AI enhancement ideas

[Murat]

Well I am guessing HanBarca has the sort of post we should be getting towards. I am not a programmer so I do not know how AI works but other posts seem to indicate that this points system may be effective.  Each level of AI needs a different skill set too. Hard AI should have omniscience (except for chit picks and aplogies to whomever proposed this as I cannot find the post). I am moving today but I will begin working on some ideas for each nation.

[j-s]

ORIGINAL: Marshall Ellis

The single corps attacks (or just stupid attacks) seem to be happening quite a bit. Does anybody have a good test game where multiple corps could have been sent to battle BUT only one was actually sent?

In many game AI uses a good leader (Napoleon, Charles, ect) and a good corps (artillery, guard, cavalry) and attack against major stack. This is pointless and silly, becouse corps will be wiped out and leader captured every time.

AI should at least understand, that it don't need to destroy all russian cavalry by attacking with one cav corps one per turn against turkish feodal stack. And let few militia corps stand and watch [X(]

AI don't use stacks at all, at least not in my games that I have been playing. It just keeps all corps in one stack (usually in some wierd place) and defend if that stack is attacked.

[zenmaster]

#1 - The AI needs to be more aggressive in some wars or just not declare them.  Turkey frequently declares war on me as Russia, but I've never seen them send a single corp across the border.  I don't even bother putting a single factor down there because I know they are not coming.  Even if you don't intend to move to take the capitol, holdin a few provinces can hurt.
 
#2 - For AI vs AI wars, avoid selecting the "Remove Leader Option.  Focus more on Land/Money victory conditions.  I find that as a human player the computer opponents are easy to walk over because in a very short period of time all of their good leaders are long gone becore I get there.
 
#3 - Many other thoughts, but most are mentioned above.

[pzgndr]

Well I am guessing HanBarca has the sort of post we should be getting towards. I am not a programmer so I do not know how AI works but other posts seem to indicate that this points system may be effective.

Points or weighting factors would be good. Ideally some way of orienting the AI towards historical priorities would help. Here the USAFA version of EiA may be helpful for setting AI priorities.

Let the individual AI's have "personallity traits".

I recall the early draft of the game, at least the one demoed for me several years ago, did have AI personalities. This would be a nice game option to provide eventually, but I would recommend focusing on the plain vanilla AI for now. Once that's working OK, then it would be nice to see more passive or more aggressive variations.

[fvianello]

ORIGINAL: pzgndr
I recall the early draft of the game, at least the one demoed for me several years ago, did have AI personalities. This would be a nice game option to provide eventually, but I would recommend focusing on the plain vanilla AI for now. Once that's working OK, then it would be nice to see more passive or more aggressive variations.

I agree. I'd also add that "AI personalities" can be a dangerous business, as the usual definitions of "aggressive", "prudent" and similar isn't very useful in game terms. who wants a prudent france or a peaceful GB? or an aggressive austria that stay aggressive when napoleon is camping at vienna ?

[Adraeth]

AI grand strategy must focus on protect herself while attacking the enemy. So it will be useful to let the AI making a given% of units in her homecountry to counter invasions (Fra vs Uk - Aut Vs Fra or Tur), while the other units are on the march to attack the enemy.
 
If a Major State AI has multiple enemies the AI should have priorities targets (for istance Austia at war with Fra, Tur and Spain; Aut will act in offense only against Fra ... then if the war with Fra ends against Tur and finally against Spain; in the meantime the homeland will be safe (or trying to) thanks to the X% units in defense allocated in given Corps with a proper leader).

[iamspamus]

ME,

I have one where was Russia vs. Prussia. I was sitting in Lubeck or somewhere over there. I had 3 corps: a Swede, a Dane and a Russian. The Prussian had 8 (?) corps in Berlin (one space away). He attacked me with one corps at a time for 4 or 5 months in a row, as I build up more corps. Twice they were with a leader that I captured. I had no corps in the East (I think I was also fighting Turkey or something). He didn't march over there at all.

My computer is on the fritz, right now, so not sure when I can send it to you.

