Replying to Skirmish AI v1.9 beta2 now up!

Actually, the "AI not capturing certain points" is quite map-related. Most maps its fine but on some its like a curse is placed on those unclaimed LPs. Wierd. 6p_Mortis is like this especially for Player0 (If an AI is placed there). Bizarre. Not sure Arkhan can correct this "anomaly" since its rare.

I was referring to my game.

Larkin, are you referring to one of my reports ? If so I presume it's the first one ? Just curious ... I'll have to check back if I can notice any un-capped LP's in both tested games.

Thanks.

I guess it was the capturing plan. Threat on path detection isn't really the strength of the AI.

There were two uncapped points just near their hq which it could capture easily.

B

P P







E

B:Base
P:uncapped Points
E: Enemy

How long is the harassing phase ? It was way 6 minutes into the game.

4 Minutes!

I also gave the eldar example were a single squad (the only squad eldar had) followed the enemy instead of recapping its points.

I guess it was the capturing plan. Threat on path detection isn't really the strength of the AI.

Is it possible to give some more control for setting these attack values ? I want the orks to gather a big army but they waste their guys too much running around.

I'll send you the new version tomorrow which should have less chaotics in the harassing phase. I think I also add the attack / retreat points to the race parameters. This should give you a bit more freedom in this area.

Armies are attacking though they have barely a value of 300:
Should only happen in harassing phase where the AI has some more tolerance during attacks.

How long is the harassing phase ? It was way 6 minutes into the game. I also gave the eldar example were a single squad (the only squad eldar had) followed the enemy instead of recapping its points.

Is it possible to give some more control for setting these attack values ? I want the orks to gather a big army but they waste their guys too much running around.

Don't know when it crept in but I get the feeling there are too many move commands. Once commited orks should stay at CC. Looking at the battles it seems that CC troops and jumpers commit for a short time just to retreat - just to reengage - just to retreat. As if the code isn't sure about the best action and they are trapped between two different tactics.

Attaching problems:
There were indeed some problems with the min strength of the squads to attach. Further, the Orc commander had the wrong attach code. Probably a copy/paste problem.

Capturing back and forth moving:
In general very rare. But I increased the required squad strength for guarded crits, which should help to avoid this problem.

Capturing of critical locations:
The AI doesn't capture crits in the harassing phase to get more harassing power. This was a design decision because Slash reported that they capture too much instead of harassing. This is a good compromise.

Reduced commander health limit:
In WA commanders seem the only ones which are able to really hurt a LP. I did this after I saw a replay of Zenoth where chaos lost about three squads in a row to a LP because the commander was attached and no unit able to effectively hurt the LP.

CC combat is still weak...:
I tried to reach some more "Do it or not" behaviour, but these are fundamental problems which can't be fixed so easy.

Armies are attacking though they have barely a value of 300:
Should only happen in harassing phase where the AI has some more tolerance during attacks.


New version should be ready tomorrow. Further issues are delayed to DC since 1.9 is intended mainly as a fix for 1.5 balancing and performance.

WoW! Fantastic detective work there Zenoth! Bravo! Very comprehensive. Very much like your use of screenshots!

Yes the CL thing is a little dicey. Having the AI send infantry to capture them backed up by another force especially vehicles may/may not be possible until DC. See your point though and good analysis - we need that kind of observations. Thank you!

Report #2 for 1.9 Beta 2

Match: 2 Vs 2 (Me as CSM, allied with SM Vs. 2 x Orks)
Map: Mountain Trail
A.I Difficulty: Hard
Resources Rate: Normal/Standard
Resources Sharing: Off/No
Winner: Me & Ally

Comments: Here I wanted to see how the heroe attachments would end up, and as I thought, it ain't constant. In this test only one of the three A.I players never attached its heroe (Ork). All others did. And that in itself isn't good in my opinion. Also, I noticed a strange "back and forth" behavior by the Orks trying to capture one of the CL.

1 - As I mentioned, in this test, two A.I players out of three attached their heroe to infantry. One of them didn't, and it was a disadvantage, since all other players around were skulking their heroe around with infantry ready to help him right away. In my previous report, none of the A.I players attached their early heroe units. It's not the type of issue that is easy to reproduce, but it happens regularly enough as it seems.

