How does the AI in M2:TW compare to R:TW?

[Ichon]

Looks promising, this can bring up some exciting battles!

If I see this I can imagine him using that cavalry instead to battle my cavalry, or trying to flank me.

~Fluvius

XAI and some other working along similar principles have been succesful in getting the cavalry to flank and even avoid stakes that some archers in MTW2 can place but so far none have been able to get cavalry to function as screen vs other cavalry or stop the AI from throwing all its cavalry ahead of its infantry without support if it has an advantage in cavalry numbers. The AI will hold cavalry in reserve if your own army has equal or greater amounts of cavalry but if not... throws away its cavalry and usually its general as well then its an easy battle to flank and finish off the remaining infantry. The only time battles are really tough is when AI has way more cavalry than your own army and even if it loses 1/2 of it stupidly the other half does some damage.

The problem seems to be to make the AI cautious enough to hold back its cavalry in the face of having superior numbers also means it sits there when attacked at range by missiles and does not react very well. I think some versions will chase the unit targeting them with missiles but then that is almost as bad as it is quite easy to draw out single unit or even one I tried that all cavalry within a certain radius would converge on the missile unit firing on any cavalry so it was a simple matter to line up horse archers or whatever on a hill and run 1 HA down to fire on the AI cavalry and then a whole string from one flank of the AI army chased that single cavalry back and forth on the bottom of the hill until all were dead.

Making sure AI armies are relatively well composed and that using replenish rates that players have a problem creating 100% heavy cavalry armies is key to more often good interesting battles than just improving BAI alone. If the AI percieves the forces as equal is when it seems to act most intelligently under all the BAIs and when paired with one of the better BAIs that has been developed can lead at least to some interesting battles on VH where AI gets a slight morale advantage that can sometimes make a difference. Usually it still fails somewhere but to exploit that failing takes more work and thought by the player.

[QuintusSertorius]

Been playing some more EB1 and getting annoyed with the army composition and battle AI again. I just fought a Makedonian full stack comprised of twelve units of classical hoplites, a single levy phalanx at a tenth of it's normal strength, a unit of cavalry, a unit of akontistai, a general (who charged the centre of my line; I had to make an effort not to kill him) and the rest useless Illyrian levies. Who of course came forward piecemeal, not even as a single line. Seriously, how hard is it to at least stick to a single line?

[Tuuvi]

Been playing some more EB1 and getting annoyed with the army composition and battle AI again. I just fought a Makedonian full stack comprised of twelve units of classical hoplites, a single levy phalanx at a tenth of it's normal strength, a unit of cavalry, a unit of akontistai, a general (who charged the centre of my line; I had to make an effort not to kill him) and the rest useless Illyrian levies. Who of course came forward piecemeal, not even as a single line. Seriously, how hard is it to at least stick to a single line?

Yea that's one of my pet peeves to, the trick is to play as Hayasdan and be at war with AS early on in the game, you'll be glad the AI is so dumb.

[QuintusSertorius]

Yea that's one of my pet peeves to, the trick is to play as Hayasdan and be at war with AS early on in the game, you'll be glad the AI is so dumb.

There's really nothing much fun for me winning because the AI is incapable of the most basic tactic of holding a line.

[QuintusSertorius]

Does M2:TW use the naval AI from BI? Meaning that if you happen to have a coastal boundary, you'll be spammed by insignificant but annoying stacks from factions you don't even have a meaningful relationship with.

[antisocialmunky]

Does M2:TW use the naval AI from BI? Meaning that if you happen to have a coastal boundary, you'll be spammed by insignificant but annoying stacks from factions you don't even have a meaningful relationship with.

I think this was actually a big problem in Vanilla M2TW.

[vollorix]

Yea, it´s annoying when a powerfull faction like Carthage tries to "raid" Latium with a "3 unit stack", though i suppose it has some reasons i might have observed in my campaigns:
- the AI seems to be designed to send an "adequate" force to fight you at a designated place, means: if you have only couple of troops in that province, the AI will send you a small raiding party instead of a whole army
- the same one can often observe when AI tries to take Eleutheroi towns - it sends a half way decent army while trying to take the chances on the even odds; but it mostly sends captains against 5-10 star gouverneurs of that towns, and we all know the results.. But in the same time there are huge stacks guarding homelands, led by capable generals. One should take this into account too, imo.
- that funny AI behaiviour when you see all those reinforcements lining up along the roads toward the final target, and the often observed siege with a pathetic force ( a captain, a quoter stack, too light units - all that against a multiple star commander of yours with a decent army; the AI sallys mostly on the favorable odds, so the attacker AI seems to think - hey, why not besiege the town earlier, build some siege equipment, while the reinfocements are on the way? I really would like to know what is responsible for the choices of the AI to let a FM lead the army instead of a captain! This would make things not only more realistic, but also more challenging, i think ( for example: the AI never recruits generals on BI.exe, it´s a feature only the player might use, afaik. If one could change that, or script a general for every decent - 1/3 stack at least, or so - army, this would spice up things too ).