Jason

ORIGINAL: Marshall Ellis

Wow! I see that I have asked the right crowd, the right question LOL!
The single corps attacks (or just stupid attacks) seem to be happening quite a bit. Does anybody have a good test game where multiple corps could have been sent to battle BUT only one was actually sent?

Appreciate the help guys!

[pzgndr]

the usual definitions of "aggressive", "prudent" and similar isn't very useful in game terms

True. But with the historical priorities given by that USAFA version link, personalities could then be more historical (ie more aggressive at pursuing and defending those historical national goals) or less historical (ie more flexible and unpredictable). Whatever may be adopted, I would also think the personalties should be subject to change whenever a nation surrenders. That would be a time for reassessment and possible shift in behavior.

[Dragonheart]

It´s obvious that there is a plenty to do to make EIA also a nice Singleplayer game. I agree with most opionions.

Concerning the single corps moving i have to add that the AI is not supplying its corps at all. For example i saw so many times moving a russian corps all the way to France..maybe starting full......foraging all the way ....and arriving completly depleted with one factor in it.

When defending preussia they do the same with a couple of corps...moving without building up a suplly chain and foraging to dead.

[dulsin1]

One thing I keep seeing is that when France picks up the Sweedish and Danish fleets. The english should make a priority of clearing the sea of all French controled ships even if that means ungaurding a few ports.

The last game I defeated the english fleet in the channel and england kept moving more ships to block the channel but never enough to defeat the combined neutral fleets. If the english had moved two of the blockading forces to attack he would have decimated everything floating.

Instead the AI gave me 2-3 PP each turn while I whitled away the english fleet to a token force.

[eske]

Wouldn't it be possible to train a small neural net to pick combat chits?
- or two actually. one for attacker and one for defender.

Inputs could be no of inf and cav corps on each side, up to a max. Moral in steps of 0.25. Estimate of opponent moral same way. Leader strategic ratings. Die modifiers from comparing tactical ratings. Terrain table modifiers, for attacker/defender. An indication of possible cav. superiority, maybe presence of cav. leader. Chance of reinforcement, both sides. Possible guard commitment steps both sides.

I probabliy forgot something, but maybe this is allready too detailed.(?)

Test data could be generated randomly: A lot of battles with random chits and for a start just who won.

The wieghted output from the net should not be used for picking chit directly, but the chance for each chit, I guess.

- hmm, a little more complicating issues here than I imagined. But for starters simplify as much as possible.

Anyways, should be easy to obtain better results than totally random picks [:)]

[Grognot]

Probably a good idea to code up templates of what a stack 'should' look like, to avoid lots of vastly suboptimal stacks running around.

Ex. -- saw a Russian corps march from Moscow, across Prussia (access), towards Mecklenberg (at a time when the French army is heavily concentrated in what used to be Prussian or Austrian territory -- well within range).  It turned out to be an artillery corps w/ 10 factors.  Mistakes made in that march incl...

Russian artillery corps should never, ever, be operating without infantry corps support.  A/C and A/G stacks are too expensive.  If an infantry escort all dies due to foraging losses, that means the artillery corps needs to hie over to some accessible infantry stack.

No attention was paid to supplying it on the way.

Russian corps shouldn't operate solo on offense.  Other than Cossacks, they have no business going solo anywhere near an enemy major power -- one corps might suffice to take a minor, but other majors are going to be able to smash it pretty easily.   Only go solo for extreme tactics like 'recon by death' (which would require the AI to have some memory) and sacrificial rear-guard screens, or blocking a long and critical supply route (still best done with Cossacks, but *shrug*).

Should have something in mind like ex. Gds corps, A corps, several inf corps and pref. a couple cav corps under a decent leader like Kutusov -- for a primary army, not one for a defense of St. Petersburg unless it's really going that badly.  With an objective of linking up with Au/Pr forces if it's going to hit a large French stack... and if that stack's decimated, with Kutusov at risk of being captured, he should probably go back in the reinforcement pool.

[fvianello]

I completely agree. BTW, the problem of the "piecemeal" approach (never attack with everything but send little useless forces every turn) seems to infect many game AIs, not only EiA.

