Skynet AI

**okzersal** · October 06, 2021, 11:23 AM

Good afternoon. Can you please tell me, have you fixed such a bug when the AI sends siege stairs and towers to dilapidated walls during an assault? After all, because of this, the AI begins to freeze and lag.

**Dominick** · December 10, 2021, 06:53 PM

Good evening gentlemen.

I'm having a problem implementing the battle script of Skynet AI into the campaign_script of a modded Teutonic campaign, as what I've done so far breaks the campaign in the sense that no faction can upgrade castles to fortresses and citadels anymore and some other campaign problems (probably has something to do with counters).

If you could tell me what I've done wrong in implementing the battle script I would appreciate it a lot.

Here's the code of the campaign script:

Spoiler Alert, click show to read:

Thank you in advance!

**Jadli** · December 11, 2021, 04:31 AM

Skynet doesnt itself edit files that deal with upgradability of settlements and so on...

So my only guess is that you broke the campaign script by importing this incorrectly, as you suggest indeed (though maybe it has more to do with AI not having the bonus money). If there is an error in the script, while campaign_script stops working. You can check this for example by whether some welcome historical/event scrolls show up or not (if you know when the scroll is supposed to show). If it doesnt show up, no script is working. And of course, if there is an error, it should show in the log.

Anyway, by briefly looking at this, I believe I see the first (and probably only mistake that you could have done when importing the script). You imported the script between "wait_monitors" and "end_script". Afaik, "wait_monitors" need to be at the end of the script (before the "end_script"), so move the line after the Skynet battle script. That should make the script work correctly... though, you will of course need to start a new campaign.

**Dominick** · December 11, 2021, 09:18 AM

Originally Posted by Jadli

Skynet doesnt itself edit files that deal with upgradability of settlements and so on...

So my only guess is that you broke the campaign script by importing this incorrectly, as you suggest indeed (though maybe it has more to do with AI not having the bonus money). If there is an error in the script, while campaign_script stops working. You can check this for example by whether some welcome historical/event scrolls show up or not (if you know when the scroll is supposed to show). If it doesnt show up, no script is working. And of course, if there is an error, it should show in the log.

Anyway, by briefly looking at this, I believe I see the first (and probably only mistake that you could have done when importing the script). You imported the script between "wait_monitors" and "end_script". Afaik, "wait_monitors" need to be at the end of the script (before the "end_script"), so move the line after the Skynet battle script. That should make the script work correctly... though, you will of course need to start a new campaign.

Thank you! Your resolution solved my problem. I now went to check my old Retrofit mod installation in which I also applied Skynet and I did it wrong there as well, thus also breaking my campaign script, however I didn't notice as I finished the campaign in Retrofit in about 75 turns, so I thought for example that the Mongols just didn't arrive yet.

Good I noticed something's wrong in the Teutonic campaign.

Maybe including a short explanation on implementing the battle script in the campaign_script file in the README for modding newbs such as myself wouldn't be a bad idea.

On another note, any idea why the AI would be doing this? VH/VH difficulty, many small armies instead of a larger relief force. It took the AI a couple of turns to do this, and the next turn it brought in the final small army to attack.

Spoiler Alert, click show to read:

**bitterhowl** · December 11, 2021, 09:43 AM

What about descr_campaign_db section about siege and attack parameters?

**Jadli** · December 11, 2021, 10:11 AM

Yea, probably, these lines specifically in descr_campaign_db.txt.

Code:

      <att_str_modifier float = "0.70"/><!--modifies the effective attackers strength when determining the priority of making attack decision (i.e. att_def_strength_ratio = ((att_str*att_str_modifier)/def_str). z3n 0.75 . Set lower to attack with larger force -->
      <siege_att_str_modifier float = "0.55"/><!--modifies the effective sieging attackers strength when determining the priority of making attack decision  z3n 0.75 Set lower to attack with larger force  -->
      <crusade_att_str_modifier float = "1.2"/><!--modifies the effective crusading sieging attackers strength when determining the priority of making attack decision  z3n 1.00 Set lower to attack with larger force -->
      <sally_att_str_modifier float = "1.0"/><!--modifies the effective sallying attackers strength when determining the priority of making attack decision   z3n 0.75 Set lower to attack with larger force -->
      <ambush_att_str_modifier float = "1.25"/><!--modifies the effective ambushing attackers strength when determining the priority of making attack decision -->
      <str_limit_weak float = "0.80"/><!--min ideal strength ratio modifier for determining when an army is far too weak for an attack ( att_def_strength_ratio < (ideal_str_ratio*str_limit_weak) ) z3n 0.75  Set HIGHER to attack with larger force -->
      <str_limit_strong float = "999.0"/><!--max ideal strength ratio modifier for determining when an army is far too strong for a fair attack ( att_def_strength_ratio > (ideal_str_ratio*str_limit_strong) ) -->

Im using Skynet as well in my mod, though significantly changed some aspects to improve/better fit my mod. I had the same issue as you, so I modified the lines to have larger armies, so just feel free to to copy those lines, should work fine for you, considering this is loosely based on original Skynet values as well.