Such raids from less powerfull, and less rich, factions seem more then apropriate to me. The full scale warfare is something that didn´t happen every year, unless a huge conflict between some local superpoweres was goning on - like the First Punic war, for example.
It could be a matter of diplomatic presets, how much one faction is suspicious about another, but when choosing the campain difficulty, one, sadly, has to choose between a passive, poor ( no mercs etc. ) AI and the super agressive, insane factions with the top priority to take the player out. That´s why i mostly play on "hard" as a bit more balanced setting. I hope, the settings for the AI on MTW2 engine are more sofisticated.

[QuintusSertorius]

What's the latest on this? It was some time ago that I originally asked the question, has AI been addressed now? Have any other mods borne fruit in ternms of getting campaign AI able to recruit and mobilise properly, and battle AI capable of holding a line?

[Moros]

There are a lot of mods that have custom AI adn there are many AI mods out there. For example Lusted on is quite well known. Currently we have a few people working on it. And as you can read on our twitter we've tweaked a mongol faction AI from Lusted to fit to our Saka faction. So AI especially in second or later release will probably even be customized for each faction/fation type/culture. In other words it should be a huge improvement over RTW1 and MTWII:k.

[moonburn]

from eb1 empire posting it seems the koinon hellenon are very naval invasion prone so now having acess to the sea is no longer just a way to increase revenues trough merching or an highway to invade other lands but also a weak point that must be adressed guess that the river fort strategy to secure your core lands won´t be enough anymore without a decent navy to protect your shores so the strategic ai as improved

lets see how the diplomacy ai and how the tactical ai compare to real players

[Kull]

What's the latest on this? It was some time ago that I originally asked the question, has AI been addressed now? Have any other mods borne fruit in ternms of getting campaign AI able to recruit and mobilise properly, and battle AI capable of holding a line?

One thing you have to keep in mind is that only very recently did the team finally slay the bugs that made it impossible to play for more than a few turns without some sort of CTD. And you really cannot test a campaign AI without, well, a campaign! So from that standpoint we are definitely playing catch-up, but it is a key focus area, and we have made progress.

[RAWROMNOM]

Right now it's all to hard to say. Other than we have a lot of studying, questions to ask fellow modders (I'm sure CA as well), "Follow AI" for 100000000's+ of turns, internal discussion, and work, before we're done with regards to the CAI. We're going to aim high that's for sure. But for this instance of the mod, it's likely, "done when it's done."

The some problems ahead of us are because of the fact that EB2 is a total conversion; concepts like religion are gone, and other unrelated concepts have been introduced. How do we get the AI to work in concert with those new concepts? Specialty scripts? Gentle nudges/hard pushes in already established AI files? Etc? Some things are hardcoded and out of reach, some things are more softcoded than they were in RTW.

Anyways, the best way to help get that "Dream AI" in as fast as possible is to point knowledgeable people in the teams' direction, or take some days to learn about it and help us out yourself if you have the time and motivation.

[Rex Somnorum]

Concerning the naval AI in Vanilla: playing as Egypt on VH/VH, the Byzantines would frequently land small stacks near Alexandria. The armies were usually slightly larger than my garrison and composed of more professional units. However, due to the BAI's incompetence, I always defeated them, but they also mauled my force in the process. Whenever the landing force was weaker than the garrison, they wouldn't attack at all but evacuated in the same turn. In my England campaigns, the Portuguese frequently landed huge, powerful stacks in Wales.

In other words, naval invasions are possible and are generally well-calibrated. In my experience, the bigger problem is the tactical AI.

Edit - For extra challenge, it's easy to integrate the Hardcore ReallybadAI mod into any build. It essentially introduces a command-chain, ensures the AI uses stakes and limits the deployment time.

[fallen851]

In other words it should be a huge improvement over RTW1 and MTWII:k.

This sounds like... hype from R2TW.

