TPWC AI suppression system

[pellejones 1]

Not in the version that TPW released today...

[fabrizio_t 58]

Not in the version that TPW released today...

I meant you are correct, it should be fixed.

[ollem 4]

instead of soldiers going prone after x nearby shots, might it be better to handle suppression probabilistically? I wot not how much overhead that would introduce, but it would make the behavior that much more realistic.

Not really statiscally, but for sure easier to implement could be the following approach:

Right now every time a bullet is detected, in 'tpwcas_fnc_supstate' the enemy bulletcount is increased with +1:

line 28:

_enemybulletcount = _enemybulletcount + 1; 

based on a specific enemybulletcount threshold value (bullets counted) it's determined the soldier should go prone.

Instead of increasing bullet count by 1, we could randomize this value with e.g.

_bulletIncr = [1,1,2,4,2,1,1] call BIS_fnc_selectRandom; 

Or using: (untested!):

_bulletIncr = [1,1,1,2,2,3,4] select (round(random 6)); 

It would be possible to make this skill dependent. E.g:

if skill high then _bulletIncr = [1,1,1,2,2,3,4] select (round(random 3));
if skill medium then _bulletIncr = [1,1,1,2,2,3,4] select (round(random 5));
if skill low then _bulletIncr = [1,1,1,2,2,3,4] select (round(random 6));

Other approach could be to base the 'random x' on the current bulletcount number, which would allow to increase the 'probabilistically'.

e.g.

_bulletIncr = [1,1,1,2,2,3,4] select (round(random _enemybulletcount)); 

[Guest]

Updated release frontpaged on the Armaholic homepage.

TPWC AI Suppression system v3.02 beta

TPW AI LOS (TPWLOS) v1.02 v1.02



Community Base Addons

Edited August 10, 2012 by Guest
updated and enabled again

[XinSavior 1]

This sounds awesome! It's always been a pain trying to work around the fact that suppressing fire doesn't work on these guys. I can't wait to give it a try!

[ollem 4]

Talk about glacial pace of development, we haven't released a new version in 3 weeks....

TPWCAS 3.02: http://filesonly.com/download.php?file=367TPWC_AI_SUPPRESS_302.zip

I confirm there are some bugs in this release: both in script version and the pbo version.

so please ignore this version and wait for updated fixed version.

For those who can't wait: I'm able to provide a fixed interim script version if you send me a PM.

Updtaed official version will be released by tpw soon. Stay tuned..

[tpw 2314]

Many thanks to Ollem for bugfixing and additional stance modification stuff. And thanks to our testers too - your cheques are in the mail.

TPWCAS 3.03: http://filesonly.com/download.php?file=348TPWC_AI_SUPPRESS_303.zip

Changelog:

• Correct bDetect call in script version
  
• Correct wait for bDetect initialisation
  
• Cleanup of config variables in addon version
  
• Changes to stance modification code to make it CBA independent
  
• bDetect 0.73

I won't frontpage this until you can confirm it bug free.

And to all those complaining about filesonly: yes I know it sucks, most file hosters do. But at least these guys don't arbitrarily remove files. As soon as the release is confirmed stable then we'll provide an armaholic mirror (thanks Foxhound).

[kremator 1064]

Could you not just put it onto a Dropbox ? Would be SOOOOO much better.

[ollem 4]

Mirrored direct download link: TPWCAS 3.03: TPWC_AI_SUPPRESS_303.zip

[tpw 2314]

Not really statiscally, but for sure easier to implement could be the following approach:

Right now every time a bullet is detected, in 'tpwcas_fnc_supstate' the enemy bulletcount is increased with +1:

line 28:

_enemybulletcount = _enemybulletcount + 1; 

based on a specific enemybulletcount threshold value (bullets counted) it's determined the soldier should go prone.

Instead of increasing bullet count by 1, we could randomize this value with e.g.

_bulletIncr = [1,1,2,4,2,1,1] call BIS_fnc_selectRandom; 

Or using: (untested!):

_bulletIncr = [1,1,1,2,2,3,4] select (round(random 6)); 

It would be possible to make this skill dependent. E.g:

if skill high then _bulletIncr = [1,1,1,2,2,3,4] select (round(random 3));
if skill medium then _bulletIncr = [1,1,1,2,2,3,4] select (round(random 5));
if skill low then _bulletIncr = [1,1,1,2,2,3,4] select (round(random 6));

Other approach could be to base the 'random x' on the current bulletcount number, which would allow to increase the 'probabilistically'.

e.g.

_bulletIncr = [1,1,1,2,2,3,4] select (round(random _enemybulletcount)); 

Some nice ideas there.

I actually had a go at implementing count_roland's success = 1 - (f^n) formula.

Let's say a given enemy bullet has a 20% chance of causing suppression. f = 0.8.

suppression_threshold = 1 - ( 0.8 ^ enemybulletcount )

1 bullet = 0.2

2 bullets = 0.36

3 bullets = 0.49

...

