The AI thread

[frogbeastegg]

I have a few questions for those of you who have the game.

1. How does battlefield difficulty affect the AI? Does it again access to extra strategies a la STW and MTW, or does it merely gain a fat stat bonus a la RTW and M2TW?

2. I've seen some brief comments about unit size affecting the AI's performance. A few people have said it plays a far better game with medium or small units, and tends to engage in milling about and other bad behaviour when set to large. Can anyone expand on this?

3. How is the campaign map AI in the grand campaign? Is it sane? Does diplomacy work?

Any further comments on the AI would be welcome; it's the most important aspect of the game for me. No awed gibbering or disgruntled ranting please; let's keep this one semi-analytical.

Any comments from the developers would be welcome.

[Negative]

I'd like to add on to this.

I've been playing on medium. When I attack a fort the AI places half its army outside the fort and makes some very questionable decisions as to where it stations its men. They're just spread out all over the place. It doesn't even use the walls or the height advantage for placing cannons. Is it smarter/better on high or very high?

[pevergreen]

1. It is back to MTW/STW.
2. Smaller units are always easier to control for the AI.
3. Diplomacy works better than we could have hoped for. There is reason for everything they do. They will try to steal and kill with agents, they will retreat to a better position for their armies. This game is just awesome.

[Shadow_Wolf33]

I have a few questions for those of you who have the game.

1. How does battlefield difficulty affect the AI? Does it again access to extra strategies a la STW and MTW, or does it merely gain a fat stat bonus a la RTW and M2TW?

2. I've seen some brief comments about unit size affecting the AI's performance. A few people have said it plays a far better game with medium or small units, and tends to engage in milling about and other bad behaviour when set to large. Can anyone expand on this?

3. How is the campaign map AI in the grand campaign? Is it sane? Does diplomacy work?

Any further comments on the AI would be welcome; it's the most important aspect of the game for me. No awed gibbering or disgruntled ranting please; let's keep this one semi-analytical.

Any comments from the developers would be welcome.

1. The battlefield AI seems fairly improved to me, my very first large land battle was against the huron and I was expecting it to be a cakewalk. Instead of the usual "wait for me to get to their town and seige it", they charged out of their camp and met me head on in the field which was a very surprising change to say the least. Once in the battle, they also more or less took me apart. They had an insane amount of bowmen (which out ranged my current musket troops), and more or less peppered my troops a decent bit before charging my center and breaking it. This was fairly early in the game, so I didn't have a whole ton of military tech unlocked (the turn after this I had ring bayonets researched ). That first battle REALLY took me by surprise (and this is on EASY no less), and I just had to re-load and try it again....and well...let the AI auto-calc it cause they did WAY better than me.

I led the assault on the actual "town" myself though, which went far better as I paid a lot more attention to my tactics and those of my enemy, and more or less goaded them into charging my musket lines using my cannons.


2. I've seen nothing that would indicate anything like that happening in my game. Frankly if they perform better with medium/small units, I'd be afraid to play at that unit size.

3. Campaign map AI that I've seen so far seems relatively sane and much lower on the stupid scale compared to previous Total War games. They have a new tactic though, instead of stabbing you in the back, they badger you to trade territories a LOT. Literally the turn RIGHT AFTER I conquered the Huron capitol, the french came along and wanted to trade me Newfoundland for it. The turn after that, they wanted to trade me Newfoundland AND an island down by northern Brazil. I know people in Newfoundland talk a bit funny and can be hard to understand, but are the french THAT desperate to get rid of it? A few turns later I managed to acquire those 2 provinces for a nice cash settlement, so it turned out nicely.

One thing I did notice though, is that the minor factions seem to have a taste for making VERY stupid decisions on who to go to war with. This isn't necessarily a BAD bad thing though, as it can make territorial expansion easier. I'll probably take advantage of it in the long run.

[Monk]

I have a few questions for those of you who have the game.

1. How does battlefield difficulty affect the AI? Does it again access to extra strategies a la STW and MTW, or does it merely gain a fat stat bonus a la RTW and M2TW?