If you're lying like they were, I'll kill you myself. But if your not... then EB2 will really be something special. For a long time I couldn't think it could possibly be better than EB1... but this...

[RAWROMNOM]

This sounds like... hype from R2TW.

If you're lying like they were, I'll kill you myself. But if your not... then EB2 will really be something special. For a long time I couldn't think it could possibly be better than EB1... but this...

It's not hype. As Moros mentioned there are other custom AI's out there that surpass vanilla. If others can do it, we'll learn how to do it as well; And being a labor of love like EB2 is for us, we're not going to settle until we're proud of every part of it and the fans are happy too.

Anyways, I wouldn't threaten Moros. I once saw him turn a camel into a balloon before flying off into the sunset... I still hear the screams at night.

[Moros]

Been playing some more EB1 and getting annoyed with the army composition and battle AI again. I just fought a Makedonian full stack comprised of twelve units of classical hoplites, a single levy phalanx at a tenth of it's normal strength, a unit of cavalry, a unit of akontistai, a general (who charged the centre of my line; I had to make an effort not to kill him) and the rest useless Illyrian levies. Who of course came forward piecemeal, not even as a single line. Seriously, how hard is it to at least stick to a single line?

Hopefully the recruitment pools and the ability to add recruitment priority to units will fix this somewhat.

This sounds like... hype from R2TW.

If you're lying like they were, I'll kill you myself. But if your not... then EB2 will really be something special. For a long time I couldn't think it could possibly be better than EB1... but this...

Doh't expect to have a product of the same quality as the final EB 1, with the first release. Don't forget it's still an Alpha/Beta release. Comparable with the 0.7x versions.

[fallen851]

It's not hype. As Moros mentioned there are other custom AI's out there that surpass vanilla. If others can do it, we'll learn how to do it as well; And being a labor of love like EB2 is for us, we're not going to settle until we're proud of every part of it and the fans are happy too.

Anyways, I wouldn't threaten Moros. I once saw him turn a camel into a balloon before flying off into the sunset... I still hear the screams at night.

I am well known for my empty threats. But don't tell Moros that.

Anyway, it is good to hear that the EB2 team plans on committing to finished EB2, but I've heard that before. In many ways, I felt EB1 was left in an unfinished state. To install the game, you have to download and then install EB 1.1, then patch it to EB 1.2, then install a bunch of hotfixes. Why was it never all put together into a single download and install? Also, there have been a lot of improvements to RTW mods since the release of EB 1.2 in graphics quality (RS2 terrain), and game stability (the no quotes fix) that have greatly improved RTW mods. EB hasn't been updated with these either. Finally, I think most of the EB community will agree that certain mini-mods should be rolled into the main game, such TWFanatic's phalanx mod (among others). Phalanx units fight in a much more realistic and historic fashion with that mod, and with one of EB's goal seeming to be historical accuracy, I feel it should be added and can't conceive of an argument why it shouldn't be added.

Hopefully, EB2 will be finished and polished.

Doh't expect to have a product of the same quality as the final EB 1, with the first release. Don't forget it's still an Alpha/Beta release. Comparable with the 0.7x versions.

Don't worry, I have realistic expectations regarding the quality of the first releases. I was around for the earliest EB releases. But if there isn't an AI improvement in the final EB2 releases compared to EB1, I'll be pretty disappointed.

[Cybvep]

In many ways, I felt EB1 was left in an unfinished state.

Can't say I agree. Sure, numerous improvements could have been made, because there is always room for improvement, but EB1 v.1.2 was quite polished IMO. Also, the team's manpower is limited. There is a point when they have to say "that's enough, the mod is finished" and move on. And there are always unofficial submods for those who want to enhance the game further. RTW1 is a very old game by now, while MTW2's engine gives the team a chance to do things which were impossible to do back in the RTW1 days.

[moonburn]

i still defend that eb2 should be released in mini mods style and then in the end when everything works fine all would come together to make a trully amazing single big game

eb steppes warfare \o
eb the barbarian world \p
eb the diadochii (eastern greeks) \o/
eb land of the hellenes (the greek world and the politiks of the city states) d/
eb the mediterranean world \~/
eb the ties of kart hadast /T\

the lst 3 overlap i´m aware but mini campaigns could be trully amazing and tweak specific war zones like the desert or the steppes and tweak political organisations like the carthies the roman/italians and the greek city states or work better in the development of the barbarian civilizations and how the barbarian diplomacy worked

[RAWROMNOM]

That's something that would be highly impossible to coordinate. We're simply a bunch of guys and gals donating our freetime. Work, family, friends, and other responsibilities come first before EB2 development. We're not a professional development team with a budget. This means that everything comes together at different times. Like for instance, right now some of our historians/programmers we had before are busy with real life issues. Thus the areas they have been working on are basically stalled.

