AI Behavior- Tactical Level

[IronWarrior]

In a Detailed Battle that I felt I should have lost, the AI seemed to not want to win. AI started off aggressive and hit my supply wagons, but after that it seemed to dance back and forth and never dealt the final blows that would have sent me off the field.

This was on Blucher difficulty level, not sure if that would make a difference.

(Starting to see why Hard Sarge wins every battle in his AAR's) just kidding HS [:D]

I'm not a big fan of AI's and prefer mp anyway, but just thought I'd mention it.

EDIT: The AI does seem to work well in naval battles, this was land combat.

[IronWarrior]

Wanted to add that the AI does the same thing on 'Bonaparte' difficulty level. It will aggressively go after the supply wagons, but then spends the rest of the battle moving units up to one hex away from mine, but then withdrawing again without firing or charging.

AI makes some decent moves but doesn't seem to want to capitalize on them. Cavalry will come all the way up to my artillery and instead of charging it, it withdraws again. Infantry moves up and doesn't fire, instead withdraws again-moves up again. Rinse and repeat.

[Erik Rutins]

Hm, haven't seen that regularly, could be something specific to your battle? Could you e-mail me a save file please to erikr@matrixgames.com?

[IronWarrior]

Hi Erik,

email sent, not sure if it's the files you need but i sent a couple that I saved (iirc) during those battles.

Bill

[Mus]

I have seen the same behavior of Cavalry charging towards my guns and then moving back.  I thought it was trying to draw out a reaction fire so it could freely move up another unit, but Im not sure if the AI is that complicated in its actions. And then like you I noticed they were never actually driving home, even if I didnt fire.

[ericbabe]

Erik has forwarded your save file to me.  Thank you for it. 

I have seen some cases where the AI plays a little too defensively.  The AI doesn't like to move into local situations in which (locally...) it is outclassed, and sometimes this makes it a little too timid when it has an overall advantage.  I'm experimenting with tying the local timidity factor to parameters relating to the overall battlefield situation, and I'll be able to use your save files to test these changes.

The AI has a set of several different cavalry tactics among which it semi-randomly chooses one at the start of the battle, and there is a small chance it can change cavalry doctrines during the battle.  In certain cavalry modes, cavalry doesn't like to charge formed units unless they are in certain situations, and in these modes cavalry will sometimes approach a line, and then move around a bit waiting for the right situation to charge.  Sometimes this mode works very well, as the cavalry then save their charges for when they matter most, though it can be less than optimal if the supporting infantry aren't close enough behind.  I have considered adding a heuristic to keep AI away from the fronts of formed artillery units.

In my own recent testing, I've found that cavalry can become a little too predictable based on the current tactic they are using (though perhaps this is only because I know exactly what the tactics are and how they work).  I'm considering giving cavalry a small chance to break with its tactic so that the AI will be a little harder to predict.

[Hard Sarge]

well, one thing the AI is doing, is moving in, trying to draw reaction phases, most times it gets in close like that, it don't have the MP to finish the charge off, if it sits still, it is going to get clobbered, if it moves away, with out drawing reaction, it was a wasted move

overall, I think the AI does well, as long as it has options, once it gets down to do or die, it is not sure what to do (and the player, normally does his best to take as many options away as he can)

[IronWarrior]

Eric, no problem- glad to help out if I can. I double-checked and the second file was saved at the beginning of a naval battle and not what I thought it was. When I had played the first battle, the AI hit my supply wagons, but I was still able to win with absolutely no supply for the entire battle due to the units going back and forth.

Sarge, good points although I'd like to see at least the infantry be more aggressive... or at least stop moving around as much and use mp's to fire. So far it seems like AI infantry only fires when pinned. I've also seen cavalry sit in front of my artillery for numerous turns taking canister fire.

[Ironclad]

In many battle accounts there are descriptions of cavalry and infantry standing in formation for hours suffering the attention of enemy artillery, often having moved into a forward position. This could be for a variety of reasons - command incompetence/bloodymindedness, disciplinary, miscalculation, distraction, sheer necessity, justifiable risk and/or the prelude to a battle changing event. So perhaps the AI is acting in a more realistic manner than we think.

[Mus]

ORIGINAL: ericbabe

The AI has a set of several different cavalry tactics among which it semi-randomly chooses one at the start of the battle, and there is a small chance it can change cavalry doctrines during the battle.  In certain cavalry modes, cavalry doesn't like to charge formed units unless they are in certain situations, and in these modes cavalry will sometimes approach a line, and then move around a bit waiting for the right situation to charge.  Sometimes this mode works very well, as the cavalry then save their charges for when they matter most, though it can be less than optimal if the supporting infantry aren't close enough behind.  I have considered adding a heuristic to keep AI away from the fronts of formed artillery units.

