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

[QuintusSertorius]

Well, the diplomacy is far superior than RTW: thanks to the "appreciation" scale (dunno the actual ingame name XD) factions can be allied for virtually forever even if they share borders...
I remember in a campaing in a M2TW mod, i was playing HRE and even though my standings with the poles and hungarians were awful since the beginning, they became my best allies. The magyars attacked me, and after I crushed their invading army as I was approaching Buda, I offered peace, alliance and (this time) military access, they accepted and never attacked me again.
Also seems that factions are reluctant to attack if your "fame" is far superior than theirs...

That sounds promising. Like perhaps they've fixed that particular issue.

About the army composition issues; check this out.

I'm not entirely sure that addresses it; the thread talks about free-upkeep units as garrisons and control over what can be recruited where (possibly). Within what's available we could still see the AI spamming useless armies.

"You will grow tired blunting your weapons on a poorly-led horde of mindless corpse-men; and once you have reduced them to so much sausage filler, the sweet taste of success will turn to ashes in your mouth" ™


The AI in TW games hasn't been good since shogun first came out, imo at least. Though I expect the single-player campaign in EB2 to be something amazing, I will try to get most of my fun out of multiplayer battles (there should be a big enough of a community to support that on a regular basis).

Hmmm, to be honest multiplayer doesn't appeal. I'm only in this for the single player campaign, battles alone don't really do anything for me. They need that context of the campaign (and strategic maneuvering) to make them interesting.

We can take it as a fact that the AI in M2TW is better than in RTW. It's been a while since I last played it, so I can't give you examples off the top of my head, but I've played it enough to be certain of it. Also, an important point is that there are more possibilities to deal with the shortcomings of AI. One example is the feature to limit recruitment of certain units, which will do a great deal to prevent pure elite armies, but perhaps the most important is that the AI itself is actually moddable. Never done it myself, so I have no idea to what extent, though.

Again, that gives me a little hope, even if no one is really sure yet it works, that it can be modded opens some possibilities.

Edit: I'm aware that the elite army is not the most pressing matter, but there some factions are quite prone to that (played some years in a Romani campaign over the weekend and fighting elite african pikemen and other assorted elite infantry becomes tedious; same goes for the notorious grey death's argyraspides stacks).

I don't play on Very Hard campaign difficulty, so perhaps that's why the army of skirmishers is a more frequent occurence and thus a concern to me than the army of elites.

[Ludens]

I know how you feel: the predictability of R:TW's A.I. can make it seem like a chore. The A.I. of M2:TW has definitely been improved, although it was not a priority for the developers. From my brief experience with M2:TW I got the impression that the strategic A.I. is now competent. It can develop its cities and field credible stacks, something R:TW always struggled at. Partly this is due to better game-design (recruiting units does not deplete cities and prevent upgrading), but also because the A.I. is grouping and deploying its units better. The same thing applies to diplomacy: the new negotiation screen gives more feedback, allowing you some grip on what's going on; but the A.I. is also more reasonable. Neither A.I. is likely to outplay all but complete newbies, and at higher difficulty levels you simply get a load of elite stacks flung at you, but it doesn't make as many glaringly stupid moves.

The one thing I am less sure about is the tactical A.I. Again, it has been improved and I didn't see as many stupid moves, but I also didn't get much of a challenge. It's nowhere near M1:TW.

[Paltmull]

I'm not entirely sure that addresses it; the thread talks about free-upkeep units as garrisons and control over what can be recruited where (possibly). Within what's available we could still see the AI spamming useless armies.

I'm not sure what the EB team intends to do, but a possibility would be forming the unit pools so that they are proportinate to an army. For example, if regular line infantry units had high avaliability the AI would be likely to recruit and use those in large numbers, rather than units with low availiability. This way, you could control what kind of army composition that the AI (as well as the player) would use.

[bobbin]

Campaign AI is a lot better for M2TW and we are tweaking it too. Not really sure about battle AI (i don't play battles enough now to form an opinion).

[B-Wing]

As has been said, both campaign AI and battle AI have improved, though not by as much as you might hope. I've been playing Stainless Steel (a popular M2TW mod) lately and have found its AI, which was imported from other mods, to be greatly improved over the vanilla game. I would expect the EB2 team would likewise try to implement these improvements as much as possible.