2. I've seen some brief comments about unit size affecting the AI's performance. A few people have said it plays a far better game with medium or small units, and tends to engage in milling about and other bad behaviour when set to large. Can anyone expand on this?

3. How is the campaign map AI in the grand campaign? Is it sane? Does diplomacy work?

Any further comments on the AI would be welcome; it's the most important aspect of the game for me. No awed gibbering or disgruntled ranting please; let's keep this one semi-analytical.

Any comments from the developers would be welcome.

1. I wouldn't go as far as saying "it's back to MTW", but its improved upon over m2. Though i hate to report that it is still prone to moves of idiocy, which makes those moments even harder to bear in the previous series since you know it can do better!

2. No idea. I haven't done anything below Large.

3. Campaign AI is amazing. It's light-years ahead of the tactical AI and makes sense. Making friends is hard, but when you make a friend they are normally yours for a long time. Diplomacy makes sense, which is the most important thing ETW needed (aside from a good AI)

My only nit-pick is the State Gifts. While I understand them and know they are vital for showing your good will, it seems hugely exploitable. If you have a fat treasure you could easily wash away a couple decades of brutal occupation with a few china sets and a Thoroughbred Horse, something about that feels.. odd.

[Eusebius86]

I spent an hour testing N/H/VH, Custom Battle, GB, one line inf vs one line infantry. I'm at work and don't have the results on me, but the AI DEFINATELY stat stacks on VH. 80vs80 I was left with 15 troops while the AI had 45, and he had 3 columns, opposed to my two, with only the front lines firing. Definitaly stat stacking... I'll post more detailed and definate results when I get back from work.

[frogbeastegg]

Thanks for those answers.

Eusebius86, I'd definitely be interested in seeing those detailed results. I can't stand stat stacking - I grudgingly tolerated it in STW and MTW because the boost was minimal and there were the extra tactics to compensate. In RTW it was ridiculous.

I suppose it is too early for people to have an idea of what tactics are enabled at the higher levels.

My only nit-pick is the State Gifts. While I understand them and know they are vital for showing your good will, it seems hugely exploitable. If you have a fat treasure you could easily wash away a couple decades of brutal occupation with a few china sets and a Thoroughbred Horse, something about that feels.. odd.

"Here's a horse to make up for my killing your father. This china tea set is a special comemorative edition in memory of your mother and auntie Bertha, whom I accidentally poisoned while trying to kill your grandfather. Speaking of him, this deluxe hat with feathered plume is the latest fashion in my captial and it's to say sorry for splattering him with a cannonball. Yes, sorry about that. Now, you might be thinking 'What about my 15,000 brave soldiers who died in this war?' and you'd be right. It doesn't do to forget the courageous dead. That's why I started work on this delightful series of collectable figurines ..."

[PanzerJaeger]

I'm not sure where you guys are coming from with the AI.

I haven't been able to play the campaign at all, but I played quite a few custom battles with the AI set at very hard and it was absolutely atrocious.

It sent units at me piecemeal quite a bit, and when it did finally bring a sizable force against my main line, it repositioned it self constantly, which did not allow for firing and reloading, which meant that they were fish in the proverbial barrel. This one issue seriously made the game completely unenjoyable for me.

Also, I saw no flanking or anything that could even come close to being described as tactics. It couldn't even organize itself for a basic frontal attack. It consistently left its artillery unprotected, too.

If this is the way it acts on the campaign, which I have no reason to believe it will be any different, I will be very disappointed.

On the upside, MP does have potential - which has always been my main focus.

[Eusebius86]

I'll post the result's in about 7 hours, when I get off of work. Interesting stuff.

[Liberator]

I'm not sure where you guys are coming from with the AI.

I haven't been able to play the campaign at all, but I played quite a few custom battles with the AI set at very hard and it was absolutely atrocious.

It sent units at me piecemeal quite a bit, and when it did finally bring a sizable force against my main line, it repositioned it self constantly, which did not allow for firing and reloading, which meant that they were fish in the proverbial barrel. This one issue seriously made the game completely unenjoyable for me.