Eric, I was wondering if my infantry supporting the gun was in fog of war so the AI couldnt see it and charged down to take the artillery in flank but then "saw" that it was supported and backed off, but Im not sure how the AI and fog of war works. Also not sure if the AI remembers the gun was supported or not so it moves back up the road, infantry is in fog of war again, AI cant "see" its supported, so goes back down the road next turn to try to charge it flank only to discover its supported etc etc.

[ericbabe]

The AI uses FOW rules to a certain degree, so that's a possibility in this case.  It's kind of hard to explain how some of the AI behavior works since it's based on a lot of number crunching and probabilities.  The AI does have kill-zone detection routines that are supposed to prevent it from moving into or sitting in murderous locations.  However, these things are somewhat hard to balance.  For instance, the two biggest complaints in this thread so far is (1) the AI isn't moving forward to attack as much as it should and (2) the AI is moving foward too close to artillery.  I've been considering adding a special check just for enemy artillery, and perhaps changing the game rules to make it harder to move artillery forward into point-blank contact with the enemy.

[IronWarrior]

Ok it looks like I was a bit hasty, and should have qualified my previous posts as first impressions. I stayed up late playing the game, and had a really fun, massive battle between Napoleon and the main Austrian army. The AI was quite masterful (as much as can be expected from an AI), and the infantry moved up and fired, cavalry charged when it had clear advantages/flank/rear attacks.

So don't listen to me, listen to Hard Sarge and others who actually know what they're talking about. [:D]

Also (and maybe this belongs in the wish list), I like that the AI uses various different tactics, but perhaps instead of it being random the AI could chose tactics based off of the commanding officer's tactical rating.

[saintsup]

I'm a bit disapointed with the tactical AI. I think specifically that it use Cav very badly.
 
1) Cav units are very often going in front of my inf/art units doing nothing. Then I fire doing hudge damages and the Cav retires.
2) Cav units are often wasted before infantery comes

[Anthropoid]

I agree that the AI does not seem to use Cav as effectively as it could. I've played about 6 or 7 different games now, for a total of maybe 20 tactical battles ranging from Normal to Diff settings, all of them with me as France 1792 start, so I've only played battles up to about 1793. Almost all of my playing has been with Brigade Level. Here are my observations:
 
1) AI uses Cav to form screens that are mostly pointless. A screen is useful if it is used either as a ruse or as a way to hide a mass of troops. There are almost never well-organized masses of troops behind the AIs cav screens.
 
2) AI uses Cav to 'probe' enemy line, with _FAR_ too much attention to the center of the enemy mass. Here I'm mirroring many of the preceding comments about Cav moving in to close to try to cause reaction or squaring up, but then either not succeeding or getting in too close to Arty or high firepower infantry and getting mangled. If the same tactics were used on the flanks, rear-flanks or REAR of the human's formations, they might work a lot better, but this leads to a third point.
 
3) AI 'tries to flank me' but never actually uses enough brigades to actually threaten it. One infantry and one cav that are positioned 5 or 7 hexes to the left of my left flank are just not enough to 'flank' me in most instances. Meanwhile, I later see that the AI has multiple 'reserve' units that get clumped together in a feeble mass right in front of my main facing where I've set up a terrible kill zone.
 
4) As long as I don't shoot at him, AI lets me surround him very readily. This seems different from FoF where the AI seemed to back off if I started moving units around to flank and encircle him even if I wasn't pounding the units from the center front facing.
 
5) AI does not get the most out of his arty. Here I think the key issue is that the AI seems to act as if the arty is his silver bullet, and will send it forward even when there are a lack of lined up infantry to support an arty brigade as by gaurding his flanks.
 
Not knowing how the decision-algorithms work it is difficult for me to make sound suggestions but just in the interest of promoting brainstorming I'll give it a stab:
 
1) AI uses Cav to form screens that are mostly pointless & 2) AI uses Cav to 'probe' enemy line, with _FAR_ too much attention to the center of the enemy mass.
 
Perhaps a solution here would be to make the AI favor different Cav tactics based on how many brigades he has and what their morale is? For example
 
A) if I (the AI) have fewer than Sum=14 Cav morale points, and Sum=3 Cav brigades, there is really not much point in doing anything with Cav other than treating them like infantry other than perhaps doing a couple of screens.
B) if I have an intermediate level of MP and brigades (maybe 4 to 7 brigades and 15 to 21 MP) then sending three brigades to try to flank the enemy while, another one to 4 engage in screens seems reasonable. The best brigades, the ones most able to change formations, should be the ones who try to flank, and if all possible they should skirt a wide margin around the enemy to try to avoid being seen. If B routine is 'decided on,' then the main force should wait till the right time and engage the front of the main enemy facing, i.e., one or two turns before the Cav are in position to do a flank charge.
 
3) AI 'tries to flank me' but never actually uses enough brigades to actually threaten it & 4) As long as I don't shoot at him, AI lets me surround him very readily.
 
