How to improve the tactical AI

[Foz]

Didz and Cheetah: I actually assume the difference is in what version you are playing, as opposed to play style. Cheetah describes the state of the game in 1.0 or 1.1, where Didz describes the game w/ a shield fix applied, or leaked 1.2 (which has the official shield fix). Flanking is much more important in 1.2 because the combat DOES slow down a lot when shields work properly, and spears do not simply get butchered in a few seconds. I actually had some very poor spear units who just held a city gate for maybe a minute or two against a charging mongol horde that had broken it down. The mongol force was lightyears better than the troops they were encountering, but it still took some time for them to get through them. Had it been a field battle, I'd have had plenty of time to flank the opposing force. It never would've gone like that pre-1.2.

The long and short of it is that anyone with correctly working shields should be able to get some good mileage out of flanking forces and reserves too now, which I'm of course very happy to see since it actually gives us some play room tactically.

[Cheetah]

@Foz is correct, as long as I am using vanilla 1.1. I will try the leaked patch though to see the difference.

[Didz]

Not really....I agree that our playing styles are probably a lot different. My approach is very much based upon laying down heavy missile and skirmish fire on the enemy line rather than charging it, but nevertheless once the battle lines close a melee is a melee. Perhaps, the big difference is that my melee troops are usually on 'stand-ground' orders and most frequently spear or pike armed, so their role is basically to hold the enemy charge and survive rather than inflict heavy casualties.

Certainly, in seige battles against the Mongol horde single units of Russians survived for most of the engagement under constant melee attack whilst my archers on the walls behind them did all the killing.

I suspect also that we are comparing the campaign game with the PvP game and whilst I've not played PvP since STW I remember there was a huge difference in unit performance between the two.

In the campaign I regularly zoom in close to watch the melee action which is really only viable because they take so long to resolve.

Another possible difference is that I always play with Huge Unit sizes on which probably means units are more resilient due to their greater frontage and depth.

[Cheetah]

Yes, huge unit size can make a lot of a difference too, I am using normal. Also S&S infantry kills faster than spears, pikes, so it can be an other reason for the difference.

[Yun Dog]

I think its important to keep the various components separate, many of the things people mention are more related to the strat map tactics than the battle map

recruitment is a strat map problem and was always a problem but with MTW it wasnt as influencial to the the gameplay, because whatever the AI recruited joined the battle if you attacked their province. There are a number of adjustments that now need to take place to accomodate for the new RTW open strat map.
1) recruitments - the AI now needs a recruitment and expenditure script
2) castles vs citys - the AI now needs to accomodate for both of these settlement types
3) the AI needs to 'build' armies ie recuit x from castle y from city and then combine (a) ranged + (b) spear + (c) sword and cav + (d) general to make an army - this was done automatically when you attacked an enemy province in MTW
4) the AI must maintain a full stack in its settlements - previously when you attacked a province the castle units and the field armies all engaged - now the AI needs to understand the strategic importance of holding settlements - if it has spare troops then these hide in ambush en route to the city.

I think it woud also be useful for the AI to cluster its armies so more often it will out number or reinforce against the human

it wasnt uncommon in mTW to face a 6 stack assalt from the AI, now that is rare

The Tactical AI - the AI in battles is a shadow of the battle AI employed by MTWVI - I used to fear this AI to the level of ...maybe if I Autocalc Ill stand more chance!!!!

Its difficult to break down its behaviour into steps - but that is what has to be done
you cant just program in a players moves and hope they work because it has to respond to the players moves whatever they may be

but as said previously there needs to be gold rules followed by silver etc etc

hold the high ground (defence always - attack when possible)
on attack - outflank outflank outflank - and if that fails outflank
the army should keep its shape - in this case keep the AI relatively defensive - archers just in front of infantry wall - swords and cav on flanks waiting to outflank
army and unit facing - units should always face the persons main army with some units held on the flanks for defence
walls engage unit to unit, spare units move around the sides
threats from the player must be met with no more than necessary - spears and cav vs cav, swords vs spears
The Ai needs to seetle on a position and not continually readjust
once engaged the AI needs to keep track of spare units and send these to areas it is weakest - it needs to break off from routers and attend to the main battle with the majority of its men - as I say keep its shape and fight off the minor skirmishes - then the player has to break it down at least

the big thing is the AI needs to keep its shape and make sure all units are engaged 1-1 or worst 2-1

finally the general should be sent either round the back to hit the archers or where the AIs troops are about to turn

these are the only things I can think the MTW AI used to do - it kept shape, it moved into a superior position, it always seemed to find a way to outflank me - in the end I was building boxes of spears to try and stop it

if they did it once then surely it can be programmed again

[Cheetah]

Paradigm shift alert.

One of the big problems identified with AI armies is the number of cheap units they keep around. Another is its inability to disband units. A third is the hard time it seems to have in reinforcing cities/castles that are being attacked.

I think I have a solution that is not only relatively easy to implement on a coding level but also has a very realistic aspect to it. Militia units are a Militia, not a standing army.

As such, if you are fighting on your own territory you will have an option before battle to “Call up militia”. From a coding point of view this is a button on the pre-battle screen (one that shows how well you should fair against the enemy) that will open up a “Buy militia” screen. You can then pay a recruiting cost for militia units that can be produced in the province. This should be about 1/4th or 1/5th of what the current production price is. This would also work like mercenaries, in that you can call up more than the recruiting slots that province has. Unlike mercenaries, the amount you pay is deduced from your taxes for that province. As such you don’t have to have the cash handy, but you do pay for it the next turn.

How the program will decide what is available is pretty straight forward. Divide the population by 10. This is the maximum number of troops that can be called up. As they are called up, the population of the province is reduced by the same amount. After the battle these units automatically are disbanded and the population in the province is increased by a like amount.

Since all of the militia units would now be defensive only, you would no longer be able to build them for your standing armies. There may need to be some new units added or some renamed, but this should allow a city to call up town militia for defense but still allow for spear troops to be built there.

What this should do is allow the AI to build an offensive army while still being able to keep up on the defensive.

I like your idea but (a) I think the free miltia units in towns represent this "call up the milita" function; (b) I would go one step further.

Namely I would suggest that by building walls one would get the militia units of the given level of barracks in that town automatically. So if you build large walls you get 1 more militia unit automatically (assuming you already had 3). It would be town milita if there is no barracks or archery range in the city; it would be spear militia if there is a level 1 barracks; it would be xbow if there is a level 1 archery range and finally the player could chose if there are both barracks and archery range. However, these units would be "stickied" to the city so they cannot be moved out and would fight only if the city is under siege. This way the AI would have a minimal garrison in all of its cities and wont be able to move them out.

Also these free upkeep garrisons could be given for catles as well (perhaps fewer slots). Towns have already many advantages over castles, giving free upkeep slots for castles would perhaps give more incentive to keep castles and not to convert all (or almost all) to cities (ofc my previous idea would give an other incentive too). Of course, castle slots would be filled up by the appropriate level castle units, so this way, just in case of towns, the AI would have a fix high level (assuming it upgrades its barracks, which it does) garrison in all of its castles.

[Cheetah]

Now, lets move forward.

Assuming that the AI gets an even ground in terms of cavalry and morale what kind of armies you would expect from the AI?

First of all I would suggest that the AI should build differetn siege and field stacks, and of cuorse guard siege stacks with field one (i.e. march them together) but it should fight out the siege on the battle map only if it has a siege stack there.

So what I am interested in is suggestions for siege and field stacks for all factions.

Here is my basic idea, feel free to disagree : I think that given the ability (or inability) of the AI to handle shooters the AI is better off without archers, or just with a few. Also, the fewer unit types it has perhaps the better it will keep the army together.

So lets say for Byzantines I would suggest the following siege stack (or something similar in composition):

1 BG
3 katapracts
6 VG
4 byz swordsmen
4 byz spearmen
2 trebuchet

Byz field stacks:

1 BG
19 vardariotai

... I would love to see such stacks from the AI.

1 BG
6 katapratcs
4 byz lancers
6 VG
3 trebizond archers

or just

1 BG
10 katapratcs
9 VG

perhaps a bit more realistically (in terms of upkeep):

1 BG
3 katapratcs
3 byz cav
4 byz lancers
3 VG
3 byz swordsmen
3 trebizond archers

The idea is to have lots of melee in field stacks, mostly cavalry and S&S and 2H axe units.

[R'as al Ghul]

The AI should be able to outcalculate the player in tactical battles.