[Ichon]

Campaign AI is improved in vanilla and many mods take that a good step further. The main issues are AI going bankrupt and requiring huge money scripts to be competitive and AI tends to leave many cities on coasts with a single garrison unit making surprise invasions very easy to do which some mods address by garrison spawn scripts. Also the sea invasion AI is not very good and the AI tends to waste many of its naval ships in strange places- IE, depending on the mod AI often deploys its entire navy in a small area of water unreachable to anywhere else like Red Sea, Persian Gulf etc.

Battle AI is a little better than RTW but still exhibits many of the same flaws. There are some good attempts to make it better and to a degree they are successful but a veteran human player should still be able to defeat the AI on 1 to 1 odds every single time. The biggest problems with BAI is that AI does not get the idea to use infantry and cavalry together and charges ahead with all its fast units when on offensive or holds them in place and doesn't move when on defensive so missiles or charge to the rear easily wipe out their army.

Since most generals in MTW2 and mods are mounted that means within 1-3 minutes of battle start the AI commander is dead and another 1-3 minutes battle is over unless its a siege or there is a 2nd reinforcing army. EB2 having more commanders as infantry should help this issue as well they could try different things such as give AI generals high defense but low charge and melee so the AI doesn't charge so quick and when it makes mistakes it doesn't always kill its commander right in the start of the battle.

[Arjos]

Also with the allied forces' "command", many things can be done in battles: as deploying HA or cavalry forces, while you move the infantry and viceversa, or having the reinforcements not standing still while in offense...

[B-Wing]

Also the sea invasion AI is not very good and the AI tends to waste many of its naval ships in strange places- IE, depending on the mod AI often deploys its entire navy in a small area of water unreachable to anywhere else like Red Sea, Persian Gulf etc.

I'm hopeful that this particular point won't be an issue in EB2. I recall that EB1 placed some serious restrictions on where ships could be produced, and I suspect that they can and will make it so that regions with access to the Persian Gulf and Gulf of Oman will at least not be able to produce anything more advanced than the most basic ships, which will hopefully save the AI some money.

Since most generals in MTW2 and mods are mounted that means within 1-3 minutes of battle start the AI commander is dead and another 1-3 minutes battle is over unless its a siege or there is a 2nd reinforcing army.

Though I've heard it many times before, I've actually not seen this happen often at all. My most frequent experience is that generals linger around behind the main lines for most of the battle, occasionally charging my flanks, only to retreat when I send someone to after them. And then sometime after half their men have died, they charge into the center of the already engaged line and eventually get killed. I've rarely seen the kamikaze behavior so often brought up. Maybe I've just been lucky, but I did want to mention it.

[amritochates]

Well its great enough that EB(RTW+BI+Alex) has been permanentely uninstalled- as the old saw goes "There's no going back"

I won't comment on vanilla which is a disaster, but the mod that I am playing DLV 6.2 (http://www.twcenter.net/forums/forumdisplay.php?f=309) is amongst the best.

The Tactical AI is competent, no more single units of pike-men doing solo attacks on your entire army thank goodness!

Though some do comment on the AI in MTW-VI being superior, IMHO the battle AI is the same, but the battle mechanics were better in MTW-VI.

The real culprit here is the animations issue, where the unit animations play an equally imp role as its stats, which leads to units with slower stats to severely under perform. Additionally clumping is still an issue. Also units with dual weapons aka RTW under perform severely.

The Strategic AI is miles ahead, if played on H and not VH.

For a better look at that wonders may be done with the AI have a look at this:
http://www.twcenter.net/forums/forumdisplay.php?f=637

[Ichon]

I'm hopeful that this particular point won't be an issue in EB2. I recall that EB1 placed some serious restrictions on where ships could be produced, and I suspect that they can and will make it so that regions with access to the Persian Gulf and Gulf of Oman will at least not be able to produce anything more advanced than the most basic ships, which will hopefully save the AI some money.


Though I've heard it many times before, I've actually not seen this happen often at all. My most frequent experience is that generals linger around behind the main lines for most of the battle, occasionally charging my flanks, only to retreat when I send someone to after them. And then sometime after half their men have died, they charge into the center of the already engaged line and eventually get killed. I've rarely seen the kamikaze behavior so often brought up. Maybe I've just been lucky, but I did want to mention it.