These two are very difficult to reconcile because (3) calls for the AI spreading out his troops more and sending them into red zones more but then (4) calls for him keeping them lined up closely more and out of red zones more. About the only thing I can suggest here is that the AI's probability to move in for a frontal assault or try a flanking should DEFINITELY depend on him having numerical superiority if not parity. There is no point in a force of 22K Brits trying to outflank a defending force of 28K Frenchmen who have reinforcements on the way! Last two battles against Brits, this is exactly what he tried, and I wound up capturing most of his Divisions by slowly patiently moving my 65K reinforcements to encircle him, but not firing at him, thus not causing his WTF to drop, disordered count to rise, and propsensity to retreat/escape to climb.
 
5) AI does not get the most out of his arty. Here I think a better 'understanding' of terrain, and of how to form up a mass of infantry and arty into lines would help, though I have no idea how you get an AI to have that!
 
Hope it helps. It is one of the best games ever, and I have no doubt in coming patches you guys will make it even better [:)]

[Randomizer]

My experiance so far has been somewhat different, I have found the naval AI indifferent but the land AI has presented some challenges.  In one battle as Russia I was defending with 57,000 men against a Polish army of around 65,000 and the AI put up an excellent fight.  I found some good ground and deployed defensively anticipating the FoF tactic of allowing the overly aggressive and left flanking AI to break itself on my line but it didn't.  A Polish infantry/cavalry force demonstrated to my front and suddenly two Polish divisions appeared square on my right flank.  This was at last light and the onset of night allowed me to disengage and find new ground, had it happend two or three turns sooner it might have been decisive.  Next day was a bloody slug-fest that turned into a race of cripples that I won only by a narrow margin.  It was one of the best AI fights I have ever had in any computer game (actually cannot think of a better one off hand).
 
This is really a great game.
 
Best Regards

[IronWarrior]

All in all I find the AI too easy playing on Bonaparte/Bonapate... but that doesn't really mean anything, I find the AI too easy in any game. I still dream of seeing the detailed battles be able to be imported/exported to LAN in mp, and the results impoted/exported back to pbem. A lot to ask as far as coding goes I'm sure, but one can dream right?

[Mus]

ORIGINAL: IronWarrior

All in all I find the AI too easy playing on Bonaparte/Bonapate... but that doesn't really mean anything, I find the AI too easy in any game. I still dream of seeing the detailed battles be able to be imported/exported to LAN in mp, and the results impoted/exported back to pbem. A lot to ask as far as coding goes I'm sure, but one can dream right?

That would be awesome. If WCS does something centered around the detailed battles I hope they put this feature in. The ability to play out detailed battles from imported data from a PBEM.

Which battles would be fought detailed battle could be handled by house rules. Say only engagements involving 100,000 total men, or whatever the group wants.

[Anthropoid]

ORIGINAL: Randomizer

My experiance so far has been somewhat different, I have found the naval AI indifferent but the land AI has presented some challenges.  In one battle as Russia I was defending with 57,000 men against a Polish army of around 65,000 and the AI put up an excellent fight.  I found some good ground and deployed defensively anticipating the FoF tactic of allowing the overly aggressive and left flanking AI to break itself on my line but it didn't.  A Polish infantry/cavalry force demonstrated to my front and suddenly two Polish divisions appeared square on my right flank.  This was at last light and the onset of night allowed me to disengage and find new ground, had it happend two or three turns sooner it might have been decisive.  Next day was a bloody slug-fest that turned into a race of cripples that I won only by a narrow margin.  It was one of the best AI fights I have ever had in any computer game (actually cannot think of a better one off hand).

This is really a great game.

Best Regards

I recently started up a new game on Difficult settings. I had a few battles, but then earlier today, the AI started to show its true abilities. I'm having a similar experience to the one you recount of just barely holding on by my fingertips. It will be interesting to see if I can get my reinforcements deployed quick enough to save the day against the AIs nasty artillery brigades.

[ubik]

ORIGINAL: IronWarrior

In a Detailed Battle that I felt I should have lost, the AI seemed to not want to win. AI started off aggressive and hit my supply wagons, but after that it seemed to dance back and forth and never dealt the final blows that would have sent me off the field.

This was on Blucher difficulty level, not sure if that would make a difference.

I can confirm this on normal level. The effect is most noticeable with cavalry as often the results of a commitment would be more dramatic for me, but it seems infantry also suffers the same problem.

Basically I think under some situations the AI should risk more as the loong term tactical benefit is bigger than the drawbacks. In one instance, I had a very vulnerable infantry plus two artilleries within 2 hexes and the AI did not charge with its cavalry.

On other instances, I don't know if it is coded to trick the player into some kind of trap, because that happened to me the first time I went after it.

Anyway I think it is doing more harm than good overall.

[ericbabe]

I have made some changes to AI aggressiveness in the patch we're working on (based on overall field conditions), so hopefully this will help the AI act more aggressively when it should be dealing the "final blows."