The root of this problem is always in the AI atomic decisional approach; it evaluates the "best move" for every single unit never considering them as part of something bigger.

Even in the acclaimed Total War, the battle AI is practically unable to execute the most basic attack maneuvers, for example a simple all-out frontal assault.

[adrianthomson]

You should find a way for the AI to actually strategize with other AI to conquer....

[Grognot]

...or defend.

I'm in four simultaneous 2-human/5-AI games, proceeding at a fairly decent clip (PBEM combat to avoid some glitches and reduce turn-around), and I think I've yet to see two allied AIs stack in the same region.  Much of that is probably because AI Austria and Prussia seem relatively indifferent to the other's being under threat.

AI Russia responds to calls from Prussia, and Turkey to calls from France, but the eastern majors don't maneuver very well.  e.g. seen a Turkish stack sitting in Greece, with a fleet, and staying there while Russians are besieging Constantinople.   Russia isn't quite as passive (the corps sometimes move) but they're uncoordinated.

[Grapeshot Bob]

Please allow the Spanish and French (or anyone else, for that matter) to combine their fleets and fight together.
 
I miss Trafalgar.
 
 
 
GSB

[Pans]

Hi together,

a good idea would be to modify the ally request info-box.

What happened was that I (Spain) asked the British for 30$. The roll was a eight and the Brits agreed for a loss of 3pp for me. I agreed and got 10$ ?!

If the money a MP lend another MP will not be over 10$ in total so please limit the input in the request field on that

cheers
Andreas

[Marshall Ellis]

ORIGINAL: HanBarca

I completely agree. BTW, the problem of the "piecemeal" approach (never attack with everything but send little useless forces every turn) seems to infect many game AIs, not only EiA.

The root of this problem is always in the AI atomic decisional approach; it evaluates the "best move" for every single unit never considering them as part of something bigger.

Even in the acclaimed Total War, the battle AI is practically unable to execute the most basic attack maneuvers, for example a simple all-out frontal assault.

Hanbarca:

My problem was not necesarrily the atomic approach BUT the "last stand" approach i.e. France has a stack heading for Berlin and I will bring everything I can in a desparate attempt to defend Berlin. If that means one corps then one corps it is. No denying that this needs to be changed.

[Norden_slith]

There are horribly many angles to a decent AI. One is better diplomacy, another is coordinating with allies, the next is defensive objectives, offensive objectives, supply, when to give up, when to take the peace-offer and so on.

One angle is fielding an effective force:

Generally most countries fielded one big fieldarmy. The possible exeptions are France and Russia.
This should be a guideline. It is of course very dependent on the total number of available corps and the average number of troops in those.

Each country should know its most effective General and try to fit him with the best units at least up to his corps-max. Best units as in some priority of quality (cav-guard-inf-mil) and quantity- depending on the country (Turkey might go for quantity). A certain % of cavalry should also be in the mix, this varies widely from country to country.

On top of that, certain corps should be assigned permanently to certain generals, like the Guard and the Artillery to Napoleon.

If there are more corps and generals, start to fill out the next general.

Most of this involves a lot of statistics, which is where computers excel.

Example:

Austria:
Prime leader: Charles, 6 corps or at least 2/3 of available. Cavalry is only between 10 and 15%. Allways attach him 2 a guardcorps or infantrycorps.

Secondary leader: Comes into play if at least 2 corps available. Preferable guard or inf-corps.

John: nononono!

1st priority: collect corps under charles
2nd priority: get the biggest corps
3rd priority: get as much Cav as possible
4th priority: attach Charles to a guardcorps.


Is this a usable approach? Somehow, I imagine, that if you focus your forces around a few Generals, principal tactics are a little easier to develop. If Prussia has Blücher and Hohenlohe you have to find out what 2 units are doing, not all those separate corps.

[Marshall Ellis]

I've seen the "main army" idea a few times and it seems that this could be a good approach. This could be done with a stack buildup like you're talking about (force mix) and leader and goals (defensive or offensive). This could also help minimize the suicide attacks by preventing individual corps actions. I like it...
 
Tell me what would be the optimal force composition in EiANW?