Barbarossa - analysis of AI at different command levels

[CharonJr]

AG level AI, month 2:

Even after switching the AI to Attack mode there are still a fair number of low odd attacks like this one - this one here actually has decent odds, but I saw much worse, too.

And I noticed that the AI is actively using interceptors above enemy airbases here (it seems to choose one and mostly stays there till either the base is overrun or the Soviet planes are rebased).

After having closed the gap to its north HG Süd is starting to focus more on the south (as it should due to its objectives), but it is still operating further to the north than I would have thought at first.

This might be caused by trying to compensate for HG Mitte's focus on the northern part of its front.

HG Nord has taken Riga and similar to HG Süd it is spread out right to the middle of the frontline, too. Basically HG Nord and Süd both seem to ignore that HG Mitte is actually (at least supposed to be) handling the center.

HG Mitte itself is starting to spread more to the south now and seems to move troops into position to attack its objectives in the southeastern part of the marshes.

And 2 weeks later, even before the 3 objectives are taken a fair number of units of HG Mitte are moving to assist in crushing the pocket to the west of Riga - far away from any of its objectives again.

The army group AI is lagging even further behind the theater AI now, especially in the south.

Casualties for the month are 79k (SU) vs. 85k (GER), bringing the totals up to 208k vs. 182k - the theater AI was at 211k vs. 183k by this point. So we have very similar casualties with a fairly large difference in provinces gained.

This might be a first indication that the lack of coordination of those army groups is making a difference. It will be interesting to see if this trend will continue.

And it has to be kept in mind that I am comparing a 1 month Blitz 1 month Attack AI to a 2 month Blitz AI here at the moment. The progress is fairly similar to what the pure Attack AI at the theater level achieved after 2 months.

Heeresgruppen Nord and Süd stay spread out up to the middle of the front with some of their forces while HG Mitte is fairly concentrated at the center at the moment.

Despite HG Mitte looking fairly centered at the moment I suspect that its strat redeploying all over the front that I have been able to see before (first to the south, later to the north again) is the result of the lack of coordination mentioned earlier.

If this trend continues (and I suspect that it will) I hope that not only coordination between allies will be improved with 1.4, but coordination between lower level HQs inside one country will be improved, too.

From what I have seen the main problem is the combination of the lack of coordination on the one hand with a fairly high degree of freedom for the AI concerning how it tries to go after its objectives on the other hand.

CharonJr

 

[Modo]

Less objectives might allow the AI to control the offensive better. You might be forcing its moves. Specifically, I saw the AI switching the bulk of its forces (at all levels) from one objective to the next, not spreading units around except to cover the front. This means that multiple objectives create a very bad yoyo effect. It's the main reason why Japan does so badly in most games—it has too many objectives that aren't in one general direction. It's also the reason AI armies need mobile units—the yoyo effect will still be there, but units will reach their destinations faster, with less time spent in idle redeployments.

Two things about low odds attacks:

1. Sometimes they turn into overruns as more units enter the fight. It looks similar to a player selecting a bunch of units and telling them to attack at once, without checking if they are a) in position and b) still under the attack delay. They get the job done, but it causes more losses than coordinated attacks.

2. Pinning attacks are ok in my opinion, especially if done with armor (even if you lose, you cause more casualties than you take). I like that the AI does them.

 

[CharonJr]

You might have a point with HG Mitte having too many objectives, sadly the other 2 HGs just have 2 (Süd) or 3 (Nord) objectives remaining - all well away from the center of the front, but still the AI has a fair number of units around the center.

Especially HG Süd should focus on Odessa now (the 2nd objective is fairly close to it, too).

And the attempt of HG Mitte to help with the crushing of the pocket west of Riga is very troubeling to me, too, since HG Mitte has no objectives around Riga, all its objectives are around the marshes.

So IMO essentially the AI takes too many freedoms and is not stringent enough in its pursuit of the objectives it has - even when setting a direction for the attack. But giving the AI a direction does improve its following the orders, just not as much as I would have liked it.

