Barbarossa - analysis of AI at different command levels

[CharonJr]

Just finished the test with the HQs detached from the theater and have to say that if you really want to play a hands-off game (including not micromanaging objectives fairly often) this was the worst setup I have tested so far.

Since it is 2am around here and I have to go to work today I will write up some more about this run when I get back from work.

CharonJr

 

[CharonJr]

AG level AI, detached from theater

Nothing really different during the first month, including armored divisions supporting attacks that have not taken place yet and might not for some time.

After 1 month HG Nord and Süd have extended/spread their positions again up to the middle of the front (no change from being attached to the theater) and HG Mitte is focused on the south of the middle part of the front.

Overall progress is pretty similar to the AGs attached to the theater, too - the lacking progress in the north is compensated by additional progress in the south. No large gaps I am able to see and after this month I will change the attack stance from blitz to attack.

Again a large pocket has been formed west of Riga, and this time I immediately give my paras and marines an objective inside this pocket.

In the south Kyiv has fallen and here we can see an armored division trying to support an attack that is currently not taking place as mentioned before.

As usual Romania got overrun by the end of June.

The AI tries to finish off the hurt Soviet units in the Riga pocket before they are able to catch their breath and sends in some bombers as support, too.

During this month HG Süd has switched its focus more to the south while HG Mitte moved more to the middle. HG Nord is still spread out down to the middle of the front with a focus on the area east of Riga, it is just using about 1 korps at the Riga pocket.

After 2 months progress is still fairly similar to the run with the army groups attached to the theater with progress in the north still being worse while it looks better in the south. And the Riga pocket is being taken care of this time in contrast to the previous run with the army groups still attached to the theater.

With the Riga pocket crushed the marines are tasked with taking Tallinn and Leningrad while the paras are set to take Minsk.

As expected Bulgaria surrendered by the end of July which will finally free up the Italians.

In the meantime is has become fairly quiet on the German/Soviet front, most likely the combatants have to bring up their ORG which has suffered during the last weeks. HG Süd is in worst shape with HG Nord being the best.

HG Mitte stays fairly focused on the middle now except for some MTN divisions operating in the south. And HG Nord is spread down to the Minsk salient.

The northern front is very passiv now with what looks like most of the Soviet forces in Skandinavia having arrived on the front and are pretty much forming an impenetrable front (at least in the mind of the AI).

The Soviets managed a breakthrough at the southern front, this is the first instance where I think I saw a weakness in comparrison to the HQs still being attached or under actual theater control - because there are hardly any units around which are able to contain the Soviets or cut them off.

With still nothing happening in the north despite most units having fairly high ORG I switch HG Nord to Blitz mode hoping to force an attack there.

And yes, the AI is actually attacking and the odds are not even that bad.

While having been able to take Minsk and even forge a bit further northeast the AI lost Misk again and 5 divisions are cut off and about the be taken prisoner. I try to set a bunch of objectives for HG Nord and Mitte at provinces which might break throught to the encircled units, but even if the AI is willing to react I doubt that it will get there in time.

Overall without being attached to the theater the AI seems to prefer deep penetrations leaving the flanks vulnerable to encirclements unless you are willing to do a lot of micromanagement again with changing objectives.

Personally I think I prefer keeping the HQs attached to the theater since while deep penetrations and due to this encirclements seem possible with detached HQs having to reset objectives fairly often is too close to micromanagement to me.

With having broken the stalemates on the fronts and no hope the break the encirclement I switch back to the attack stance.

Dang, Romania has been liberated by Italy with hardly any Romanian provinces taken, this will throw a big wrench into my hopes for Italy putting pressure on the Soviets.

And it looks like it is about time to do something about the expanding Soviet breakthrough in the south, HG Süd objective is changed from Odessa to cutting off the Soviets for now.

But large parts of the units stay at their positions, most likely due to the large number of Soviet troops facing them.

I think this final screenshot shows quite nicely what happens when you detach the HQs from the theater without wanting to micromanage the objectives for the AI. The lack of coordination and neglecting the flanks (both on a penetration level and concerning the complete front, too) are the main differences I have seen.

Overall casualties are 312k for the Soviets vs. 321k for Germany, the worst result yet and this is without taking into account the German divisions lost to the encirclement. In addition the Soviets are still fielding 1200 brigades, which is the highest number of all the tests so far, too.

So nope, if you want a basically hands off game detaching the AI controlled HQs looks like a very bad idea to me. This is the first situation where I would be fairly doubtful if the German AI would be able to pull off a victory.