Also, I saw no flanking or anything that could even come close to being described as tactics. It couldn't even organize itself for a basic frontal attack. It consistently left its artillery unprotected, too.

If this is the way it acts on the campaign, which I have no reason to believe it will be any different, I will be very disappointed.

On the upside, MP does have potential - which has always been my main focus.

I totally agree. What enrages me a lot is that the AI is attacking, even if it should be defending
+ that while attacking, they always send infantry units after my cavalry units and follow them all over the map instead of attacking with all units at the same time at the same point (which is the only way for the AI to win, if it does not outnumber me significantly [So far, I've never lost ])

[GM1940]

Liberator, Panzer: What unit size? Might it be underperforming at large/huge?

I haven't done much beyond very unequal battles and testbeds to confirm stat imbalances at "Expert" in custom battle (which seems to be different from the usual E/M/H/VH nomenclature for TW battle difficulty.)

[Liberator]

I'm playing battles on H, unit size large

[Buckwallis]

Don't have answers to all the questions but here's a few observations:

I think unit size is definately an issue. My first 3 battles I fought with Ultra and it was just disgusting. Hard to place units, pathfinding was ridiculous, and the AI was just bound up.

I've dropped that to large and it's improved. The depth and potential of the game... and some of the campaign features... and of course, the sheer beauty of the game... have me slowly warming up.

The big thing is that the pathfinding is really bad. Especially around fences and such. Things get really screwy around fences and that just blows up the AI. The two issues are linked IMHO. Fix the problem with pathfinding and fences and I think the AI won't seem as funky...

[GM1940]

You know, I like the series; I like the devs, I'm confident the game'll turn out great. But I am gobsmacked that things which the community old hands notice in a day of gameplay completely escaped all of the professional game reviewers in the world. Does it not occur to these guys to enlist some TW experts in checking out things like AI, pathfinding, etc?

That said, I can't even go read in the TWcenter forums because of all of the "CA is violating my human rights my shipping the worst game evar" threads.

[ChaosLord]

Hmm, sounds like the AI may have actually been improved. But I think i'll wait a week or two for more results. I thought hard about just pre-ordering this one too, but i've barely spent enough time with M2TW to justify that purchase. Hearing that the campaign AI improved is great, but how is it as far as combining and composing their armies? Is it random mass recruitment or does the AI try to build balanced armies?

[pevergreen]

For 1 region factions, when under pressure it seems to mass the best units it can recruit. The Crimean Khanate have lost about 30 units of Cossacks to me.

[A Very Super Market]

I haven't been able to tell whether they are starting armies or not, but every single faction I've fought has a very balanced amount of units.

[Strategy]

Hmmm... diplomacy makes sense? Playing a campaign as Sweden on VH/VH, I declared war on Courland, causing Poland to join the war. After annexing Courland and taking a few Polish provinces, I offered it peace (it had no army worth speaking of). It refused. I continued taking provinces. Down to only two provinces, the AI finally accepted my demand of peace and all of its tech advantages as the price.

In its turn, it immediately declared war on me again. Two turns later, Poland no longer existed. Not a smart decision.

The diplomacy for exchanging provinces is also completely nonsensical. Prussia was extremely eager to trade Polish provinces for East Prussia. Just for fun, I checked whether it would trade West Prussia (badly damaged from battle and 1700 tax base) for East Prussia (major fortifications and 5000 tax base). Absolutely.

The AI is generally weird. It can show flashes of sense and then the next moment do some utterly nonsensical stuff (e.g., the above diplomacy examples, or letting its units stand and be blithely shot to pieces.

[Eusebius86]

Ok frogbeastegg, here's the results of my tests.

Tests were done with Medium Unit sizes, Great Britain, Line Infantry, no veterancy. All difficulties tested 3 times and averaged. In all tests my units were in 2 files, AI lined up units in 3. In all tests only the front row fired.