Picture 1:

http://img76.imagesh...roealonebw2.jpg

As you can see, here's the Ork heroe that ventures alone in the wilderness, eventually (but valiantly I guess) taking one of my Listening Posts by himself. And of course he never managed to do it, since I eventually killed him. Notice the game time there, 5:33 minutes in (althought all the heroes were trained ealier).

Picture 2 & 3:

http://img128.images...attachedon1.jpg
http://img128.images...ttached2nb6.jpg

By 5:33 minutes, both other A.I players had their heroe attached, and they had better success on the battlefield trying to take out enemy LP's because they were attached. Not like that Ork one. As I just said, notice we're at the exact same game time in all three pictures shown so far.

2 - During this game, I wanted to test out something I already knew how to test. Which is what I believe to be a map-specific issue, but it's very problematic. Note that it's also very easy to recreate for testing purposes.

Right when the game began, I sent one group of Cultists to the Critical Location in question. And that CL is the one directly north to my base (east side of the map).

When a human player controls that CL before the enemy, as well as leaving the troops that originally controlled the CL (in a guarding stance) there, then the A.I will wait very long before trying anything to capture it back (sometimes they simply do nothing). And when they do try something, they usually send one (extremely rarely more than one) infantry group (I've never ever seen vehicles anywhere near that CL, but that could be related to paths placed there by the map author(s) so to allow infantry only in that sector of the map, I just don't know if that'd be the case). And in this very case, it was a single group of Tankbustas.

Note, however that the enemy in that test had enough troops to take it back from my mere Cultists ...

Picture 1:

http://img134.images...lcapturebj0.jpg

All those Ork troops were waiting there, for gathering purposes I presumed. Because they eventually attacked me (they didn't stayed there for too long). But they could have resumed their LP/CL capturing-spree up to that CL so easily. They never tried. Instead they gathered as I expected (I was just waiting for them as you'll see in the replay) and they eventually tried to take me down.

Picture 2:

http://img149.images...stubbornrj4.jpg

Here notice the game time. By around 20:00 minutes in, that Tankbusta group tried in a "back and forth" fashion, to capture my CL. All they did for around 7 or 8 minutes (non-stop) was to run towards it, against my Cultists, properly attacking them on their way, but eventually turning back and retreating. Sometimes they lost one unit in the group, and they reinforced it regularly. Back and forth, non-stop, trying to get it, but never managed to. And while they tried to capture it, my forces were already laying siege to their main base (and they never turned back to help about that). I just thought it'd be worth it to show you that A.I behavior guys.

But as I mentioned earlier, that issue in Mountain Trail is inherent to it. It's surely a map-specific bug. Perhaps you Arkhan/Thudo/Larkin can take a look into it. I'm not sure if I have to call that a "bug" in the first place. But as a tester I couldn't just ignore the "back and forth" behavior. Because I know that Arkhan worked on such issues before. Well the annoying thing is that it's still happening and is very noticeable. Perhaps not happening as often as before though. But althought the frequency at which it occurs is smaller, it seems to remain something to take a look into.

So for this test I focused on looking for heroe attachment "issue" (might still be subjective). And any problems around LP's and/or CL's if I saw any. And I saw one, as reported here. That's about it for now for this test guys. Hope it eventually helps.

For some reason Arkhan reduced commanders health limit to <= 10% before they attach at gamestart. Not sure why.

We'll look into that. Thanks Larkin for noticing!

And the gathering issue isn't happening on LP's thudo (if that's what you meant by "based on where the LPs are"). It's occuring under Critical Locations. And it's never all of the CL's when there is more than one on a particular map, it's just one, never more. They gather at one of them, and that's it. If nothing else around the gathered units "un-stucks" them, they don't move at all.

Sorry.. Criticals.. Then I haven't noticed it. I run multiple games and the only issue I've seen is when the AI doesn't finish off an enemy sometimes especially when they have a leftover barracks or generator sitting in their base OR secondary base. This happens when any other ally is defeated so there is one enemy to defeat left over. The AI will sit there (whether Relic, Strat, or Critical) and do nothing. Odd..

Your problem I have never seen except right at the end of the game but its not limited to Crits. I play Mountain Trail ALOT and never saw this prior to nearing the end.