Campaign AI - Beta 1.4.0

[Dabos37]

I have been doing some tests in the first 5 days o a new Beta 1.4.1 campaign. Here is a perfect example of what we have been talking about the AI is clearly prioritizing attack over the defense, which is a negative behavior for campaign balancing and increase snowballing and lords defection.

1- As you can see, the Western Empire funded a huge army with 1200 men and start a besiege on Marunath:

17

Image 17 hosted in ImgBB

ibb.co

2- Then I went to Lagerta and a much smaller Battania army just started a besiege on this town.

18

Image 18 hosted in ImgBB

ibb.co

3- Western Empire had not started the attack when Largerta besiege was started but anyway, the WE army just ignores the Largeta siege and continue with the Marunath siege.
4- The result was WE taking Marunath while Battania taking Lagerta. It does not look too bad, both factions have just traded a town and are similar than they were before, but then you realize that it is actually really bad for campaign balancing and lords get a penalty for losing settlements and relationship penalty with the king, which increase chances of defecting.

This new issue was introduced with lasted patches due to people complaining about AI abandoning ongoing sieges in order to defend. If I would have to choose, I prefer the old AI behavior by far, where the AI just abandoned sieges in order to defend. On the other hand, it would be great if the new behavior could be tweaked and AI just does not abandon sieges if the actual attack is ongoing for some time and the AI army will conquer the settlement soon. Is this possible to do?

Currently in beta 1.4.1, it looks like once an army start a besiege (just the besiege and not the attack to the settlement), then the army does not abandon It in order to defend which I find wrong.

Plus, smaller armies (compared to attacker force) should try to help friendly besieged settlements like they did before, just going close to the attacking army and waiting for reinforcements. This was happening before and It was pretty effective for the AI to defend settlements and to avoid snowballing.

 

[Kegeyn]

Maybe change it depending on how "invested" army is in a siege? Like if it is already on siege engine building it means shortly they will start assault and take the city. Possible abandonment during siege setup stageis more logical since army is not wasting too much time yet

 

[Adijazz1706]

I have been doing some tests in the first 5 days o a new Beta 1.4.1 campaign. Here is a perfect example of what we have been talking about the AI is clearly prioritizing attack over the defense, which is a negative behavior for campaign balancing and increase snowballing and lords defection.

1- As you can see, the Western Empire funded a huge army with 1200 men and start a besiege on Marunath:

17

Image 17 hosted in ImgBB

ibb.co

2- Then I went to Lagerta and a much smaller Battania army just started a besiege on this town.

18

Image 18 hosted in ImgBB

ibb.co

3- Western Empire had not started the attack when Largerta besiege was started but anyway, the WE army just ignores the Largeta siege and continue with the Marunath siege.
4- The result was WE taking Marunath while Battania taking Lagerta. It does not look too bad, both factions have just traded a town and are similar than they were before, but then you realize that it is actually really bad for campaign balancing and lords get a penalty for losing settlements and relationship penalty with the king, which increase chances of defecting.

This new issue was introduced with lasted patches due to people complaining about AI abandoning ongoing sieges in order to defend. If I would have to choose, I prefer the old AI behavior by far, where the AI just abandoned sieges in order to defend. On the other hand, it would be great if the new behavior could be tweaked and AI just does not abandon sieges if the actual attack is ongoing for some time and the AI army will conquer the settlement soon. Is this possible to do?

Currently in beta 1.4.1, it looks like once an army start a besiege (just the besiege and not the attack to the settlement), then the army does not abandon It in order to defend which I find wrong.

Plus, smaller armies (compared to attacker force) should try to help friendly besieged settlements like they did before, just going close to the attacking army and waiting for reinforcements. This was happening before and It was pretty effective for the AI to defend settlements and to avoid snowballing.

THe rel drop has been massively decreased to a reasonable level so not a problem. Yours isnt a bad scenario as the remnants of the WE army can mop up the remaining Bat army and retake the city in a week tops considering the weak garrison that would exist in the newly taken town. Otherwise huge armies besieging a town could be forced to leave halfway because of a small enemy army besieged a castle somewhere and hardly any fiefs could be taken

 

[Dabos37]

THe rel drop has been massively decreased to a reasonable level so not a problem. Yours isnt a bad scenario as the remnants of the WE army can mop up the remaining Bat army and retake the city in a week tops considering the weak garrison that would exist in the newly taken town. Otherwise huge armies besieging a town could be forced to leave halfway because of a small enemy army besieged a castle somewhere and hardly any fiefs could be taken

This is exactly what should happen IMO, and It was like the game worked before and snowballing was under control. The problem was that sometimes armies stop attacking towns with only 7 defenders and things like that, but there is a big magin between this and the current situation.

