PB: "Block with hostage" ignored by Bots?

[Pierre-Marie Baty]

I've made a few changes that I knew by heart be necessary for a bot to work in CS 1.6. I've also followed TurtleRocker's advice to put the FL_FAKECLIENT flag back. Try this one and tell me if it works any better.

[botmeister]

Originally Posted by Pierre-Marie Baty

I've made a few changes that I knew by heart be necessary for a bot to work in CS 1.6. I've also followed TurtleRocker's advice to put the FL_FAKECLIENT flag back. Try this one and tell me if it works any better.

Bot works slightly better. They at least sometimes they move around, but they don't seem to follow a path and it looks like random movements. Most of the bots are just stuck in one spot. When they do move they walk instead of run. This same thing happens with Realbot under the updated CS. I'll test again without mEAN in case that has something to do with it, stay tuned.

[Pierre-Marie Baty]

oooooooooooooooOOOOOOOOOOOOOOOOOO *contained anger*



Need help here.

What is their max speed ?

the fact that they seem to do random movements only comes from that they believe they are stuck, and they are trying to unstuck by jumping/ducking/strafing, etc. They don't follow paths when they are trying to unstuck.

Could you use the PMTools plugin to display the entvars dump for a bot entity who is "walking" (trying to run) ? It's the "printent" command. Use the "logfile" CVAR to send the output to a file.

Does it look like a bad msec problem ? Do the bots seem to move in "slow-motion" ?

[botmeister]

OK, the podbot_mm.dll works great! The problem is with mEAn because the pfnSetClientMaxspeed hook does not work. I use it for the silent walking effect. I think there's a thread in here about this, where it won't work under CS 1.6 but it does for CS 1.5. I'll look up what the solution is.

btw, great job on the podbot clone. I'll put it on my bot server and see how it does in the real world.

[botmeister]

Damn! When the map changes I get this after trying to add bots

[metamod: command 'pb' unavailable; plugin unloaded]

This is the same problem we saw with realbot, so I think it may be a bug in metamod.

[>BKA< T Wrecks]

Uhm... I hate to interrupt you guys, but I have also tested the new dll (the first one posted, the one for 1.5) for several hours, and these are my impressions:
1) The bots hardly fail to respect "BlockWithHostage" flags - maybe in 1 of 25 cases or something. I tried with cs_duke2k, which is ideal for testing because there are two hostage rooms connected by a vent. Originally CT bots activated the hostages and jumped into the vent, taking the shortest way to the second map goal - but leaving the first two hostages behind. Now they respect the waypoint flag in most cases.
2) Bots DO seem to need more time to learn the map; first they seem to hesitate and move a bit back and forth rather clumsily, but then they slowly start finding and trying out ways, until in the end their navigation is ok.
3) One drawback is that they don't take ways they used to take occasionally before (tested with de_karachi01 and my own waypoints, a map full of ledges, rooftops etc., offering almost endless possibilities); their selection of ways and routes through the map seems to be a bit more "conservative", less open to experiments. Might just be my impression, though.
4) It might be (once more, maybe just my impression) that this annoying behaviour of jumping when walking around corners on narrow ledges (thus falling down and hurting themselves) occurs more often.
5) In cs_duke2k, bots sometimes froze for a fraction of a second, then jumped, although there was no reason to jump (i.e. when running through a hall or on the street). I didn't notice that so explicitly in other maps, however.
6) In de_karachi the whole game seemed to be slower than before... like in a slight slow-motion effect; fluent but slow. Hm, maybe a reboot and new start would have fixed that - dind't have time to test yet.

So all in all, I think your "quick dirty hack" did what it was meant to do - it (virtually) fixed the hostage problem! Otherwise, it might need some tweaking, but I have had too little time to perform a thorough testing routine so far.

[Pierre-Marie Baty]

Well lads, I'm glad to hear it worked that well

In answer to your comments:
1 - There is now nothing in the source code that would make a bot decide to pass through waypoints which have the FL_NOHOSTAGE flag set, unless if the bot has not noticed that the hostage was actually following it (i.e., the bot thought that he failed using the hostage for some reason). All the other cases must be waypoint problems. Either there is a waypoint which has not the flag set, or there is a pathwaypoint that skips the hostage blocking waypoint.

2 and 3 - Yes, this is a totally new pathfinder for the bots, which leads them to adopt a new behaviour. They will need to learn everything about the map again. But normally, when they'll realize they get shot easily if they follow the main route, they WILL take the other routes, and in the end it should happen much more often and tactically than before.

4 - I haven't changed anything in the bot source code related to that. One day when I'll be less lazy maybe I'll give it a go, but for now perhaps you should try to turn the problematic waypoints into "crouch" waypoints, the bots would not go as fast and hopefully they would fall less often.

5 - Probably a peak of lag in the game that made the bot think it was stuck... The experimental A* pathfinder that Count Floyd had started to write is not really optimized, that could explain why lag peaks happen.

6 - If the map has a large number of waypoints, and there are a large number of bots running around and shooting stuff at once, chances are that the A* pathfinder is spinning like crazy doing all the path computations and can't keep up with the frame rate... Remember, the pathfinder is not optimized.

All in all, great ! I think I can release the source code of this modified podbot DLL as soon as the current issue with Steam is fixed. Maybe someone will have a go at fixing the other problems in a clean way. The modifications to make are ridiculous compared to the original source code

[>BKA< T Wrecks]

Wow, that's what I call a complete reply! I'll try and do the same:
1) In the tested map, there is definitely no wp problem. However, those few exceptions are explained by what you said, and moreover occur so rarely that apart from someone explicitly looking for occasional bugs (like us) nobody will ever notice.
4) What's funny is that with the new dll, those deadly jumps occurred at different spots than before. And what really freaks me out is that there's a corner where CT bots manage to jump around, whereas T bots fall down... o_O mmmm... gonna make'em all crouch WPs and see what happens.
2,3,5 and 6) Yes, I think this A* pathfinder has its little flaws... Yesterday, when I wanted to test some difficult routes, that bot totally ignored my "debuggoal" commands and never even reached one of the points I specified :'(
Maybe it's because the map has close to 600 waypoints with lots of connections, many of them involving jump combinations etc. and the pathfinder has a hard time getting along with that?

Btw, would it be possible (without totally insane amounts of work) to include a kind of "security flag" that forces the bots to
a) not consider a waypoint as reached before they are actually standing precisely on the spot (right now, there's a certain radius, and if bots move inside that radius, the waypoint is "reached". This can lead to problems with ladders & ledges as well as with those "walk-into" buttons present on some maps).
b) approach the WP walking, unless it's a jump connection?

Don't feel rushed; I'm just curious, because this would make the bots' navigation as close to perfect as possible - from a technical point of view. From a tactical point of view, your PB plugin already leaves little room for improvement... And if it's not possible, maybe I can still use some dirty tricks like placing waypoints inside walls to force bots to run into a button etc... 9_9

[Onno Kreuzinger]

afaik CS player minutes do still rise, so no reason to stop, atleast not until hl2 ....

[Whistler]

....and if you want to have a bot can use the new weapons and the shield, you can take a look at this one (source code)....

http://groups.msn.com/weimingzhi
Goto the 'documents' link at the left, then select 'podbot.exe'
(it's also a modified POD-Bot, and it can work as both Metamod Plugin and normal hook DLL bot)