[Rex Somnorum]

Thus the areas they have been working on are basically stalled.

What areas in particular?

[RAWROMNOM]

What areas in particular?

Koinon Hellenon Unicorns, Seleukeia Opium cut to 80% with hashish, and Sweboz hookers (Back then there was no ID; so the more hair, the better).

[RAWROMNOM]

Actually, that's a lie. EB2 is finished and we've been playing it for years. Eb3 is already halfway finished, built on Rome 2 Total War.

[fallen851]

Actually, that's a lie. EB2 is finished and we've been playing it for years. Eb3 is already halfway finished, built on Rome 2 Total War.

The Warscape Engine makes me shudder.

[QuintusSertorius]

A more general question; in RTW the military and diplomacy AIs didn't talk to each other. A faction massing troops on your border for invasion could happily sign trade rights with you, then attack the next turn. Another faction you were pasting in battle after battle wouldn't accept a ceasefire (without Force Diplomacy) even though they'd been greatly weakened. And so on.

Is there any evidence the two actually interact in M2TW? Does the AI respect a ceasefire and alter what it's doing, or throw whatever it has at you as soon as it can build up an army again?

There are a lot of mods that have custom AI adn there are many AI mods out there. For example Lusted on is quite well known. Currently we have a few people working on it. And as you can read on our twitter we've tweaked a mongol faction AI from Lusted to fit to our Saka faction. So AI especially in second or later release will probably even be customized for each faction/fation type/culture. In other words it should be a huge improvement over RTW1 and MTWII:k.

That's reassuring that we don't have the exact same AI as we had in RTW (as before, I have very little trust in CA applying themselves to deeper, more complicated things than spiffier graphics).

One thing you have to keep in mind is that only very recently did the team finally slay the bugs that made it impossible to play for more than a few turns without some sort of CTD. And you really cannot test a campaign AI without, well, a campaign! So from that standpoint we are definitely playing catch-up, but it is a key focus area, and we have made progress.

Well, I was hoping more from a general modding perspective that people might have been playing with the AI and have experience that could be applied to EB2. Rather than expecting the developers to have necessarily achieved it themselves.

Concerning the naval AI in Vanilla: playing as Egypt on VH/VH, the Byzantines would frequently land small stacks near Alexandria. The armies were usually slightly larger than my garrison and composed of more professional units. However, due to the BAI's incompetence, I always defeated them, but they also mauled my force in the process. Whenever the landing force was weaker than the garrison, they wouldn't attack at all but evacuated in the same turn. In my England campaigns, the Portuguese frequently landed huge, powerful stacks in Wales.

In other words, naval invasions are possible and are generally well-calibrated. In my experience, the bigger problem is the tactical AI.

Edit - For extra challenge, it's easy to integrate the Hardcore ReallybadAI mod into any build. It essentially introduces a command-chain, ensures the AI uses stakes and limits the deployment time.

That is also reassuring. Is it possible to effectively stop naval invasions if you strongly garrison coastal settlements?

As a side-question, were the factions landing troops those you were actually at war with, or had something to gain by invading? The problem with the naval AI from RTW:BI was that anyone on the same body of water was a potential aggressor, regardless of diplomacy status (because of course, military and diplomacy AI didn't talk to each other in RTW...).

Hopefully the recruitment pools and the ability to add recruitment priority to units will fix this somewhat.

Fingers crossed.

[Kull]

Well, I was hoping more from a general modding perspective that people might have been playing with the AI and have experience that could be applied to EB2. Rather than expecting the developers to have necessarily achieved it themselves.

"People" have been playing with the AI, just not people on the EB2 team. Do a search for "campaign AI" or go browse discussions and tutorials and you'll see that many folks have spent lot of time customizing the AI either for specific mods or as "bolt-on" improvements to the original vanilla M2TW AI. What we are doing - now that we can actually TEST the AI in EB2 - is looking at the entire corpus of research, identifying existing improvements made by others, testing those, and making alterations accordingly. The EB2 AI is ALREADY customized based on that initial testing and research, but we are looking to do even more as we go forward.

That said, i would encourage people to have realistic expectations. The AI will never be a human player, and will always feature limitations of some sort that simply can't be overcome. What I can promise you is that the AI will be better than it was in M2TW, and probably quite a bit better than it was in EB1.