- Normal: I had 45 men remaining, AI had 15 men remaining. Pretty accurate if you consider the disadvantage of having 3 files instead of 2.
- Hard: I had 31 remaining, compared to the AI's 13. Once again, I won due to being in 2 files instead of 3. I should have tested at 3 files, but didn't...
- VH: I had 15 men remaining before routing, the AI had 40. VERY lopsided. AI fired very fast, and very accurately.

Interesting notes. Small difference between N and H. HUGE difference between H and VH. Also, AI behaved identically in all 3 difficulty settings. Exchanged rounds until my units dwindled down to about 45, and then charged. Very interesting.

This says nothing as to the AI's tactical abilities. Just that the AI does stat stack as far as reloading times, defense, and accuracy is considered (at a minimum). I'm sure the same applied to naval AI...

Campaign AI will take time to figure out.

[frogbeastegg]

For 1 region factions, when under pressure it seems to mass the best units it can recruit. The Crimean Khanate have lost about 30 units of Cossacks to me.

At least it's recruiting elite units for once, even if it's obviously cheating. Remember the good old days of the AI spending its bonus cash on peasant and ballista armies :winces:

- Normal: I had 45 men remaining, AI had 15 men remaining. Pretty accurate if you consider the disadvantage of having 3 files instead of 2.
- Hard: I had 31 remaining, compared to the AI's 13. Once again, I won due to being in 2 files instead of 3. I should have tested at 3 files, but didn't...
- VH: I had 15 men remaining before routing, the AI had 40. VERY lopsided. AI fired very fast, and very accurately.

From that it looks like hard difficulty is the way to go unless you don't mind watching the ridiculous. Ugh.

It might be worth going a bit deeper into this one. If it's tested out with pure melee units as well, and with different deployment for the human unit, it might be possible to pin down the bonuses.

Hmm. Doesn't research affect battlefield performance? Is it possible the AI doesn't have stat bonuses but instead has an extra level of research which makes the soldiers reload quicker? The first step to answering that is to find out if there is a research item which states it gives that bonus. The few I know about give boosted melee skills or new formations.

I think unit size is definately an issue. My first 3 battles I fought with Ultra and it was just disgusting. Hard to place units, pathfinding was ridiculous, and the AI was just bound up. I've dropped that to large and it's improved.

That looks like the next area for testing. Any volunteers?

If a balanced large army was chosen and used by each side, and the battle fought on the same map with the same set up with regards to who attacks and defends, the casualties and performance could be compared. Play it several times on large units and several more times on medium. If replays can be saved from custom battles then deeper analysis might be possible.

[Sheogorath]

The combat AI is definitely improved, however, the UNIT AI seems to have a few issues.

To name a few...

My musketeers are VERY picky about facing for some reason. If the enemy is just a few degrees off from being perfectly centered, the entire formation has to rotate and reform, which, obviously, when faced with charging cavalry is a bad thing.

Likewise, sometimes their little firing arc doesn't follow their orientation. This leads to some embarrassing moments, such as (apparently) infantry facing gleefully in 90 degrees to the left while hordes of vicious Indians bear down on them. When told to fire on a specific enemy in this 'mode' they run off in a random direction, apparently forever. Seems like some AI confusion there.

Soldiers STILL seem to prefer grapples to gates...this is an issue going back to RTW. Cant CA fix this already? It's kind of silly for my own men to run over to the grapples on the wall when the gate to myfort is standing OPEN directly behind them.

Furthermore, the AI seems a little woozy on the whole fort thing...on wooden forts it apparently has some sort of compulsive desire to scale the walls ONLY on the side with the gate.

On stone forts it has its men run to every side BUT the one facing the way it came from (most of the time).

Fort cannons seem a bit useless, IMO. They make a lot of noise and big explosions, but only kill one guy at a time in most cases and their range is too short to get off more than a couple shots unless the AI is kind enough to stall in a position where it can be shot. For cannons should, I think, have the same range as regular cannons. Their low rate of fire makes them a lot less godlike than the terrifying automatic doom towers of MTW2, which fired about 100 cannonballs per second.
(It is, however, totally awesome that we have working defenses on the walls. I completely love that.)

