state an entity is and somehow code a corresponding behaviour for the bots into the waypoints.
An example: An elevator is at ground level in idle state (0). When triggered with a button, it goes to the 2nd floor and stays there (1). Now a bot approaches the elevator at ground level. So far he'll maybe push the button (but maybe not) and then rush into the elevator shaft, thus being crushed by the descending elevator.
It's easily possible to force the bot to enter the elevator only after having passed a WP next to the button- yet the bot will only push the button randomly.
Would it be possible to get a kind of thinking process going: If entity in state 0, ignore button and go to next WP. If entity in state 1, push button and wait for n seconds, then continue? Could this be implemented by adding a real button flag to the WP and determine waiting time by a console commando (like "WP setwait n")? And this "wait" flag would only apply at the same time as the button flag, i.e. if the bot "sees" that the elevator is in state 0, he skips the button entirely. If he sees that the entity is in state 1, he knows he must use the button flag, and the WP tells him to wait 10 seconds before continuing? This would be perfect... 9_9

Hehe, I can almost see poor PMB getting angry in front of his 'puter: "Yeah, right! And would you like the bots to serve tea and coffee in the meantime, you ********??!?"

Seriously, I don't have a clue if even 10% of this is feasible - but one thing's for sure: So far, bots don't know at all what to do with buttons, period.
I heard Killaruna's ParaBot for TFC etc. can handle elevators quite well... maybe he can provide some useful information?

[[+Duracell-]]

Just a few comments that I'd like to make.

When the bot sees an enemy and runs away, it tracks through the wall too often and too long. I saw a CT bot on Italy run through the left path watching the wall; it saw an enemy eariler on the right path as it ran away.

Some of the chat flags don't work. When the bot says something like "Duracell is a damn cheater!", the name doesn't show. I believe it's the %s flag or something.

i don't know what else...you have done a GREAT job so far, PMB!

[sPlOrYgOn]

they can't see through glass..
i've only tested this on cs_siege..
let me get some demos later today..

[SoUlFaThEr]

the button thing is something me and Austin have been talking about for aq long time now.......the WAIT flag idea already exists but is not yet truth.....

we have to wait

the transparent spawn entities should show while the editor is ON......AND in noclip........so if its showing only in noclip.....theres something wrong.

[sPlOrYgOn]

I just found out they throw nades like idiots..
and also they cannot see through glass but i have not yet gotten a demo of that yet.
this demo is of them throwing nades.. very hard to get a demo unless you stick to one bot..

[edit]
crouch waypoints should have more danger weight cuz i see a crouch waypoint that is the "shortest" route, they all try to go through the crouch waypoint since it is the shortest but it is not the fastest.

on cs_siege where most of the battle takes place near that turn in middle of map..
they all try to go through the jump/crouch waypoints on the boxes because it is the shortest but not the fastest and they all crowd together trying to jump up the boxes then fall down and keep going...

and as for them not being able to see through glass.. I haven't been able to get a decent demo for it unless someone will make a map that has a glass box where 1 team spawns and the other team can look at them from the out side then you will see that they shoot by hearing and they don't actually see through the glass.
[/edit]
[edit2]
maybe different personalities and/or skill level will decide if they take the best-worst route to their destination
[/edit2]

[Pierre-Marie Baty]

I did notice they don't see through glass. I'm investigating the problem. I believe the engine PVS stuff is incriminated.

I know they shoot nades like idiots. The fault is their mathematical, statistical, discretized and thus non cognitive analysis of which waypoints are dangerous and which others not. They shoot nades preventively at the waypoint they computed that was the most dangerous for the position they are standing at now. This leads to very stupid reactions sometimes, such as bots on the roof of cs_assault happily shooting nades from the roof into the void. I'll see if I can fix this too.