10 bullets = 0.89

20 bullets = 0.99

if (random 1 < suppression_threshold) then {suppression etc....};

On paper this should work nicely. However, when I implement it into tpwcas_supstate and try shooting at a group of enemy, it usually takes 1 or 2 bullets and they're nearly all fully "red ball" suppressed. I need to set f = 0.97 or 0.98 in order for it to take more than 5 bullets to suppress them, and even then it's unreliable. I get the feeling that the random command might be the problem.

On another point, I don't think suppression should be skill linked. A properly trained and skilled unit should still hit the deck if shot at. If anything an untrained religious fanatic might even be less inclined to go prone. Just my thoughts.

[count_roland 1]

My original thought was just a fixed percent chance of suppression per bullet, no formula required. (The formula I gave was just the statistical odds of the occurrence of at least one event of a given probability in a given number of repeated independent trials.) But yes, I don't think random is as random as they say, though I found it out from the other direction. Using the snippet I posted earlier, with a 15% chance per round, the average was a little over 5, but the right tail of the distribution had way too many extreme outliers (meaning enemybulletcount > 20, which was at least 3 standard deviations from mean) to be normal.

Not sure what to do. Ollem's solution is by far the simplest, if there's no reason why enemybulletcount needs to reflect the actual number of bullets.

[dmarkwick 261]

On another point, I don't think suppression should be skill linked. A properly trained and skilled unit should still hit the deck if shot at. If anything an untrained religious fanatic might even be less inclined to go prone. Just my thoughts.

Well, I should say that suppression itself should not be skill linked as you said, in fact I would say that a skilled soldier would hit the deck sooner. But, the effects of suppression should be skill linked. So maybe there's a sort of inverse effect to think about: skilled soldiers go prone sooner but retain their skills, whereas less skilled soldiers remain upright longer but lose their skills.

[NoRailgunner 0]

I would say trained soldiers could recover from suppression and move on a bit earlier than someone who is not trained. Suppression is just what it is but its not somekind of magic skill leak/bonus.

[dmarkwick 261]

I would say trained soldiers could recover from suppression and move on a bit earlier than someone who is not trained. Suppression is just what it is but its not somekind of magic skill leak/bonus.

Well suppression has to manifest somehow, and ingame that is most effectively done by reducing the skills of the suppressed soldier. You have to imagine the purpose of suppression - it's not to make the opposition hide, that's more of a downside :) the purpose is to reduce the effectiveness of incoming fire.

[metalcraze 290]

The reduced effectiveness of incoming fire is because of opposition hiding.

[tpw 2314]

Well, skills suppression is already dependent on the initial skill and courage of each unit. A more skilled unit is affected less by suppression.

Recovery time at present is just a random number between 8-13 seconds.

I should say at this point that -Coulum- has just rewritten the whole skill suppression core and once I get a few attoseconds free I'll look to incorporating it into the existing framework.

[Variable 322]

On another point, I don't think suppression should be skill linked. A properly trained and skilled unit should still hit the deck if shot at. If anything an untrained religious fanatic might even be less inclined to go prone. Just my thoughts.

I fully agree. The accuracy of a suppressed unit should drop gradually, but the more skilled you are, the more proficient you are in dropping quickly and finding hard cover.

[kiberkiller 10]

And to all those complaining about filesonly: yes I know it sucks, most file hosters do. But at least these guys don't arbitrarily remove files. As soon as the release is confirmed stable then we'll provide an armaholic mirror (thanks Foxhound).

Mediafire doesn't have that asinine capcha or waiting time and it doesn't remove files. The only "downside" is that you have register to upload (not to download).

[cuel 25]

Or filesmelt

[tpw 2314]

Thanks for all the file hosting suggestions. Ollem has posted a mirror, please use that. Hopefully his server can be the default download location.

Ollem and count_roland:

Setting the tpwcas_st (shot threshold) from 10 down to to 5 and replacing

_enemybulletcount = _enemybulletcount + 1;

with

_enemybulletcount = _enemybulletcount + (random 1);

gives a nicer random feel to suppression, but keeps the computation in tpwcas_supstate reasonable.

If you wanted to make it skill dependent (which I don't), then you could always use

_enemybulletcount = _enemybulletcount + (random (1 - skill _unit));

Edited August 7, 2012 by tpw

[orcinus 121]

The reduced effectiveness of incoming fire is because of opposition hiding.

I think DMarkwick meant incoming fire in the sense of return fire. I don't entirely agree that their hiding is a downside, the big upside is that they cannot observe you so readily - especially flanking movements.

[kremator 1064]

I think DMarkwick meant incoming fire in the sense of return fire. I don't entirely agree that their hiding is a downside, the big upside is that they cannot observe you so readily - especially flanking movements.

As long as the enemy don't have 'spider sense' (as can happen with AI :) )

[pellejones 1]

Get dropbox! And use me as a referrer and I get some more space :P

http://db.tt/JXUAqVI

OR

http://www.dropbox.com <- is an awesome and easy way to share small files. The link above is also to dropbox but just a referal.

Edited August 7, 2012 by Pellejones

[kiberkiller 10]

Get dropbox! And use me as a referrer and I get some more space :P

http://db.tt/JXUAqVI

This is a completely wrong place for referrer spam.

[pellejones 1]

Just providing a VERY GOOD way to share files!