And yes, I actually like the armored attack, but it did the same with INF and even paras more than once, those battles were lost fairly quickly and I was not able to see any secondary purpose in those attacks.

Overall I think the ratio between usefull and useless low odd attacks was about 50:50.

CharonJr

 

[Modo]

And the attempt of HG Mitte to help with the crushing of the pocket west of Riga is very troubeling to me, too, since HG Mitte has no objectives around Riga, all its objectives are around the marshes.

That's what you get when you attach HQs to the theatre. They will also move against partisans and front breaches (be it on the line, or by separate invasions). The idea of borrowing units from other HQs is sound, but I too don't like the fact that such missions are hardcoded to always happen when under the theatre.

 

[CharonJr]

Good point about the HQs under a theater, I have not known this one before.

Similar to comparing Blitz and Attack orders I might compare army groups attached to the theater and unattached during this run.

CharonJr

 

[Kirth Gersen]

Yeah, Gladiator's AAR shows this pretty well. I prefer to see the theater HQ as the Home defense HQ. It can be too intrusive when placed in the HQ hierarchy, particularly on in a multi-front war, where it can disrupt your plans (usually for very good reasons) dramatically. I'd rather lose the Leader bonus. See here fore a drastic bit of theater AI intervention (4th picture in post 89):

http://forum.paradoxplaza.com/forum/showthread.php?t=443965&page=5

On a different note, have you been using Axis of advance with your objectives? Many people find it very useful in keeping the AI armies on target, and attacks more incisive. Would it disrupt your analysis?

 

[CharonJr]

I used axis of advance for HG Nord (can be seen in the 4th screenshot of the last turn) and while it is not running around like a headless chicken I would have prefered it to stick even closer to the axis (mainly the southern armored units).

CharonJr

 

[CharonJr]

AG level AI, month 3

I start this month by streamlining the objectives of my army groups a little bit and giving my marines and paras the 2 Soviet pockets as objectives hopefully leading to attacks and freeing up units if/when those pockets are crushed.

In the pocket west of Riga 140 Soviet brigades are trapped and I have to hope that they run out of supplies soon since I only have about half the number of units in the area. Otherwise I might have to switch objectives of HG Nord to this pocket, too and I would dislike calling off its move towards Leningrad/Tallinn. This first attack was called off a few hours after it began.

I am hoping for some bomber support by assigning the pockets as objectives for my bomber heavy special units, but so far not that much is happening. And once those bombers were damaged due to the large number of Soviet defenders even this small bombing did stop.

The AI is trying to crush the Riga pocket as I wanted from the marines, but looking at those combat odds despite the Soviet units being involved in an offensive, too, I think that I will need additional forces there. I gave HG Mitte an additional objective there and removed the other objectives for now - so HG Mitte only had the 2 pockets as objectives.

And while the AI is sending units of HG Mitte towards the pocket those might be not enough to stop a nearly full Soviet army before it is able to reach the harbour and IC in Memel.

And from what I have seen from the AI usage of the bombers - at least at Attack settings - it seems to rest them once their strength goes below 70%. I will try to test this by setting HG Mitte and my marines and paras back to Blitz in order to deal with the pocket/breakout.

A few hours after changing the attack stance the assault on the Orany pocket is started with 4 TACs in support. But after 2 hours the AI stops the attack despite 70% odds which I think is odd...

The AI seems to be worried about the Soviet Riga army, too, since it is moving units from HG Nord there, too, which has no objectives around there. Sadly it is pulling most of those units from the semi-breakthrough it has achieved south of Tallinn.

It seems like there is no difference between the strength level at which the Blitz or Attack AI will send TACs to rest since I keep seeing no active TACs at less than 70% regardless of its stance.

And the Soviets crushed Romania and if they continue south Bulgaria will fall next. Hungary might give them more problems, but the way to the Hungarian capital is fairly open to the Soviets.

Concerning the TACs I saw my KG 76 returning to action once its strength was back at 75% and it stayed grouded once it went below 70% again. So 70% and 75% seem to be the trigger levels here.

The Soviets in the Riga pocket (the breakout is contained) are out of supplies now, but the large number of units I (via objectives) and the AI (both from HG Nord and for some reason 1 armored division from Süd) have moved there (about 100 brigades now) is hurting on the remaining fronts where the AI kept being counterattacked and pushed back.

Shortly before the end of this month the Riga pocket was finally crushed and 140 Soviet brigades ceased to exist and the AI was moving the troops there back to the front/remaining pocket.

HG Mitte stayed under Blitz orders and its only objective was the Orany pocket.

Progress is OK despite my focus on crushing the pocket and having taken out 140 Soviet units in the Riga pocket. But before focusing on the pockets the AI was already on the gates of Tallinn and Kyiv, so the setbacks are partially my fault. But I would hate to think what the Soviet army in the Riga pocket would have been able to do if it had been allowed to break out.

Casualties were 62k vs. 65k bringing the totals up to 270k vs. 247k, the theater Blitz AI was at 249k vs. 224k at this time.

With HG Süd more concentrated in the south now and large parts of HG Mitte having moved north or staying right at the center the Soviets regained some provinves around Kyiv - the AI was right at the outskirts of Kyiv already.

Except for the dispersed HG Mitte (part of it my fault due to sending them in to crush the Riga pocket) and some units of HG Nord which are moving back to the front after dealing with the Riga pocket, the AI keeps its units together fairly well (for some reason HG Süd moved 1 armored division to help deal with the Riga pocket, too).

What bothered me the most during this turn was the attack on Orany which was stopped after 2 hours of fighting despite having 70% odds.

HG Süd moving 1 armored division up to help with the Riga pocket felt strange, too, but since we are just talking about 1 division here I was more surprised than annoyed.

In addition I would have liked the attack stance to have some impact on how long the AI is willing to use its TACs, but overall 70% to stop and 75% to resume bombing look like decent numbers to me.

CharonJr

 

[Modo]

But I would hate to think what the Soviet army in the Riga pocket would have been able to do if it had been allowed to break out.

This seems to be an AI trap. It creates the pocket, but is too eager to move past it, and can get into trouble if it loses Riga or Memel. I had the same problem in my last Barbarossa. Fortunately my special forces army was just leaving Britain so I had spare units to contain, otherwise something would hit the fan.

The AI probably weakened its nothern position more as it faced less dangerous enemies there (mostly broken divisions, right?). This shows how important reserve units are (and not only when using the AI). Even an understrength reserve army would be perfect for dealing with that pocket.

 

[CharonJr]

Only about 10% of the divisions were completely broken at first, about 40% were somewhere between 25-50% while the remaining 50% were somewhere around 75% from what I have seen. So basically the Soviets still had a numeric advantage which the AI decided to ignore.

IMO the AI put too much thrust in the militia units which it had around there - basically it seems to count them as if they were combat divisions - and even then 70 brigades vs. 140 Soviet brigades which are concentrated in 2 provinces and can decide between which of 5 provinces to attack seems like not such a great idea to me. And most likely it treats the HQs as combat divisions, too.

And in contrast the Orany pocket seems overprotected IMO, but I can only guess the number of Soviet units there. The main problem there is the AI trying to have a decent number of units in every/every other province and since we are talking about 7 provinces this takes a lot of units.

Instead it should compare how many units is needs/has for guarding those provinces with how many units would be needed to attack with good odds and simply attack if less units are needed this way.

I just wish I had any idea why it stopped this 70% attack...

CharonJr

 

[loki100]

I just wish I had any idea why it stopped this 70% attack... CharonJr

This is based on recent observation but could well be wrong. I noticed at the end in my war with the USA it suddenly started to break off from what were winning attacks - one on a VP province that was quite isolated started at 65%, got to 75% (at which stage I was thinking about pulling back) and then it stopped, whilst earlier it would hang in with an attack to the point of breaking.

My suspicion is that the caution coincided with severe manpower depletion and that it suddenly became very keen to protect this from being further drained.

 

[CharonJr]

This is based on recent observation but could well be wrong. I noticed at the end in my war with the USA it suddenly started to break off from what were winning attacks - one on a VP province that was quite isolated started at 65%, got to 75% (at which stage I was thinking about pulling back) and then it stopped, whilst earlier it would hang in with an attack to the point of breaking.

My suspicion is that the caution coincided with severe manpower depletion and that it suddenly became very keen to protect this from being further drained.

Hmm, interesting point. MP was at about 210 then and since it will only get worse now I will try to see anything similar happens during the next months.

This might be another reason for the AIs unwillingness to sometimes crush pockets without being essentially forced to - it fears to lose too many units by attacking such a large number of forces as can be usually found in pockets.

CharonJr

 

[Cybvep]

It seems that testing AG-level AI when assigned to Theater is a moot point, because it still contains "Theater logic", just at a lesser, more disorganized level.

 

[CharonJr]

It seems that testing AG-level AI when assigned to Theater is a moot point, because it still contains "Theater logic", just at a lesser, more disorganized level.

This sums up my current feelings fairly well, but I will play this one till October, too, then redo it with the army groups removed from the theater.

But since we are talking about a VERY small number of samples here it is hard to draw any real conclusions from one run, thats why I try to play at least 5 months for each run (maybe more if the overall result is still uncertain then).

For lower level AIs I might still leave the HQs attached to their higher level HQs and do a testrun here, but I will only write a short summary about it.

CharonJr

 

[Modo]

It seems that testing AG-level AI when assigned to Theater is a moot point, because it still contains "Theater logic", just at a lesser, more disorganized level.

I don't think it's theatre logic, it's more like theatre visibility. And the same thing happens at the army level, just with lower ranges, so you don't see such extremes. I still find it useful when having naval and air units "hidden" below the HQ level you are automating, for containing unexpected problems which a disconnected HQ would ignore, and to get the leader benefits.

 

[Cybvep]

Don't take me wrong. It's perfectly fine to include Command Structure under Theatre HQ IF you want to micromanage everything. Otherwise, it's best to use Gladiator's formula - Theatre with Garrison and MP troops to garrison the beaches, VP, arifields and fight partisans. It works fine and Barbarossa under AG detached from Theatre and with proper axis of advance is much less pain in the ass. 1.4 patch will probably modify these findings, as "Theatre logic" is changed there and allied territories are incorporated into Theatres. It will be VERY interesting to test AI in the new environment, especially if supply transfer through allied territory is fixed. I also hope that they will get rid of Theatre-involvement-into-lower-HQs-when-not-greenflagged once and for all.

 

[loki100]

I also hope that they will get rid of Theatre-involvement-into-lower-HQs-when-not-greenflagged once and for all.

I understand what you mean but fear this would bring a mass of problems with it. Without some intelligence on what is going on elsewhere, an AI controlled army/army group would probably go inert - to it there would be so many threats to its goals that it couldn't even start to sort them out, or go bonkers and just deal with its orders in a charge of the light brigade manner.

So it'll need, as now, some means to take account of the wider situation than just its collection of troops and orders. It might be ok if all the armies/army groups/corps (whatever level you use) are on AI but could break down if only some are - and I quite like mixing up AI control of the more straightforward areas with my own control of more interesting or more critical sectors.

 

[CharonJr]

I am currently taking a look at the 1.4 beta, actually I am tying to bring it up to Barbarossa, too, and see if anything has changed.

Due to this I think I will only be able to post an update at the end of the weekend.

CharonJr

 

[CharonJr]

With 1.4 beta still having a lot of problems (personally I saw mainly bugged range and vanishing supply/fuel) I am back with this AAR earlier than I thought.

AI AG level, month 4

The AI kept being pushed back in the south, but with Bulgaria being conquered by the Soviets we might see some Italian offensive now, I just hope that the Italians will not lose too many troops as "expeditionary forces".

What bothers me is the unwillingness of the AI to attack the Orany pocket despite having 30+ divisions around it and the Soviets should be running out of supplies inside the pocket by now. And while Orany is marsh terrain I still think that the AI is too fearfull of entering low infra provinces here. The divisions that could be freed up by crushing this pocket would be really helpful at the other parts of the front. And both the paras and HG Mitte still have blitz attack order for Orany...

I tried a rerun of this month with HG Mitte and my paras detached from the theater, but still under AI control - but the result was the same, the AI did not attack.

At the end of this testrun I ordered some of the units surrounding the pocket manually to attack since I wanted to confirm if the Soviets were out of supply by now and how the combat odds would be.

And as I suspected the AI would be able to crush this pocket easily, I am using less than half of the forces surrounding the pocket and the odds are at 81% - but as I saw this seems to be a problem regardless of the AI controlled HQ being attached to the theater or not.

While the units wasted by sitting around the Orany pocket seem to hurt the German AI at least the other Axis member seem to be more successfull with Hungary starting to push back the Soviets and Italy basically rolling over the Soviets in Bulgaria.

Hardly anything happened during this month and the advantage the AI had over the theater level blitz AI in the north has vanished while still being quite some provinces back in the south.

With not much happening casulaties of 50k (SU) vs. 47k (GER) are fairly low during this month, the totals are at 320k vs. 294k (318k vs. 306k for the theater blitz AI).

edit: Dang, forgot the current HG positions:

The AI pulled its different HGs closer together again, it will be interesting to see if it keeps the HGs as close together when detached from the theater.

Overall fairly poor performance from the AI this month which I think is largely due to the inactivity around the pocket.

CharonJr

 

[CharonJr]

AG level AI, month 5

I got lucky here, since Bulgaria decided to continue fighting Italy keeps all of its troops for now and should be able to put some serious pressure on the Soviets in Romania.

Starting in mid September Germany starts to have a fairly high ratio of won fights again which might be an indication of the Soviets having run out of MP.

HG Nord keeps pushing and is at the gates of Tallinn.

By the end of September Germany itself has run out of MP.

And while the there is not much happening in the middle (including the AI still wasting troops by surrounding the pocket - and most likely due to it actually) there is a fair amount of activity in the north and south.

Progress in the north and south look very similar to what the theater blitz AI had accomplished by this time. Only south of the marshes progress has been worse and (like most things not to my liking here) I tend to blame this on the units held up by surrounding the pocket. The southern marshes are supposed to be in the sector of HG Mitte, but it is still focused on the pocket.

Considering that the AI has already been as far as Kyiv this shows quite nicely what happens when you lack the forces to secure the provinces you have taken and are being pushed back.

Casualties were 46k vs. 49k this month, bringing the totals up to 366k vs. 293k which is quite a bit lower for Germany than with the blitz AI (380k vs. 374k), showing again that the Blitz AI suffers from engaging in too many close battles (a lot of which are actually lost).

Mainly HG Nord is split up between the north and the Orany pocket now, but despite splitting up its forces progress in the north have been quite decent which might be a further indication for the current Soviet weakness due to having run out of MP.

As suspected the Soviets have run out of MP some time ago and now need 240 MP to bring up its units to full strength. In addition they have lost about 400 brigades during the 5 month we have fought now.

The Soviets actually lost less brigades than at the theater AI level (about 100), but considering that the Soviets in the Orany pocket should be essentially lost the levels are fairly close, but still with an advantage for the theater AI which might be another indication of the better coordination.

So while the AIs position is slightly weaker now than under the theater AI I am fairly confident that the Axis will still win this one.

Basically the AG level AI (with its HQs attached to the theater) performs very similar to the theater level AI, right down to be too unwilling to crush small pockets.

The main difference seems to be a slightly worse coordination between the different army groups.

Next I will do a quick run with the army groups detached from the theater level to see if the results will be markedly different.

CharonJr