Though this ofc doesnt fully get rid of this, as there other thigs that would be much harder to modify, i.e., see the guide on pathfinding and so on in this subforum by z3en

**Dominick** · December 11, 2021, 08:38 PM

Originally Posted by Jadli

Yea, probably, these lines specifically in descr_campaign_db.txt.

Code:

      <att_str_modifier float = "0.70"/><!--modifies the effective attackers strength when determining the priority of making attack decision (i.e. att_def_strength_ratio = ((att_str*att_str_modifier)/def_str). z3n 0.75 . Set lower to attack with larger force -->
      <siege_att_str_modifier float = "0.55"/><!--modifies the effective sieging attackers strength when determining the priority of making attack decision  z3n 0.75 Set lower to attack with larger force  -->
      <crusade_att_str_modifier float = "1.2"/><!--modifies the effective crusading sieging attackers strength when determining the priority of making attack decision  z3n 1.00 Set lower to attack with larger force -->
      <sally_att_str_modifier float = "1.0"/><!--modifies the effective sallying attackers strength when determining the priority of making attack decision   z3n 0.75 Set lower to attack with larger force -->
      <ambush_att_str_modifier float = "1.25"/><!--modifies the effective ambushing attackers strength when determining the priority of making attack decision -->
      <str_limit_weak float = "0.80"/><!--min ideal strength ratio modifier for determining when an army is far too weak for an attack ( att_def_strength_ratio < (ideal_str_ratio*str_limit_weak) ) z3n 0.75  Set HIGHER to attack with larger force -->
      <str_limit_strong float = "999.0"/><!--max ideal strength ratio modifier for determining when an army is far too strong for a fair attack ( att_def_strength_ratio > (ideal_str_ratio*str_limit_strong) ) -->

Im using Skynet as well in my mod, though significantly changed some aspects to improve/better fit my mod. I had the same issue as you, so I modified the lines to have larger armies, so just feel free to to copy those lines, should work fine for you, considering this is loosely based on original Skynet values as well.

Though this ofc doesnt fully get rid of this, as there other thigs that would be much harder to modify, i.e., see the guide on pathfinding and so on in this subforum by z3en

I did what you suggested and tested it and...

Spoiler Alert, click show to read:

**bitterhowl** · December 11, 2021, 10:15 PM

Interesting

And what about final lines in config_ai_battle?

And it's a new campaign, with new map.rwm, right?

**Dominick** · December 12, 2021, 10:38 AM

Originally Posted by bitterhowl

Interesting

And what about final lines in config_ai_battle?

And it's a new campaign, with new map.rwm, right?

Ah, so it is necessary to do that as well for the new descr_campaign_db.xml to start working. The campaign was new, but with an older map.rwm.

Is there some list of files which are affected by map.rwm when starting a new game? Before I was deleting map.rwm by default as I was working on campaign map changes a while back, but I didn't in this instance as I didn't think it was necessary and I didn't come across the folder of map.rwm.

I'll go test it now again. Thank you!

Here's the code you asked, that whole file is from z3n.

Code:

        <battle-analyser>
            <!-- ratio of friendly to enemy strength to be considered more powerful -->
            <friendly-to-enemy-strength-ratio>1.0</friendly-to-enemy-strength-ratio>
            <!-- ratio of friendly to enemy melee strength to be considered overwhelmingly powerful -->
            <enemy-melee-strength-multiplier>1.1</enemy-melee-strength-multiplier>
            <!-- ratio of friendly strength to enemy melee strength to force attack -->
            <friendly-melee-strength-divisor>0.9</friendly-melee-strength-divisor>
            <!-- ratio of enemy to friendly ranged strength to be considered overwhelmingly powerful -->
            <friendly-ranged-strength-multiplier>3.0</friendly-ranged-strength-multiplier>
            <!-- ratio of friendly strength to enemy ranged strength to force attack -->
            <friendly-ranged-strength-divisor>5.0</friendly-ranged-strength-divisor>
            <!-- distance behind the defensive line the enemy must reach to be considered to have broken through -->
            <enemy-position-buffer-distance>100</enemy-position-buffer-distance>
        </battle-analyser>
    </gta>
</config>

<variations>
</variations>

**Dominick** · December 12, 2021, 11:50 AM

Eh, I deleted the map.rwm and the game generated a new one, but similar things are still happening.

