Lopsided AI issue

[Trexian]

So, there I was, scripting a relatively simple mission to watch AI tactics - a bunch of AK-74 shooting CDF v. a bunch of AK-74 shooting insurgents, with me as a civilian observing.  I createCenters, then WEST setFriend [EAST, 0] and everything spawns fine.  Over several iterations, it becomes clear, though, that blue absolutely dominates red.  Like, for every 10 red KIA, MAYBE there's one blue.

Also tried it with the Eotech-AK variant Russian soldiers, and the CDF still has the same result.

Again, these are units with exactly the same kit.  What is also odd, is that the AI *seem* to be using the same kinds of flanking and covering tactics, just that blue is more accurate.  Is anyone else seeing this?

TIA,
T

[Wolfrug]

Seems to be a question of your own settings there mate. Have you been messing with the friendly/enemy dispersion values for instance in your config? Or as simple as your difficulty settings not making the playing field very level (friendly AI skill being higher than enemy AI). Try making your observer on the side of the other side instead and see if the numbers switch around

Wolfrug out.

[Rommel92]

This is directly linked with friendly and enemy AI skills in your difficulty settings. I had a 'what the' moment with it about a year back, same post almost. Hopefully that saves you some time. 

[Trexian]

Holy crap.

It honestly didn't even occur to me that on "regular" settings, the default enemy skill would be "novice" and the friendly would be basically expert.

Here's what may be stranger, though.  In this mission, we script in all the specific skill settings.  Will test this more, but it may be that the in-game setting may be a modifier for the scripted settings?

[i0n0s]

WEST setFriend [EAST, 0] only sets west enemy to east, east still believes that west is nicer than nice. Until west killed enough units :>

[Trexian]

Indeed, I0n0s, just tested that on a whim, too! 

Code: [Select]

WEST setFriend [EAST, 0];
EAST setFriend [WEST, 0];

Has it always been this way?  (Not that I ever tried it quite like this before, just curious.)

For any future searchers, the difference is DRAMATIC if the setFriend goes just on direction.

[i0n0s]

It was that way already in Armed Assault

[Trexian]

Bah!  Never new that.

Speaking of things I don't know (which would fill a library) there's a new, related, issue.

This mission basically monitors the attributes of large numbers of AI.  So, we spawn 10 groups of 10 blue, and have them fight 10 groups of 10 red (2x100 units).  When one side is attrited to less than 10 units, that iteration is over, the attributes are measured, reports are logged, the units (and dead bodies) are removed, and the next iteration starts with 2x100 units.

It goes fine for 14 iterations (2x1400 units).  Then, on the 15th iteration, it only spawns ~40 troops.  After that, it doesn't spawn ANY units.  I recall a max number of groups per side at any one time in A1, but I don't think that's what's going on.

TIA.

[i0n0s]

Did you deleted or reused the created groups?

[Trexian]

MMmmmm.. I guess the question contains the answer, eh?

Code: [Select]


{
if !(player in _x) then
{
deleteGroup _x;
};
} forEach allGroups;

That, at the end of each iteration, would probably be a good idea, eh?

Thanks (again).  BTW, it looks like, at least initially, that the AI attribute "shake" is probably most important to AI effectiveness, in a darwinian sense.

More on that later, though.

[Rommel92]

Its funny, you even named the group limit (144 groups if I'm not wrong).

[Trexian]

By way of explanation, and thanks, here's kinda what we're working towards.

This diagrams, over 15 iterations, the evolutionary aspect of some of the skill attributes.



Soon, I hope to set aside enough PC time for a 30 or 50 iteration cycle to chart the attributes that are evolutionarily important for AI.

(Yes, yes, some of them like command and endurance aren't specifically tested for in the current regime, but we intend to isolate certain attributes once we have a baseline set of numbers.)

[Trexian]

Ok, new twist.

After spawning 144 groups (and deleting all groups at the end of each iteration), it will continue to spawn, but it won't assign the typical unit identifier. We're used to seeing something like B 2-3-B:10 but after 144 groups, the unit identifier becomes something like just B :10 even though there could be several groups out there from that side.  It also starts to spawn very slowly.

I guess, for my purposes here, t'would be better to just have 10 groups per side, delete the units at the end of the iteration, then replace them in the groups at the beginning....

[Rommel92]

Ok, so I see your putting effort into this, so maybe I'm missing something. Are you saying that the skill level (even if set at the same number) changes with each group spawned?

[Trexian]

hehe

Funny you should post when I'm getting really frustrated.

The genesis of the idea was evolution (pardon the pun there).  Take alot of enemy, give them roughly the same skills, make them fight, then see what the attributes were of the survivors.  Take their attributes, introduce a random element - up or down - to the attributes and repeat.  Each generation of units is 1 iteration.  I had 10 iterations (no problem), then got the 144 group problem when going to 30 cycles.

---
Edit:
To actually answer your question, no the skills didn't change by the game, they were changed by script for each iteration.  Mimicking evolutionary changes.
---

This past weekend, I tried creating 2 arrays of groups, then placing 10 units in each group.  At the end of the iteration, deleting the units, but leaving the groups.  Then, at the beginning of the next iteration just re-cycling the groups - spawning new units in the groups.  Worked fine the first iteration, then failed remarkably.

The "goal" such as it is, is to test which AI skills are more important.  I see lots of people "tweaking" AI skill, but haven't really seen any analysis as to how the skills interact.

BI has created an incredibly complicated simulation.  Even in the limited testing I've done, I can see that many of the skills are related.  But the relationships can be complex.

Anyway, any help you can give is much appreciated.