Happily losing towns should not happen so often like It is happening now IMO. And this bring issues like lords defecting and increase snowballing.

 

[mexxico]

@Debos I examined situation at the point in your save game.

When I try your save game Ergeon's army first decided to help Druimmor castle and it did not go besiege at half way but they go defence of Lageta. I compared two situation's scores in below image :

Because Lageta is more important settlement, Ergeon's army choose defending there. If I see other decision on way to Lageta I will also compare now I will continue examining.

About siege in Druimmor there seems 2 party is on way to help there. AI knows that info. Thats why powerScore seems 0.62 not 1 because with Ergeon's Army there will be so many defenders defending that castle. So you do not see these defenders and think that army choose wrong AI. However even there are lots of defenders maybe still sending extra defenders can be logical. I will think about increasing neededDefencesMax variable so 0.62 will be closer to 1 even there are enough defenders sent to that target. Also will think of reducing effect of settlementImportanceScore at defending especially otherwise armies easily give up defending castles if there is a siege at further town. Still there should not be big changes towns are very important in terms of finance of kingoms.

Currently in beta 1.4.1, it looks like once an army start a besiege (just the besiege and not the attack to the settlement), then the army does not abandon It in order to defend which I find wrong.

There was a bug increasing recent selected behavior's score more than 1.2x as mentioned above; I fixed it maybe this helps armies to change mind more frequently. However we should be careful armies should not change idea so frequent. It will be disturbing.

Plus, smaller armies (compared to attacker force) should try to help friendly besieged settlements like they did before, just going close to the attacking army and waiting for reinforcements. This was happening before and It was pretty effective for the AI to defend settlements and to avoid snowballing.

They are still helping. Smaller parties go around besieger party and wait for reinforcements. There is no change in this behavour.

By the way if you provide more save games with weird situations I can check more cases.

 

[voronius]

@mexxico Wow, good job communicating with players like that, respect !
Sorry for offtopic.

 

[Dabos37]

@Debos I examined situation at the point in your save game.

When I try your save game Ergeon's army first decided to help Druimmor castle and it did not go besiege at half way but they go defence of Lageta. I compared two situation's scores in below image :

Because Lageta is more important settlement, Ergeon's army choose defending there. If I see other decision on way to Lageta I will also compare now I will continue examining.

About siege in Druimmor there seems 2 party is on way to help there. AI knows that info. Thats why powerScore seems 0.62 not 1 because with Ergeon's Army there will be so many defenders defending that castle. So you do not see these defenders and think that army choose wrong AI. However even there are lots of defenders maybe still sending extra defenders can be logical. I will think about increasing neededDefencesMax variable so 0.62 will be closer to 1 even there are enough defenders sent to that target. Also will think of reducing effect of settlementImportanceScore at defending especially otherwise armies easily give up defending castles if there is a siege at further town. Still there should not be big changes towns are very important in terms of finance of kingoms.



There was a bug increasing recent selected behavior's score more than 1.2x as mentioned above; I fixed it maybe this helps armies to change mind more frequently. However we should be careful armies should not change idea so frequent. It will be disturbing.



They are still helping. Smaller parties go around besieger party and wait for reinforcements. There is no change in this behavour.

By the way if you provide more save games with weird situations I can check more cases.

Oh thank you very much for this, you are the best!

Concerning more save games, I am waiting for endless wars issue being fixed before playing the game again, but I will share more save games for sure. Sounds reasonable what you say and giving towns more relevance is something good and this should not change much. Maybe the best way to do It would be taking into account distance. Not sure if armies are thinking on distance when taking decisions currently but It should be something relavant to take into account, especially when deciding to attack.

 

[mexxico]

Sounds reasonable what you say and giving towns more relevance is something good and this should not change much. Maybe the best way to do It would be taking into account distance. Not sure if armies are thinking on distance when taking decisions currently but It should be something relavant to take into account, especially when deciding to attack.

I made several minor developments while examining your save game. I give a bit more bonus to defence score over hostile actions. It was already there (armyTypeScore) but was very minor advantage. So in next versions probably it will be better. Distance is already a variable as you can see in picture I shared (distanceScore) and this score reduce faster when it comes to hostile action scoring.

 

[Dabos37]

I made several minor developments while examining your save game. I give a bit more bonus to defence score over hostile actions. It was already there (armyTypeScore) but was very minor advantage. So in next versions probably it will be better. Distance is already a variable as you can see in picture I shared (distanceScore) and this score reduce faster when it comes to hostile action scoring.

Oh yes you are right, I had missed the distance variable.

Thanks for detailed explanation, and for the changes. Looking forward the next patch ?.