Spoiler Alert, click show to read:

**Jadli** · December 12, 2021, 12:08 PM

Well, follow the comments (after hashtag) in the lines I provided, and based on the explanations there try to put the values towards even larger armies

**Dominick** · December 12, 2021, 12:44 PM

Originally Posted by Jadli

Well, follow the comments (after hashtag) in the lines I provided, and based on the explanations there try to put the values towards even larger armies

I feel the problem isn't in the size of all forces combined, but in the fact they are in different stacks, as if the AI often doesn't take the time to put them all in the same stack, but attacks with divided stacks attacking at the same time. So the AI seems to attack taking the numbers you provided in consideration, but it doesn't attack with only one powerful stack, and that is the problem.

Can we tell the AI to prefer merging its units in one stack before attacking?

**Jadli** · December 12, 2021, 01:00 PM

Originally Posted by Jadli

...
Though this ofc doesnt fully get rid of this, as there other thigs that would be much harder to modify, i.e., see the guide on pathfinding and so on in this subforum by z3en

Not super experienced in this tho, perhaps bitterhowl knows more...

**bitterhowl** · December 12, 2021, 10:29 PM

Haven't seen last version of Skynet at the moment. Probably the deal with decisions parameters.

invade_start known to gather stacks together, at least near capital. Posted link about that somewhere around AI Workshop.

**z3n** · December 14, 2021, 10:02 PM

Honestly this problem is not solvable by the modder, unless they wish to neuter the AI completely.

This is brought out by the fact CA later consolidated all stacks into "forced" general stacks in all the newer TW games; while also limiting the amount of armies factions can have depending on their size. If CA gave up (the ones who can program the code behind the decisions we used), then I'd say it's something of a lost cause...

bitterhowl is correct however, invade_start can work pretty well. But every decision comes with a sacrifice, either in aggression, compactness (to some degree), or reactivity in terms of defense.

Maybe, just maybe, you could find a correct rhythm to flip the AI in and out of these long term plans properly by using the campaign script (we already something similar to this in EBII for peace and war ai label "personalities"; so mechanically it's 100% possible). I would be interested in how a test like that would turn out, but I'm currently busy with some RL things.

Since we know that frontline_balance is completely borked. Free_strength is also kinda messed up from what I recall and military_balance is only moderately useful.

So an idea using ai_labels flip the invasion decisions within a few turns (maybe use a counter, then reset the counter to 0 once it hits 25, and start the whole pattern again), e.g.
1) From turn 0-5 turns use invade_start ai label
2) From turn 5-10 use invade_buildup
3) From turn 10-15 use invade_opportunistic
4) From turn 15-20 use invade_immediate
5) From turn 20-25 use invade_start again

That way you won't be stuck using 1 invasion decision against a target all the time. Then and again, from what I've seen the AI does also respond pretty well to using just one invasion decision against targets... Like I said though, our main problem is that frontline_balance and free_strength is completely off the walls incorrect sometimes, it just doesn't make sense to use those tests. From what I remember the vanilla file uses frontline_balance + free_strength + military balance pretty often, but I don't think that's even good either. Like 1 unit heads over to the frontline and then suddenly their frontline balance is 50000000000000000000000 : 1. free_strength is often just completely incorrect (since according to cavalrycmdr it uses the slave faction to calculate against as well, or other powerful factions)

Sometimes their AI even freezes, so "free_strength" is really questionable. It must have all sorts of problems with infinite loops or something pretty off. Their programming is just bugged. Basically the only programming that isn't super bugged might be something somebody creates on their own, within the campaign scripts, and then use that to flip ai labels which seem reasonable. If you want to try that out. That'd be a real challenge though and my opinion is that it isn't solvable. We can probably only at best mitigate weird behaviours.

**R.P. Gryphus** · December 18, 2021, 11:29 AM

Is there any text description of what this mod/resource do specifically, especially regarding the campaign AI?

**z3n** · December 20, 2021, 02:13 PM

Well basically it's a complete rewrite of the decision tree the LTGD uses to plan its invasions and defense decisions. I never really kept any complete list of changes since even those get rewritten.