Also, I notice that the terrain around forts changes every battle. It was a little disconcerting when my first fort went from a wooded hillside to a cleared field.

Another minor glitch, it's possible to put cavalry (or generals, at least) on the walls if you stretch them out on/next to a ramp. It looks fairly silly

[Haxorsist]

I'm playing H/H and the AI behaved weirdly in one battle. I deployed in a very long line, but I forgot I had a unit of pikemen which was placed near the middle of the line when I ended deployment. The AI outnumbered me, but what it decided to was concentrate its attention on my unit of pikemen. Fair enough, but they decided to send most of its units into melee against my pike wall which they were obviously unable to break. My units quickly lined up in a square around the enemy and they were killed off pretty quickly from the fire from all directions.

[Bob the Insane]

I have been trying out some test battles in the custom land battles...

In all cases I used the Britsih for both side, unit size large.

First for 1v1 I had a unit of line infantry. I would just let the enemy appraoch and start the fight. They would fire unit they lost around a third of their unit, attach bayent and charge. I would pause as soon as i saw the enemy attach bayents and hit the command for my own men and them once the action started again have the charge.

On Normal victory was 50-50, it could go either way.

On Hard the AI always won, but only just with large casulties on both sides.

On Expert the AI would slaughter my men in both the firefight and the melee every time, 2 to 1 casulties usually.

I repeated the above with melee only units and the result was the same.

This would imply significant stat bonus on Expert and a slight Bonus on Hard...

I then try battles. On the Anatolia map, Early period, Medium funds, Dry, etc...

Again British v British using the present armies with one alteration, I removed the artillery on both sides. Once the battles started I left my troops in their default formation, I grouped them and the only movement was to rotate my line to match the AI's once they had formed up but before the fight started.

Normal, the AI approached in reasonable formation, formed up outside gun range in line with cavalry on the winging (identical to my formation) moved into range. Once the shooting started it moved two units of light infantry off the line, moving one to each wing where they deployed stakes and then moved to attack (firearms) the ends of my lines. Meanwhile it moved it's cavalry to the ends of my lines and then charged them in where they met my cavalry. For the duration of the fight the AI did a good job of maintaining the line and the battle could really have gone either way. It appears to move it's general near to the battles but never actually into battle.

Hard, the approach was identical to the above until the shooting started. The it got a little odd, the AI concentrated it's units in the center (overlapping a bit). The cavalry did the same as above. The infantry does a lot of reforming to face individual units in my line making the AI's line rather ragged.

Expert, interestingly it's tactics and movement was idential to the Normal battles. A little sharper maybe. It's control of it general was sharper too, keeping him back until the fight had started. The battle itself was a little onesided as the 1v1 tests above would imply. It lasted a fair time and some AI units did run away but I simply took way more casualties than the AI and my cavalry was over run on both wings.

Obviously there are loads more varaibles on the maps and i didn;t use artillery which is a huge part of battles in thie era. But so far if you looking for a "fair" battlefield I would say pick Normal. Hard is good for battles where the play must do something to win, but I have some suspicious about the AI's tactics on Hard. If Normal and Expert moved almost identically on the field, why would the AI's movement on Hard be a little odd?

[quadalpha]

Frog: as far as I know, only experience affects reload rate.

[zarkis]

In my opinion the campaign AI is a step backward from MTW2. Strangely small nations and the natives in America are sending big stacks at you, while big nations tend to follow the peacemeal approach. I played Maratha on VH/VH and crushed the great Moghul empire without serious resistance. The only tense time I had when Mysore stabbed me in the back. The AI often counterattackes your forces with insufficent strength, so you can win easily by just using autoresolve. Also it never evades battles it can't win. And so far I haven't seen any sea invasions.

Here are some pics taken by other players, that tell you more about the AI then thousand words..

Spoiler Alert, click show to read: 

Spoiler Alert, click show to read: 

Spoiler Alert, click show to read: