Campaign AI Observations and Suggestions.

[kammm]

Hello! First of all I would like to say that I really appreciate all the job that modders did since RTW, I play the TW series since shogun, but started using
mods only on RTW, and I think you guys brought the game to almost its full capacity! Much was done with the battle AI and diplomacy AI, cosmetic and historical corrections, but I think theres a main flaw that still needs some corrections(if possible) to bring the game to another level and thats the military campaign AI.

I love challenging strategy games and this last 3 weeks I played lots of
campaigns with SS 5.1b, and made some observations on WHY this game after all, is still too easy.

I played 4 campaigns total with lusted's AI, with HRE, Spain, Moors and Lithuania, and 2 campaigns with Xeryx XAI mod for SS 5.1b with Moors and Lithuania again.(always on VH/VH)


I didnt tried GrandViz Ultimate AI cause i saw complains about its AI on forums, so i didnt bother testing.

Finally, my observations:

1. AI still does not know how to face the player, it brings weaker armies
to the field, with the power score sometimes being 2:1 in player favor, but most of times around 5:4, 4:3 (player).

2. AI does not understand that sieges are harder than field battles, and still do suicide sieges completely outnumbered, and with crappy troops(like javelinmen)

3. AI does not garrinson border settlements correctly, sometimes even the capital is abandoned with only the faction leader and 3 random units(sometimes even siege units) inside.

4. AI almost always have a bigger military power, but its forces are divided in lots of small and half stacks, unable to face the player army. For example, i had a game with Lithuania(the hardest faction to play IMO) i was sieging kiev and an AI army(half stack) comes to save the capital, they were outnumbered 5:4 in power and nearby on a bridge there was ANOTHER half stack, sitting around fishing, if they had merged both stacks i would be doomed for sure(i checked later and that army was there since 4 turns earlier), the kievian rus(rutherians) were in war with me only.

5. Finally, the AI dont know how to make decent armies, sometimes they bring lots of javenlimen(like 7-8), lots of siege engines, tons of light infantry when there is lots of better untis avaliable to them.

All that said, i have some suggestions that in my opinion would make the game ALOT better, and close to its full power.

1. First of all the AI should understand that the player faction is a complete different faction, and only attack player forces when they have the upper hand, always retreat when outnumbered(even in situations like 10:9). With the latest major improvements to Battle AI, its not easy to beat a powerfull AI Stack that outnumbers you in power, this behavior alone would make the game a lot harder.

2. The AI should field LESS but MORE powerfull armies, instead of 4 half stacks wandering around, why not 2 full stacks?

3. Less siege engines, more heavy cavalary, AI sucks at using sieges weapons, but knows how to use heavy cavalary in a decent way. More heavy/light infantry and less ranged units also.

4. Better garrinsons, AI has billions of florins avaliable, so it should garrinson its border cities/castles in a decent way.

Thanks for everyone who read this post till the end, I would like to state that since I dont understand **** about modding/programing, I dont know if those changes are even possible, but I think thats the best way I can help the modding community in improving the game, too bad CA made such a great game but left its AI unfinished.

Thanks for your hard work, I know its not easy, by the way, SS 6.0 preview looks great!

[Dave Scarface]

Hi Kammm,

Let me reply to some of your observations here.

First of all it will never be possible to get the AI to play like a human. But I believe you can make the campaign a lot harder two ways. First you need a good enough money script and debt script. For the AI to be able to continue to build good armies. Lusteds campaign AI likes to make big armies but then it will go broke. When AI factions are at war they go bankrupt pretty quick as well. You can have the best AI in the world but it will be useless when the AI cannot afford more armies. Secondly I think all mods should use the recruit_priority_offset. Stainless Steal forces the AI to train better troops by making the replenishment pools slower for less powered units. But this is not enough, you must use the recruit_priority_offset as well. In my Kingdoms Grand Campaign Mod the AI builds the best armies all the time, with no exceptions. King Kong is very welcome to copy this into Stainless Steal if he wishes.

Also we cannot stop the AI from wanting to attack settlements, since it is the AI's main objective to capture a certain number of regions. About Lusteds AI not garrisoning its borders, that is true normally when the AI is not losing a war. However when an AI faction is losing a war it does defend its settlements deep and go fortified as well. Some mods use garrison scripts and this solves the problem completely. The downside is though that a large script slows turns down considerably. Therefore in my mod I have a garrison for capitals only. And because thats only about 20 settlements this has no effect on slowing the game at all. Anyway I hope that I have helped answer some of the points you made. And I believe that you definately can make this game more challenging.

best wishes
Dave

[messenger42]

Hi Dave!

I remember your discussion with Lusted over his final AI before he retired from moddling. Probably you're the one who can continue his work since you've been moddling and testing his AI from the very early stages. If I want to use your AI for SS 5.1b how would I do that (what files do I need to copy and paste). now that your AI is probably more improve from the last Lusted AI since you've done a bit of work on it. Hopefully it will be in SS 6.0

and in your fixpack Echad!

Thanks

[kammm]

Hi Kammm,

Let me reply to some of your observations here.

First of all it will never be possible to get the AI to play like a human. But I believe you can make the campaign a lot harder two ways. First you need a good enough money script and debt script. For the AI to be able to continue to build good armies. Lusteds campaign AI likes to make big armies but then it will go broke. When AI factions are at war they go bankrupt pretty quick as well. You can have the best AI in the world but it will be useless when the AI cannot afford more armies. Secondly I think all mods should use the recruit_priority_offset. Stainless Steal forces the AI to train better troops by making the replenishment pools slower for less powered units. But this is not enough, you must use the recruit_priority_offset as well. In my Kingdoms Grand Campaign Mod the AI builds the best armies all the time, with no exceptions. King Kong is very welcome to copy this into Stainless Steal if he wishes.

Also we cannot stop the AI from wanting to attack settlements, since it is the AI's main objective to capture a certain number of regions. About Lusteds AI not garrisoning its borders, that is true normally when the AI is not losing a war. However when an AI faction is losing a war it does defend its settlements deep and go fortified as well. Some mods use garrison scripts and this solves the problem completely. The downside is though that a large script slows turns down considerably. Therefore in my mod I have a garrison for capitals only. And because thats only about 20 settlements this has no effect on slowing the game at all. Anyway I hope that I have helped answer some of the points you made. And I believe that you definately can make this game more challenging.

best wishes
Dave

Thanks for your reply! I know that the AI can never be as good as a good human player, but its like on civilization where the AI cheated like hell to compensate its "dumbness" and we all had a good fight!

Nice to know that youre working on the AI, as soon as im able i would like to test your AI, i will be off for 3 weeks from now on, im making this observations also on XAI forum, and i wish you guys good luck, on "fixing" the AI.

[pwijnands]

First you need a good enough money script and debt script. For the AI to be able to continue to build good armies. Lusteds campaign AI likes to make big armies but then it will go broke. When AI factions are at war they go bankrupt pretty quick as well. You can have the best AI in the world but it will be useless when the AI cannot afford more armies.

I don't think that the AI factions should receive so much extra money that they will never go broke. One of my tactics is to massively sabotage their most expensive buildings, forcing them to go broke. I'd hate it if a script would then give the AI so much money it can constantly repair and build armies at the same time. I know scripting doesn't support much in the terms of variables and functions (you can't do a function to see how much money a faction has and then store that in a variable and use that for a add_money command).

[hhawks12]

I have not had too much trouble when it comes to the AI and money. However I have a few issues with the AI in general. While they do not have extreme amounts of money except in a few situations, the money that they do receive encourages them to produce huge numbers of siege weapons in their army. In my Norwegian Campaign I (H/VH) I have been confronted with well balanced armies, but this is always eclipsed by about 5 or 6 (or even more) artillery units per army. Obviously, as a slow moving heavy infantry faction like Norway, I endure extreme, and unecessary, casualties from these weapons. Not to mention they tend to be expert snipers and seem to have perfect accuracy even when using flaming missiles. I have lost many a general this way even when he is not engaged in combat.

Generally though, I have noticed how superior the battle AI is. They flank, charge at weak points in the line, etc, though I have noticed that enemy cavalry units pursue my archers with a fervor.

My last gripe is siege AI. At times it is decent, other times it is awful. Pathfinding is poor most of the time for both sides, but when the enemy is assaulting my settlement, they tend to stop half way through and just do nothing. For example, i was defending Toulouse from a crusade, and watched as six or seven enemy units tried to mount the walls with ladders/towers, while the other infantry units sat well out of range, frolicking in the fields with four good battering rams sitting idle. And when the AI brings multiple armies to a settlement assualt, in all my experiences thus far, the other army does nothing and I am forced to sally out of my own castle and attack them. Usually I just auto_win defender and save myself the trouble.

[Echad]

Well, I thought about recruit priority and asked Lusted about it myself... Dave, can you blend your recriut_priority values to SS EDU file? =\ We all would be really grateful for it!

[Dave Scarface]

Okay guys I will need to get King Kongs permission first of all though. I would be very happy to try and help make the army recruitment in Stainless Steal as good as I can make it.

best
Dave

[Echad]

You may do it just for my Fix Pack, you don't need to ask KK's permission just to make a modification of his file for yourself, me and the community...

And he prefers just to collect everything that is ready, AFAIK, so, will you be so kind to make a file for us? =)

And, is there a priority value for buildings as well? =\ Cause having Ballista factories in every town really kills the AI.

Edit: There was a project (AFAIK only on discussion level) to add some heroes to the campaign, we might ask you for your achievments in this field.

[Dave Scarface]

Edit: There was a project (AFAIK only on discussion level) to add some heroes to the campaign, we might ask you for your achievments in this field. :wink:

It wouldn't be a problem for me to do that as well. But it would just take me a little longer. Would you guys like every hero added like in my Kingdoms Grand Campaign Mod?

Dave

[Dave Scarface]

Btw does anybody know yet how long it might be until Stainless steal 6.0 gets released? And will it include any new factions and units? If so it might be more sensible for me to wait and release this then.

I would also like to point out that I am not working on the campaign and battle AI right now. That would require a huge amount of time and effort. I am only proposing to modify the EDU and maybe change the money script. And add the heroes from Kingdoms, Im not making new heroes as well. Someone else can carry on with that project if they wish.


Dave

[Echad]

Yes, something like that, don't ask me, I know nothing of the project, and STILL haven't tried your mod (will do it in a few days), the idea was about making heroes for every faction, like Wallace is now for Scots in SS; Your barbarossa won't be bad and, of course, everyone wanted dear Joanne...

[Echad]

Just do the EDU edits for now, I'll include them in Fix Pack, KK may include it in 6.0, better do it for now... KK may append himself to the new units later. Because if you'll wait for 6.0 it will be only in 6.1 or 7.0, see? So do it now and we'll have it in 6.0!

6.0 will have Khwarezm, Cumans and Genoa, but it's the same as Milan still.

[wolfslayer]

It would be great to have the recruit_priority_offset and the garrison script; anything to make the AI as challenging as possible without hobbling the Human player. Long turns while the AI does it's thing is no prob for me..lol.

I've been trying to test forts in SS and desperately trying to get the AI to attack them, they sure won't stay inside them oncce they capture them, and the AI has only built 2 forts the entire campaign; 1 by HRE and 1 by Lithuania.

I've been very carefully using the kingdoms trainer to only give enemy captains and generals 10 star ranks, to make them more agressive for my testing, and in the process getting some of my best stacks surrounded and completely wiped out on the battlemap, the past few battles have been some of the best I've ever played.

[Echad]

I asked Dave Scarface of the Kingdoms Grand Campaign mod to do recruit_priority_offset for SS, he agreed... You should look above for this.

The only question is will he do it for 5.1, or we'll be forced to wait until' 6.0, which is baaad!

[Dave Scarface]

The only question is will he do it for 5.1, or we'll be forced to wait until' 6.0, which is baaad!

Oh alright then I wont make you wait I will be releasing the new EDU for SS 5.1 later today. And I am looking forward to getting some feedback on my changes.

Cheers

Dave

[Dave Scarface]

Hello guys,

I am pleased to present a small improvement for the AI in SS. When King Kong first made his Stainless Steal mod his army recruitment was the best in the community. By using the export_descr_buildings.txt, he was able to force the AI into building better units. But he has not taken full advantage yet of the EDU I believe. Now not only is the AI limited to what it can build but it is also programmed to like certain units more. Using the recruit_priority_offset you will see even better quality armies than before.

How this works, some of the ordinary units have been left with a recruit_priority_offset 0. This means they stay the same. Very weak units such as peasants are set as low as recruit_priority_offset -50. Siege weapons are also set lower than 0. But even with this in my mod the AI still builds siege weapons. Most archers are set to recruit_priority_offset 5 on average, with some going as high as +10. Knights and dismounted knights go as high as +20. Basically Ive made sure all the best units with the best armour get recruited the most. This will work best in the late era campaign because there are more higher quality units available from the start. Also if the AI is going to keep on making better armies it must have more money. So I suggest increasing the money script on VH difficulty in the campaign_script.txt.


Dave

[Gotee7]

Thanks Dave. Is this compatable witrh Save Games??

[Dave Scarface]

Yes mate changes made to the EDU will work in a saved game.


all the best
Dave

[wolfslayer]

Yes mate changes made to the EDU will work in a saved game.


all the best
Dave

Thanks Dave, going to have to check this out.