 

[Major_Heartfire]

When the AI is taking into account how many allied parties are going to defend a settlement, do they look at how far away those parties are?

For example: if the Party A (AI controlled) is 12 hours away from Lageta, but sees there are two more AI allies (B and C) heading there to defend, it might decide not to go help defend Lageta, correct?

In this situation, would A decide to keep going to Lageta if B and C were super far away? Let’s say they are 4 days away.

 

[mexxico]

When the AI is taking into account how many allied parties are going to defend a settlement, do they look at how far away those parties are?

For example: if the Party A (AI controlled) is 12 hours away from Lageta, but sees there are two more AI allies (B and C) heading there to defend, it might decide not to go help defend Lageta, correct?

In this situation, would A decide to keep going to Lageta if B and C were super far away? Let’s say they are 4 days away.

Good question. Actually 4 days is huge travel time. This is something like travelling western Vlandian coast of map to Khuzait areas maybe more. Normally that further parties do not decide to defend a settlement from that much distance. So we do not take this distance into account while calculating how many defenders are already assigned. This lack of information while calculating how many parties already assigned can create problems of course. Some closer parties can give up defending because already further parties are assigned. Anyway I am increasing neededDefencesMax variable to reduce these situations. Even there are enough defenders more to send is not so big problem because as you know if there is bigger strength difference at simulation powerfull side lose less men. This can be only problem if player make fake attacks to collect parties of AI to one point.

 

[Bannerman Man]

@mexxico Hey maybe you were aware already, but there seems to be something weird going on with how the AI weighs the value of raiding Tevea. You can see examples of what I mean here, here, and here. Maybe it's just a coincidence, but it seems like there's something wrong.

Also, I posted a bug report here about the AI that causes NPCs to get trapped in a town if they are seeking food and recruits from a neutral settlement when war is declared between the two factions.

 

[Dabos37]

Hi @mexxico , can you please confirm if the changes mentioned on this thread has been already applied? I have not read anything in patch notes but I am noticing a bit less snowballing after the lasted hotfixes, but not sure if I am just having a lucky campaign or something has really changed. Thanks.

 

[mexxico]

Hi @mexxico , can you please confirm if the changes mentioned on this thread has been already applied? I have not read anything in patch notes but I am noticing a bit less snowballing after the lasted hotfixes, but not sure if I am just having a lucky campaign or something has really changed. Thanks.

Less snowballing can be related to reduce at number of wars. I have less info about final situation at "kingdoms having lots of wars" problem but it must be a bit better after hotfix.

I did developments at prioritizing defence over hostile actions but these changes are not sent yet to live or beta yet. fisrt "kingdoms having lots of wars" problem is need to be fixed otherwise 1.4.1 cannot be live and new beta includes ai developments cannot be send.

 

[Dabos37]

Ok thanks for the info, good to know that It will get better in next patch ?. And yes, I was just having a lucky campaign in the first couple of days, and now Khuzait are starting to conquest everything. It is day 130 in my new campaign and Northen Empire has gone.

 

[Dabos37]

Have these changes be applied in 1.4.2, right? I think yes because I am noticing a huge improvement for the AI and now It is able to defend settlements properly, I am just curious about why there is no mention in patch notes.

It is the best campaign AI I have seen so far since release. Please do not change It except if you want to make It better .

 

[somethinglightthat]

As far as AI defending sieges goes, what I noticed in my recent play on 1.4.2 was that every time I would siege if there was an army available the AI would show up to defend. This was so reliable that if I wanted to start a fight I could just go siege a town and it would draw the army to me, then I could just chase them down. I used this several times. A word of warning, tho, if you siege a town make sure the defenders don't outnumber you by too much, if they sally forth and defeat you then you get captured.

It was not as effective when sieging castles. The AI seemed to ignore this sometimes. This suggests to me that more weight is given to towns which is as it should be.

The AI still seems to attack far away castles for some reason. It also seems to get stuck on attacking the same fief sometimes. I hope to get around to some video and saved game files soon.

I don't think its really the AI that is the problem for Northern Empire. They are just in a bad position. All hell breaks loose around Epicrotea and then you get Khuzait from the other side, and maybe Southern Empire too. It's quite possible for them to be at war with half the world right from the start.

 

[mexxico]

Have these changes be applied in 1.4.2, right? I think yes because I am noticing a huge improvement for the AI and now It is able to defend settlements properly, I am just curious about why there is no mention in patch notes.

It is the best campaign AI I have seen so far since release. Please do not change It except if you want to make It better .

I think we forget to mention it. Yes we improved campaign AI according to your feedbacks. I am glad you liked new more defensive AI.