**Jadli** · December 21, 2021, 04:11 AM

Originally Posted by z3n

Honestly this problem is not solvable by the modder, unless they wish to neuter the AI completely.

This is brought out by the fact CA later consolidated all stacks into "forced" general stacks in all the newer TW games; while also limiting the amount of armies factions can have depending on their size. If CA gave up (the ones who can program the code behind the decisions we used), then I'd say it's something of a lost cause...

bitterhowl is correct however, invade_start can work pretty well. But every decision comes with a sacrifice, either in aggression, compactness (to some degree), or reactivity in terms of defense.

Maybe, just maybe, you could find a correct rhythm to flip the AI in and out of these long term plans properly by using the campaign script (we already something similar to this in EBII for peace and war ai label "personalities"; so mechanically it's 100% possible). I would be interested in how a test like that would turn out, but I'm currently busy with some RL things.

Since we know that frontline_balance is completely borked. Free_strength is also kinda messed up from what I recall and military_balance is only moderately useful.

So an idea using ai_labels flip the invasion decisions within a few turns (maybe use a counter, then reset the counter to 0 once it hits 25, and start the whole pattern again), e.g.
1) From turn 0-5 turns use invade_start ai label
2) From turn 5-10 use invade_buildup
3) From turn 10-15 use invade_opportunistic
4) From turn 15-20 use invade_immediate
5) From turn 20-25 use invade_start again

That way you won't be stuck using 1 invasion decision against a target all the time. Then and again, from what I've seen the AI does also respond pretty well to using just one invasion decision against targets... Like I said though, our main problem is that frontline_balance and free_strength is completely off the walls incorrect sometimes, it just doesn't make sense to use those tests. From what I remember the vanilla file uses frontline_balance + free_strength + military balance pretty often, but I don't think that's even good either. Like 1 unit heads over to the frontline and then suddenly their frontline balance is 50000000000000000000000 : 1. free_strength is often just completely incorrect (since according to cavalrycmdr it uses the slave faction to calculate against as well, or other powerful factions)

Sometimes their AI even freezes, so "free_strength" is really questionable. It must have all sorts of problems with infinite loops or something pretty off. Their programming is just bugged. Basically the only programming that isn't super bugged might be something somebody creates on their own, within the campaign scripts, and then use that to flip ai labels which seem reasonable. If you want to try that out. That'd be a real challenge though and my opinion is that it isn't solvable. We can probably only at best mitigate weird behaviours.

BTW, this is not fully related to AI scripting (in fact, it kinda "overrides" the AI), but what I noticed that is the best and most reliable solution for AI to use full stacks (also led by generals, not just captains), instead of a bunch of small armies, is to spawn those full stacks for them... I noticed that they usually dont split those armies.. and if you spawned a general with the army, he remains leading it ...

**z3n** · December 25, 2021, 02:18 PM

Originally Posted by Jadli

BTW, this is not fully related to AI scripting (in fact, it kinda "overrides" the AI), but what I noticed that is the best and most reliable solution for AI to use full stacks (also led by generals, not just captains), instead of a bunch of small armies, is to spawn those full stacks for them... I noticed that they usually dont split those armies.. and if you spawned a general with the army, he remains leading it ...

Interesting, what happens after they conquer a settlement; the armies split or remain?

**jupsto** · January 02, 2022, 10:50 AM

Am I right in thinking if I have installed the mod as instructed and want to switch to "bloodthirsty" I can just extract and rewrite existing files only.

What exactly does the bloodthirsty version do?

I have been testing this AI it feels a bit passive so will try bloodthirsty, but it might mainly feel that way because I have the occupy relations bugfix applied which I think is a huge change vs vanilla.

edit:
I had to set ai_label for papacy back to papal_faction from default because was too op to get excommunicated by aggressive pope on turn 2-3
however it doesnt fix it, papacy still aggressive and is not giving player quests such as cease hostilities etc.

Thread: Skynet AI

Thread Tools

Search Thread

Display

Re: Skynet AI

Re: Skynet AI

Re: Skynet AI

Re: Skynet AI

Re: Skynet AI

Re: Skynet AI

Re: Skynet AI

Re: Skynet AI

Re: Skynet AI

Re: Skynet AI

Re: Skynet AI

Re: Skynet AI

Re: Skynet AI

Re: Skynet AI

Re: Skynet AI

Re: Skynet AI

Re: Skynet AI

Re: Skynet AI

Re: Skynet AI

Re: Skynet AI

Posting Permissions