AI getting stronger I think

Koshling

Vorlon
Joined
Apr 11, 2011
Messages
9,254
Well, I just got utterly crushed in me last game. Admittedly I was unlucky with resources (no metals anywhere near), but normally I can survive that with good use of great generals and mounted units. This time however, I really got crushed. Couldn't keep up technologically, and a surprisingly effective AI SoD smashed through a one tile bottleneck where I had 7 or 8 units with a great commander in a fort on a forested hill, which I thought was utterly secure.

This was a game started as Immortal and increased to Deity on turn 2. That's not a setting I've ever really had a problem winning from eventually for the past year or so at least. Finally starting to see AI work paying off - V29 will be a step up from V28 (I'd say about one difficulty level worth)
 
and at the same time i have the most easiest game i've ever had >.> but again i got really lucky and gained 2 Tribes from Goody huts (1 i used while i was at the 6th tech orso) ruined my Economy for a couple of turn's got Capital up it's feet got Hunters out and now i'm at the begin of the Second era no AI has build any wonder as only 1 of the 3 i started with (i was hoping i would get an easier start while a late game fun run with BarbCiv's) but the only problem i had was 1 Barb stack that could have became much more of an issue then they did (probably their odd's on my city was crap Hill+river) and so they where driven into the desert and they all died there as i could only enforce their retreat up to the border of the desert my "standing army" isn't capable of negotiating that terrain without damage
 
and at the same time i have the most easiest game i've ever had >.> but again i got really lucky and gained 2 Tribes from Goody huts (1 i used while i was at the 6th tech orso) ruined my Economy for a couple of turn's got Capital up it's feet got Hunters out and now i'm at the begin of the Second era no AI has build any wonder as only 1 of the 3 i started with (i was hoping i would get an easier start while a late game fun run with BarbCiv's) but the only problem i had was 1 Barb stack that could have became much more of an issue then they did (probably their odd's on my city was crap Hill+river) and so they where driven into the desert and they all died there as i could only enforce their retreat up to the border of the desert my "standing army" isn't capable of negotiating that terrain without damage
The difficulty levels in which you can get tribes from goody huts don't really count (that goody hut outcome should probably just be removed).
 
New players to CIV and BtS like the 1st 2 Difficulty levels just for the chance to get a free settler/tribe.

And since C2C is like an Expansion to BtS, more than a Mod, they probably would like to keep that option alive.

JosEPh
 
@Koshling: Does this mean that you are more or less done with AI work for this cycle? Because given that Primem0ver says that GeoRealism will be ready for merge later this month and things have slowed down a bit recently for development I think it might be nice to start the freeze soon (middle of next week maybe?).
 
@Koshling: Does this mean that you are more or less done with AI work for this cycle? Because given that Primem0ver says that GeoRealism will be ready for merge later this month and things have slowed down a bit recently for development I think it might be nice to start the freeze soon (middle of next week maybe?).

I am still shooting for 15 Mar, because DH has some stuff, that needs to be done yet, and i am hoping that DH or AIAndy has looked at some of the stuff i wanted to get in before the freeze.

And YES, teh AI is 10 times better this version, the way it belongs, but the WORKER still needs attention, as i mentioned in the Bug thread.

Plus NO one has commented on the Sub-Combats for Heroes, and what to change those Civic_Slavery posts yet??
 
@Koshling: Does this mean that you are more or less done with AI work for this cycle? Because given that Primem0ver says that GeoRealism will be ready for merge later this month and things have slowed down a bit recently for development I think it might be nice to start the freeze soon (middle of next week maybe?).

I'm almost done with the beat commanders ai work, but I haven't done anything about surround and destroy yet. Given the deadline, I think I will leave that for the next version though.
 
and at the same time i have the most easiest game i've ever had >.> but again i got really lucky and gained 2 Tribes from Goody huts (1 i used while i was at the 6th tech orso) ruined my Economy for a couple of turn's got Capital up it's feet got Hunters out and now i'm at the begin of the Second era no AI has build any wonder as only 1 of the 3 i started with (i was hoping i would get an easier start while a late game fun run with BarbCiv's) but the only problem i had was 1 Barb stack that could have became much more of an issue then they did (probably their odd's on my city was crap Hill+river) and so they where driven into the desert and they all died there as i could only enforce their retreat up to the border of the desert my "standing army" isn't capable of negotiating that terrain without damage

V28 or SVN? I'm talking about the SVN, which is noticeably improved over V28.
 
I'm almost done with the beat commanders ai work, but I haven't done anything about surround and destroy yet. Given the deadline, I think I will leave that for the next version though.

I'm willing to move back the deadline to give you time for SAD AI, till the 15th or so. Then we could release on the 24th.
 
S&D is a much, MUCH harder problem than great commanders. It's much more than an extra week. It can wait.
 
Oh well, guess we'll see it in v30 then.
But from what koshling said, the ai is already going to be more... violent in v29 XD I can't wait...
 
Oh well, guess we'll see it in v30 then.
But from what koshling said, the ai is already going to be more... violent in v29 XD I can't wait...

Believe you ,me, it is back to the very basic's of BtS again, the way it was intended to be played(only better with C2C:)
 
I'm almost done with the beat commanders ai work, but I haven't done anything about surround and destroy yet. Given the deadline, I think I will leave that for the next version though.

Personally I would like to see you forgo surround & destroy completely, I imagine it'd be hard work and I feel like you'd better off teaching the AI some useful little things instead - like blockading with ships, pillaging with rogues/thieves etc, even something like using transport ships to drop troops behind enemy lines for multipronged attacks, teach them the value of seige weapons ;)
Probably a lot of others I can't think of right now, but you get my point. I hope you don't think im trying to tell you what to do its just my opinion. Depending on the degree of difficulty and amount of time required, i'd like to see you just attack it generally to see what improvements can be made to AI for the smallest time investment kinda thing.

Who really cares if S&D works or not honestly? I personally would just prefer you to remove the damn thing :sniper:
Trying to fix it so that AI understand and can use it will probably just cause additional problems for the AI. Perhaps these problems might not show up immediately, but deviously hide in the background somewhere hindering the AI for years to come. Totally not worth the risk for such a sh1tty feature imo. I remember you said yourself the AI strategy generally revolves around SoD, and you only just recently managed to optimize the AI's stack usage. Now your considering possibly undoing some of this good work. I just hate the thought of it to be honest, but hey that's only my 2c on the matter.
 
Personally I would like to see you forgo surround & destroy completely, I imagine it'd be hard work and I feel like you'd better off teaching the AI some useful little things instead - like blockading with ships, pillaging with rogues/thieves etc, even something like using transport ships to drop troops behind enemy lines for multipronged attacks, teach them the value of seige weapons ;)
Probably a lot of others I can't think of right now, but you get my point. I hope you don't think im trying to tell you what to do its just my opinion. Depending on the degree of difficulty and amount of time required, i'd like to see you just attack it generally to see what improvements can be made to AI for the smallest time investment kinda thing.

Who really cares if S&D works or not honestly? I personally would just prefer you to remove the damn thing :sniper:
Trying to fix it so that AI understand and can use it will probably just cause additional problems for the AI. Perhaps these problems might not show up immediately, but deviously hide in the background somewhere hindering the AI for years to come. Totally not worth the risk for such a sh1tty feature imo. I remember you said yourself the AI strategy generally revolves around SoD, and you only just recently managed to optimize the AI's stack usage. Now your considering possibly undoing some of this good work. I just hate the thought of it to be honest, but hey that's only my 2c on the matter.

You're joking, right? When I fight wars, they really seem like proper invasions. The combat in this mod really shows how tactics can be accomplished without any sort of game-breaking 1UPT solution.
 
Personally I would like to see you forgo surround & destroy completely, I imagine it'd be hard work and I feel like you'd better off teaching the AI some useful little things instead - like blockading with ships, pillaging with rogues/thieves etc, even something like using transport ships to drop troops behind enemy lines for multipronged attacks, teach them the value of seige weapons ;)
Probably a lot of others I can't think of right now, but you get my point. I hope you don't think im trying to tell you what to do its just my opinion. Depending on the degree of difficulty and amount of time required, i'd like to see you just attack it generally to see what improvements can be made to AI for the smallest time investment kinda thing.

Who really cares if S&D works or not honestly? I personally would just prefer you to remove the damn thing :sniper:
Trying to fix it so that AI understand and can use it will probably just cause additional problems for the AI. Perhaps these problems might not show up immediately, but deviously hide in the background somewhere hindering the AI for years to come. Totally not worth the risk for such a sh1tty feature imo. I remember you said yourself the AI strategy generally revolves around SoD, and you only just recently managed to optimize the AI's stack usage. Now your considering possibly undoing some of this good work. I just hate the thought of it to be honest, but hey that's only my 2c on the matter.

The thing that makes S&D hard is also why some of the other things you mention would be hard, so it's the common underpinning that needs tackling sometime anyway. Specifically, the underlying issue is that the AI has absolutely no concept of central control, or really of cooperation. In the original BTS AI, every stack makes totally independent decisions about what to do, so essentially it's locally greedy but has no concept of a bigger picture.

For S&D this manifests in a few ways:

1) If there are multiple stacks in an area, which should perform attacks and which should take up surrounding positions to aid another stack's attack?
2) When should an attack happen - better to do it immediately, or to wait for some other stacks to get to surrounding positions (even on the same turn that is)?
3) Since the AI is very SoD-oriented, when to break up a large stack to allow independent movement (and then when to put it back together also)?

However, for things like a pincer attack from a different direction, or sensibly coordinated pillaging to support an attack, much the same considerations would apply - basically anything that requires more than one stack to coordinate activity falls into this category, and currently the AI has no good mechanism to allow for this.

Since about V26 I've been making changes to alleviate this situation, mostly with the unit contracting mechanism, through which in C2C now:
  • workers, settlers, great generals now advertise actively for escorts when needed, and units without high priority work advertise themselves as available (the contracting system then optimizes the hook ups based on how well units meet the requested criteria and how easy it is for them to get there)
  • great commanders actively contract for hunters to stack with, in order to go get early experience
  • cities contract for defenders, rather than always building them for themselves (cities without high priority builds can also respond to contracts by building units, and in that case the contract broker picks the best placed city, based on the unit it can make, the promotions it would provide, and the distance it is from where it is needed)
  • City attack stacks (aka SoDs) actively advertise for attack units (always), healers (if they have none), and a great general (if they don't have one) to join them and keep doing his until they are ready to roll and have left home turf

All of the above (apart from city defenders which the cities always used to just build) used to be handled exclusively by each stack looking around its local area for something to do/someone to join (so SoDs would only build up as quickly as attack/city attack units happened to move close enough to see them and decide to join them, and so on). This has been a big step forward (and the same system almost certainly has further uses), but we need to go a step further to achieve S&D (and other cooperative action) AI. I have something in mind, but right now it's only an idea, and it's going to be quite complex to implement and take a long time of seeing how it behaves and tweaking in response to that to really make it work well.
 
The thing that makes S&D hard is also why some of the other things you mention would be hard, so it's the common underpinning that needs tackling sometime anyway. Specifically, the underlying issue is that the AI has absolutely no concept of central control, or really of cooperation. In the original BTS AI, every stack makes totally independent decisions about what to do, so essentially it's locally greedy but has no concept of a bigger picture.

For S&D this manifests in a few ways:

1) If there are multiple stacks in an area, which should perform attacks and which should take up surrounding positions to aid another stack's attack?
2) When should an attack happen - better to do it immediately, or to wait for some other stacks to get to surrounding positions (even on the same turn that is)?
3) Since the AI is very SoD-oriented, when to break up a large stack to allow independent movement (and then when to put it back together also)?

However, for things like a pincer attack from a different direction, or sensibly coordinated pillaging to support an attack, much the same considerations would apply - basically anything that requires more than one stack to coordinate activity falls into this category, and currently the AI has no good mechanism to allow for this.

Since about V26 I've been making changes to alleviate this situation, mostly with the unit contracting mechanism, through which in C2C now:
  • workers, settlers, great generals now advertise actively for escorts when needed, and units without high priority work advertise themselves as available (the contracting system then optimizes the hook ups based on how well units meet the requested criteria and how easy it is for them to get there)
  • great commanders actively contract for hunters to stack with, in order to go get early experience
  • cities contract for defenders, rather than always building them for themselves (cities without high priority builds can also respond to contracts by building units, and in that case the contract broker picks the best placed city, based on the unit it can make, the promotions it would provide, and the distance it is from where it is needed)
  • City attack stacks (aka SoDs) actively advertise for attack units (always), healers (if they have none), and a great general (if they don't have one) to join them and keep doing his until they are ready to roll and have left home turf

All of the above (apart from city defenders which the cities always used to just build) used to be handled exclusively by each stack looking around its local area for something to do/someone to join (so SoDs would only build up as quickly as attack/city attack units happened to move close enough to see them and decide to join them, and so on). This has been a big step forward (and the same system almost certainly has further uses), but we need to go a step further to achieve S&D (and other cooperative action) AI. I have something in mind, but right now it's only an idea, and it's going to be quite complex to implement and take a long time of seeing how it behaves and tweaking in response to that to really make it work well.

Hold on. This shouldn't be true without Barbarian Generals active, should it? :confused:
 
Hold on. This shouldn't be true without Barbarian Generals active, should it? :confused:

You're right, it should. I'll make that change shortly, so they'll just stay home until needed fro 'real' duties if barb generals is not turned on (at least for now - if I do more work on it they should probably sit in a defensive position on a border hoping to pick up xp from attacks)
 
Back
Top Bottom