[Rex Somnorum]

and Sweboz hookers (Back then there was no ID; so the more hair, the better).

I'm useless at modelling, but I could certainly script in some modified princesses, if you need help.

Is there any evidence the two actually interact in M2TW? Does the AI respect a ceasefire and alter what it's doing, or throw whatever it has at you as soon as it can build up an army again?

I spent half an hour typing out an in-depth answer to your questions and suddenly my connection timed out. Once more unto the breech.

I'll be concise: yes. Keep in mind, though, that a faction with bad relations with you will probably violate the ceasefire after re-building their forces, unless another faction distracts them. But that's only natural.

That is also reassuring. Is it possible to effectively stop naval invasions if you strongly garrison coastal settlements?

As a side-question, were the factions landing troops those you were actually at war with, or had something to gain by invading? The problem with the naval AI from RTW:BI was that anyone on the same body of water was a potential aggressor, regardless of diplomacy status (because of course, military and diplomacy AI didn't talk to each other in RTW...).

To the first question, yes but it's expensive. The second question is a bit more complicated. Yes, they both had quite a lot to gain, especially since my armies in the theatre were generally weak. However, the decision to invade is balanced against the target faction's standings (global standing, or reputation, and faction standing, or relations) and risk. The Byzantines viewed Egypt - an ally of their nemesis, the Turks - as a weak and obvious target. Portugal is essentially sandwiched between two continental powers - the Moors and Castille - and take to the sea seeking Lebensraum. So, even though the English and Portuguese might have lukewarm relations, the rewards of a successful invasion far outweigh the risk. In some cases, the AI seems to behave irrationally when it is operating on a logical game mechanic. For example, Poland may start a war with Russia by blockading Russian ports simply because their navy is too weak to defend their trade-lanes, the lifeblood of the Rus' treasury. Basically, neutrality means nothing, standing means everything and weakness will be exploited. A close ally will likely remain a close ally. But neutral factions are wild cards.

To be fair, I've developed a theory base on my trawls through the descr_campaign_ai_db.xml file that the game has hardcoded "enemies" - or factions that will always remain rivals, which would explain a lot of the AI's behaviour.

On an endnote, the AI can be vastly improved by altering a few values in the game files. There's one line in the config_ai_battle.xml file that's massively powerful:

<friendly-to-enemy-strength-ratio>0.8</friendly-to-enemy-strength-ratio>,

this measures the balance of forces to determine whether an army should attack another army, on the campaign or battle map. In Vanilla, an army needs to be only 80% the size of an enemy's force to attack. By increasing the value to, say, 1.5 the AI will only attack with a 50% numerical advantage, which would eliminate most of the small-fry offensives.

By the way, why doesn't the medieval period interest you?

[RAWROMNOM]

The Warscape Engine makes me shudder.

If you're a fan of Amnesia: The Dark Decent and other horror games (Silent Hill 1, etc.), EB3 will be right up your alley!
We're also taking design cues from this fighting game:

I'm useless at modelling, but I could certainly script in some modified princesses, if you need help.

If you're clever enough, it just might make it in as an Easter egg. Who knows? (No points if you steal the dwarf marriages image from Third Age!)

[Kull]

On an endnote, the AI can be vastly improved by altering a few values in the game files. There's one line in the config_ai_battle.xml file that's massively powerful:

<friendly-to-enemy-strength-ratio>0.8</friendly-to-enemy-strength-ratio>,

this measures the balance of forces to determine whether an army should attack another army, on the campaign or battle map. In Vanilla, an army needs to be only 80% the size of an enemy's force to attack. By increasing the value to, say, 1.5 the AI will only attack with a 50% numerical advantage, which would eliminate most of the small-fry offensives.

If you want to destroy both the Campaign AND Battle AI, use that setting. The King of M2TW AI feels differently:

You can do that, and the AI would build impresive armies that would probably do nothing, and would definately do nothing in battle. Anythong much over 1:1 tends to cause unreasonable passivity in battle. Also note Lusted's post, even a 1.2 caused expansion dificulties for the AI on the campaign level.

Like I said, we are conducting phd level research on this topic. Trust us.

[Ibrahim]

If you want to destroy both the Campaign AND Battle AI, use that setting. The King of M2TW AI feels differently:

impressive what a single number will do. it's clear improving the AI to an ideal or near idea level will be a challenge, if even possible.