CharonJr

 

[Modo]

Sounds like a breakthrough/exploitation army or two could use being separated from the command structure. Prodived, naturally, that units under the theatre are moving along to cover the flanks.

 

[Cybvep]

That's disappointing, although I can't understand two things.

1. Why no axis of advance?
2. Why no objective micromanagement? The whole point of AG detached from theatre is AI that is actually following orders instead of following its own "Theatre logic". From my experience (and from Gladiator's thread, too) I can tell that it's harder with AG under theatre.

Sounds like a breakthrough/exploitation army or two could use being separated from the command structure. Prodived, naturally, that units under the theatre are moving along to cover the flanks.

It's an interesting way of player-AI interaction - give AI infantry and only some armored units while the player handles airforce and breakthrough armies. Might work, considering that AI creates "stable" frontline.

 

[Modo]

If I read this part of the AAR correctly, the mixed approach should make the disconnected units less sticky when covering the front. More than once did half of my panzer army sit around covering a sector just because the infantry: a) hadn't yet caught up, or b) was stuck covering another sector with too many units. If the panzers would just move on without watching their backs so carefully, the infantry might be forced to redeploy more aggressively, creating a domino fix.

This makes me wonder. Would detaching HQs and giving them objectives some distance back from the front force the AI to do some actual retreating? It would be interesting to test that, especially at the lower levels.

 

[GUNNM]

That's disappointing, although I can't understand two things.

It's an interesting way of player-AI interaction - give AI infantry and only some armored units while the player handles airforce and breakthrough armies. Might work, considering that AI creates "stable" frontline.

I tried that in a game. As Soviet I built forts along the first river to try a trench war. I controlled the retreat back to this line and then gave all the troops to theatre ai. After 2-3 months of neither side wanting to attack, I gathered 25 tank divisions and first tried to break through in the north. I succeded, but did not manage to penetrate more than a province or two.

Then I relocated to the south and there I managed to break through and surround about 10 divisions. The Germans tried to counterattack but due to the forts and river they soon reached 0 ORG. At that point my AI troops counterattacked and broke through. I then proceeded to drive my tanks north-west with the AI troops following behind and surrounded almost the entire German army.

Worked quite well. Better than expected really.


But when it comes to AG-AI and objectives, I usually change objectives every week or two. I dont consider it micromanaging since here I give 2-3 objectives to an armygroup of 20+ divisions every week, compared to babysitting all those divisions every hour.

 

[Cybvep]

This makes me wonder. Would detaching HQs and giving them objectives some distance back from the front force the AI to do some actual retreating? It would be interesting to test that, especially at the lower levels.

IMO AI doesn't know the meaning of "retreat". I've never seen HOI AI that actually withdraws its forces from part of the front. It always tries to protect all its provinces.

But when it comes to AG-AI and objectives, I usually change objectives every week or two. I dont consider it micromanaging since here I give 2-3 objectives to an armygroup of 20+ divisions every week, compared to babysitting all those divisions every hour.

Yep, that's the point. Manually managing Barbarossa takes too much time.

 

[loki100]

IMO AI doesn't know the meaning of "retreat". I've never seen HOI AI that actually withdraws its forces from part of the front. It always tries to protect all its provinces.

This is part of the major problem with the AI on the defense. It doesn't use the logic - I have to defend Minsk, between where I am now and Minsk is a nice river line, I don't have to defend where I currently am, I'll retreat a small way and defend there. At best, a defense line behind the front might give it an idea what to do, once the current front is broken.

 

[CharonJr]

I should have added that I was using axis of advance for HGs Nord and Süd, Nord towards Pskov and Süd towards Odessa, but the units were still pretty spread out.

And strangely the only major AG without any axis of advance (HG Mitte) performed the most reckless penetration...

And yes, using the AI at theater level with the special units and 1 armored army under manual control works fairly well (this is the way I personally play most often) if you keep adjusting the objectives and dont set them to deep inside enemy territory.

But the actual point of those tests is to get rid of micromanagement as much as possible and see how the AI fares this way (badly as could be seen here ). Constantly having to adjust objectives is too close to micromanagement to my taste. But yes, the results would surely be better this way.

But as I had to see during this run even if I micro objectives the AI will refuse to follow orders if it feels it has to weaken its defenses too much by doing so. And once the Soviets have set up their defense due to a fairly slow moving front this is what will happen most of the time (the 99 divisions I added manually surely helped there, it might be a lot easier without those additional defenders).

Even if detached from the theater the AI is still worried about offering easy openings in its line and since it does not know the concept of luring some enemy units in and then cutting them off and crushing them this might actually be for the best.

Essentially this means you need a sizeable reserve which can be used to plug holes or exploit openings since your committed units will not be able/willing to do so (at least not in meaningfull numbers).

CharonJr

 

[Kirth Gersen]

That's disappointing, although I can't understand two things.

1. Why no axis of advance?
2. Why no objective micromanagement? The whole point of AG detached from theatre is AI that is actually following orders instead of following its own "Theatre logic". From my experience (and from Gladiator's thread, too) I can tell that it's harder with AG under theatre.


It's an interesting way of player-AI interaction - give AI infantry and only some armored units while the player handles airforce and breakthrough armies. Might work, considering that AI creates "stable" frontline.

+1 CharonJr, could you just confirm if your approach is truly hands off? Ie give the Army Groups an objective and then leave them to it. If so, I'm not surprised you're getting mixed results. The AGs will be acting independently in their own sectors and without a Theatre overview, the flanks will be vulnerable. For best results I reset objectives for AGs and Armies when necessary... as generals would do in real life.

Coincidentally, and this opens up a can of worms: there is a vehement critic of the HQ AI approach in the main forums (forget his name at the moment), who thinks that this approach is about setting objectives once, then sitting there armsfolded and seeing the outcome. "Its a wargame and it's all about warring", I think he said. And I agree with him if that extreme is taken. What would be the point? However, for me, life's too short for the other extreme of micromanaging every freaky deaky brigade. Which is why I think this game has such brilliant potential, as each person can pick the right level of granularity for them to manage. For me, it's Armies and AGs, but to be effective, you still have to actively manage them.

Sorry about the diversion - I just think this is an important point.

 

[Kirth Gersen]

+1 CharonJr, could you just confirm if your approach is truly hands off? Ie give the Army Groups an objective and then leave them to it. If so, I'm not surprised you're getting mixed results. The AGs will be acting independently in their own sectors and without a Theatre overview, the flanks will be vulnerable. For best results I reset objectives for AGs and Armies when necessary... as generals would do in real life.

Coincidentally, and this opens up a can of worms: there is a vehement critic of the HQ AI approach in the main forums (forget his name at the moment), who thinks that this approach is about setting objectives once, then sitting there armsfolded and seeing the outcome. "Its a wargame and it's all about warring", I think he said. And I agree with him if that extreme is taken. What would be the point? However, for me, life's too short for the other extreme of micromanaging every freaky deaky brigade. Which is why I think this game has such brilliant potential, as each person can pick the right level of granularity for them to manage. For me, it's Armies and AGs, but to be effective, you still have to actively manage them.

Sorry about the diversion - I just think this is an important point.

Answered, we wrote our posts at the same time!

 

[Modo]

And strangely the only major AG without any axis of advance (HG Mitte) performed the most reckless penetration...

Nothing strange about that. The AI found and exploited a weak spot. It should be better at it if not forced to go along one line, so perfectly reasonable in this setup.

At the army level, I've seen units facing Smolensk that were being redeployed south of Kiev to hit Moscow after a large opening was created near Kharkov. Adding an axis of advance in addition to the Moscow objective fixed that.

But the actual point of those tests is to get rid of micromanagement as much as possible and see how the AI fares this way (badly as could be seen here ).

The AI does as it is told. If you refuse to give new orders, why do you expect your generals to act on their own?

Constantly having to adjust objectives is too close to micromanagement to my taste. But yes, the results would surely be better this way.

A few clicks every two or more days is far from daily controlling a hundred or more divisions. Very far, even if you make it a little more when using armies during heavy fighting.

But as I had to see during this run even if I micro objectives the AI will refuse to follow orders if it feels it has to weaken its defenses too much by doing so.

There's also the resetting bug. Cancelling all objectives and giving new ones (even at the exact same locations) seems to do magical things in some situations. Even this is better than touching all divisions, but definately could use some improvement.

 

[CharonJr]

Acutally I did change objectives in between (regular: about once per month here, getting rid of taken objectives mainly, but sometimes setting new ones as well; emergency: whenever needed, but only when it gets really ugly and usually it is too late by this time ).

For "good" results going to extremes often works best in my experience and it quite nicely showed the shortcomings of the detached HQs from what I have seen (since I never tried it this way before it was actually quite interesting to see this).

Actually I am currently thinking about a run using 3 HGs under AI control still attached to the theater with 1 AG under manual control (including all the ARM, MAR and paras).

This would result in the following OOB:

AI

HG Nord - I. Armee (20 INF, 4 TAC, 2 INT), II. Armee (20 INF, 4 TAC, 2 INT)
HG Mitte - III.Armee (14 INF, 2 TAC, 4 INT), IV. Armee (12 INF, 2 TAC, 4 INT)
HG Süd - Kaukasusarmee (15 MTN, 4 TAC, 2 INT)

Manual

Spezialtruppen - I. Panzerarmee (6 LARM, 18 ARM), Fallschirmtruppen (10 para, 11 TRP), Marinetruppen (5 MAR, 10 TRP), Support: 12 TAC, 4 INT, Kriegsmarine, 10 NAV

While I believe that this one should work nicely I am wondering how much longer it would take than the earlier full micro run.

So what do you guys think would be more interesting for now, AI at army level (more objective management now) or a mix of 3 AI and 1 manual AG?

CharonJr

 

[Cybvep]

There's also the resetting bug. Cancelling all objectives and giving new ones (even at the exact same locations) seems to do magical things in some situations. Even this is better than touching all divisions, but definately could use some improvement.

I can confirm that. I do it fairly often, as it seems to really help the AI. Reloading the game every three months is also good.

So what do you guys think would be more interesting for now, AI at army level (more objective management now) or a mix of 3 AI and 1 manual AG?

I vote for THEATRE AI and one detached AG, manual. It sounds like a good setup for Barbarossa and we may be close to learning a nice way of playing the game on the large scale. I mean, what's the point if you just click Theatre-AI button and do nothing for the rest of war? It sounds like a setup for occupation of France, but it's not fun for war. Player-AI mix sounds too good to miss such an opportunity to test it.

 

[CharonJr]

Theater level AI sounds good, too. The bonus from the theater itself will not really be missed.

I forgot 12 Sicherungsdivisionen (2 MIL, 1 POL) in the OOB above, with 10 additional divisions under contruction which should help (actually they already have helped in the earlier runs) in freeing up combat divisions for frontline duty, even if the AI tends to use them in attacks, too.

CharonJr

 

[Cybvep]

Use Garrisons. AI likes to use them in their proper role - garrisoning vp provinces and ports.

 

[Kirth Gersen]

I vote for THEATRE AI and one detached AG, manual.

+1 - worth a shot.

Acutally I did change objectives in between (regular: about once per month here, getting rid of taken objectives mainly, but sometimes setting new ones as well; emergency: whenever needed, but only when it gets really ugly and usually it is too late by this time ).CharonJr

Hmm, monthly changes seem a bit too hands off. A lot can go wrong in a month!

 

[Modo]

Use Garrisons. AI likes to use them in their proper role - garrisoning vp provinces and ports.

Exactly.

I'd like to see a proper panzer army detached from the theatre. Don't care if the rest is in the theatre, army groups, or armies.

 

[CharonJr]

I'd like to see a proper panzer army detached from the theatre. Don't care if the rest is in the theatre, army groups, or armies.

Actually I. Panzerarmee is anything proper since it has way too many armor and no pure MOT/MEC divisions But basically this just means that I have to us armored divisions to keep open my lines of supply and I actually like the additional tactical flexibility this offers me.

I will try to run this one in three stages.

Stage 1:
I. Panzerarmee cuts of the Soviet forces in the south by rushing to Odessa and prevents the Soviets from reopening lines of supply while the marines and 1 armored corps will land in/near Leningrad to prevent/delay Soviet reinforcements from Scandinavia. The paras will take IC centers, provide "highways" for my armor (Red Market Garden) and fill gaps in the line as needed.

Stage 2:
Cut off the defenders in the marshes and head towards Moscow while trying to liberate the Baltic States.

Stage 3:
Crush the pocket in the marshes while parts of I. Panzerarmee take Stalingrad and rush for Baku followed by taking the remaining important VP.

During all stages I will try to take as many Soviet IC centers as possible.

Concerning garrisons, I dislike their speed, due to it I mostly build MIL/POL to take care about garrison duty, they take care about protecting my VPs and harbors fairly well from what I have seen.

CharonJr

 

[Modo]

Concerning garrisons, I dislike their speed, due to it I mostly build MIL/POL to take care about garrison duty, they take care about protecting my VPs and harbors fairly well from what I have seen.

Garrisons are not there to hunt partisans, they are supposed to sit on objectives. And due to their low speed, the AI often just redeploys them. I use armies with corps of 4 garrison + 1 mobile division, and give them only as many objectives as there are garrison divisions. This is enough to keep partisans away from important places. I never really worry about partisans capturing (and dying in) some remote forest/marsh/desert/whatever.

For Germany, I recommend 1 ARM + 1 SP ART + 1 ENG for automatic partisan hunting. Manpower cheapness with CA.