I think the solution to the ships is likely very restricted ports but if pirates still spawn there has to be some ability to make ships somewhere. Relatively low replacement rate should be more important than low cost to keep the AI from spamming ships because it can be a huge issue. I've screen shots from 2 different mods that included ports in Red Sea where the AI faction there had built over 30 ships that were filling the sea so much they could barely move. That AI was going bankrupt just from its fleets.

I've experienced the AI general hiding behind its frontlines ONLY when its on the defensive and lacks cavalry. Usually playing on mods where AI has money script it is rare it is on the defensive as the human player is always undermanned and fighting usually 2 AI armies on H or VH difficulty levels. If the AI has even 1 or 2 other cavalry units it usually charges with them and then the general right behind. Plus- when the general hides behind the line it often sits there and ignores HA or other missile units concentrating fire on it and does not move away until all the BG nearly dead.

Honestly the most difficult AI to deal with is usually the super aggressive one which charges everything right at your lines because your army will take some losses no matter what wereas if the AI hesitated even for a few minutes it would give human player ample time to setup flanking maneuvers etc and accomplish nearly bloodless victories.

My hope is that with EB2's stronger focus on infantry melee and less on powerful charges by heavy cavalry there is more tactical importance to lengthy maneuvers then simply eliminating enemy cavalry and charging from the rear whilst using infantry as a distraction meatshield. The numbers in the units might make a large difference as well... most mods follow the vanilla formula of cavalry being half the strength of infantry unit which means that it would be 2,000 infantry vs 1,000 cavalry. Not that such battle were impossible but generally infantry greatly outnumbered cavalry- even horse culture factions which fielded almost all cavalry did so in numbers much smaller than an infantry faction could field. Of course infantry not being as mobile those extra numbers weren't as important but they still mean something. If infantry units are usually 150 men I hope most cavalry is 50 or less. So cavalry used carefully is still decisive but requires some care in use and HA might actually run out of arrows before killing the entire enemy army and have to engage in some melee or have infantry hidden on the field somewhere to mop up the survivors.

The real culprit here is the animations issue, where the unit animations play an equally imp role as its stats, which leads to units with slower stats to severely under perform. Additionally clumping is still an issue. Also units with dual weapons aka RTW under perform severely.

The Strategic AI is miles ahead, if played on H and not VH.

I agree animations are as or more important than stats but is that really part of AI? AI is where units move and how they react to your own moves.

I've heard a few people make this claim about H vs VH but I've never seen anyone offer proof why H makes better strategy choices than VH?

[Andy1984]

I'm not sure of what you meant Paltmull, but it gave me an idea anyway. Suppose you have an AI-city which is controlled by a faction whose army should consist of an equal amount of phalanxes and cavalry. You might script it that the AI in each of her settelements can either recruit a phalanx-unit or a cavalry-unit, both with a priori chances 0,5. Thus: in half of the settlements, the AI recruitment options are 'cavalry', and in the other half 'phalanx'. All other recruitment options are grayed out (maybe by an invisible event?). Next turn, the recruitment options are recalculated. Some cities get phalanxes, others get cavalry. The AI might still opt not to train one kind or another, but the problem would manifest herself less.

[Paltmull]

Andy: M2TW uses a recruitment system with unit pools of various sizes, along with a certain level of replenishment for each pool. Pool size and replenishment rate is individual for each unit type.

My idea was that you would use larger recruitment pools with higher replenishment rates for the units that you would prefer the AI to recruit many of and smaller pools with lower replenishment rates for those that you want the AI to recruit fewer of.

If an ideal army for example has lots of standard infantry, some cavalry, a couple of skirmishers and very few elites, you could, through unit pool sizes and replenisment rates, arrange the availiability of those unit types so that the AI's recruitment - and therefore also its army composition - follows that model.


EDIT: I have this bad habit of using awfully long sentences. I hope the above is readable

[Ichon]

That works well in other mods and sounds like what EB2 will at least make some use of. I just hope EB2 also takes account of mercenary recruit pools in the same way as AI tends to recruit all mercenaries available if it has the money which it probably will with a money script.