It should be able to adjust its own formation and tactics depending on the players units' strength and formation. Before the armies make melee contact the AI should know what the outcome is, because it can just calculate it. Given on the foreseen outcome of the individual matchups it should plan flanking maneuvers and execute them in perfect synchronization with the frontal attack. It should frustrate the hell out of the player because you're never able to catch a skirmishing unit because the AI will always withdraw in the last possible millisecond. It should frustrate the player because it always calculates the higher ground, it always keeps a good formation, it's aware of the importance of the general, it knows the little details of the units better than the player and uses them to its advantage.

On the campaign map it's similar.
The AI should always be the richest and best developed because it knows exactly where the richest provinces and the best trade resources are. Timbuktu? We should've learned that by following an AI merchant. It would build huge trade empires and maintain them. It would use governors to best effect. It's agents wouldn't just linger around your cities achieving nothing, they'd start by sabotaging your entire building structure until they'd valoured up enough to be able to kill all your family and your best generals. It would always attack when you're weakest, with the best possible units available. Siege engines? It would only bring them deliberately to attack castles. In field battles you'd face powerful armies that outtech yours. The composition of the Armies should be balanced but favoring factions' specialties, like Horse Archer heavy armies. (Apropos HA, the most difficult thing for a player to beat should be a HA heavy army because here the AI can really shine in that it can control this bag of flies that 20 independently acting HA are.) The AI should keep with the time in so far as that it disbands old units and replaces them by new ones. It should be the first to field Gunpowder armies and it should show the player how to use a Pike and shot army by destroying the player.



Feel free to add some more shoulds

[RickooClan]

Outcalculation can be done on game with turn base like chess, but it wont work on the TW battle field at least.

Since even the AI could get the right match out or tactical move against human, it will take times to work that out (different with turn base) and the situation might already be changed while their move is half way done.

[Noir]

originally posted by RickooClan

Outcalculation can be done on game with turn base like chess, but it wont work on the TW battle field at least.

Since even the AI could get the right match out or tactical move against human, it will take times to work that out (different with turn base) and the situation might already be changed while their move is half way done.

Perhaps you are reffering to the AI of M2TW?

The tactical AI of MTW with tweaking only of available options as far as i understand it is outcalculating the player and is one of the best rushers (if not the best) i've ever seen (including my limited experience in MP) combining all units to a punch that routes even the most melee infantry heavy armies almost 50% of the time after a long series of reloads for the same battle and a few buckets of sweat, blood and tears, in Samurai Wars. His general never rushes needlessly nor risky to the melee, but simply supervises his blitz by adding key touches as engaging just enough to ensue routes in key melee nuclei during the battle.

The player feels at times helpless whether has deployed in a long line or in a tight packing as in the first case it cannot match the AI controlling all units simultaneously and continuously and thus misses many parts of the action including significant flank attacks/charges that are out of his field of vision and is outmatched in speed of maneuvers and overall coordination. In the second the AI uses effectively the ridiculed by M2 players blobbing penalty to take out the player's army by superior match ups as the player's formation essentially proves an obstacle to his own unit maneuvers.

Noir

[Puzz3D]

Outcalculation can be done on game with turn base like chess, but it wont work on the TW battle field at least.

It does work in TW. If the AI's unit is stronger, it will initiate a frontal attack on your unit. If the AI's unit is equal or weaker, it will initiate a flanking attack. If you take the initiative, you can get favorable matchups against the AI. If you sit back, you are going to get into trouble unless you have some well placed reserve units with which to respond to weak spots or enemy flankers. MTW/VI wasn't the best showcase for the old engine tactical AI because the RPS was weakened in that version.

Since even the AI could get the right match out or tactical move against human, it will take times to work that out (different with turn base) and the situation might already be changed while their move is half way done.

If you play mtw/vi, you'll see the AI change its plan if the tactical situation changes. For instance, as the attacker the AI will start moving its army laterally into a position to make a flanking attack. If you rotate your army to face the AI's army, it will start moving back the other way to set up an attack on your other flank. If you are persistent about turning your army to face the AI, it will eventually decide to make a frontal attack with only the cav sent to the flanks. If you put your cav on the flanks, it will counter by sending spears to the flanks to attack your cav. If you overcompensate to the outside, the AI will strike at your weakened center. One problem with mtw/vi is that the AI didn't know how to manage fatigue, so the large maps introduced in the mtw/vi campaign weakened the AI as an opponent.

[nikolai1962]

One problem with mtw/vi is that the AI didn't know how to manage fatigue

I think this is caused by the variable move speed on the march. The slower units are running all the time to keep up with the faster units.