Game AI discussions

[BattleGoat Studios]

Pulling this out of another thread that veered off topic, so that we can respond to some of the questions/comments:

In SRCW ship AI cant even plan its route around land, Ships go directly to their designated target untill they stumble into land, then then recalculate route and usually then drag along shoreline untill they reach their destination, which is usually far from optimal route. That especially noticeble in indonesia region with lots of islands and peninsulas.

While pathing around the shape of the real world has its challenges, for most routes the naval pathing takes the best routes based on the criteria it is designed to use. Such criteria include a desire to use supplied water areas, to stay out of hostile waters when possible, to generally stay close to coastlines, etc. There are also some technical reasons why very long or tricky paths are approximated, sometimes resulting in strange bottlenecks, because it would slow down the game to use more precise pathing. Remember that we allow thousands of units to be moving and acting simultaneously, and each unit is checked 1440 times per game day (once per "minute" of world time). The SR game engine has the most efficient and accurate pathing code of any game that features a map of this size, combined with the fact that the pathing code is "generic" (not hard-wired to the world, but allowing mods etc).

If you think it might be desirable to offer a game option for more accurate long-distance pathing at the cost of game performance, we can look at adding this as a game option for players.

Another thing about AI being terrible is how it moves units chaotically (which partly responsible for bad performance). Basically AI just mobilizes all/most of their units and moving them around aimlessly while burning country fuel reserves. While player can take manual control over his units to prevent it, same cant be said about all other AI controlled nations. So unless player plays a minor indipendent nation without allies or shared visibility, he ends up observing a a mess of hundreds of flags/units aimlessly wondering around. And before you say that "AI is patrolling/repositioning/doing whatever" - sorry, I dont buy it. Argentinian troops have no reason to patrol in united kingdom, as well as Norwegian navy doesnt need to patrol coasts of USA.

A couple of things here - first of all, the performance is not impacted by whether you see units moving or the movements are hidden by fog of war, in the Supreme Ruler games all units operate by the rules of the game, no AI cheating, so they keep moving/looking/acting whether you see them or not.

The reason the AI (and ministers with initiative) move units so much is that there's a number of things going on:

- Units are rotated in and out of reserve to keep the efficiency level of the units up. Units left in reserve will lose efficiency over time, and this takes time to recover once units are deployed.
- The AI is trained to be suspicious, so it will move recon and mobile units (infantry, tanks) around as both defensive protection and recon to watch for surprise aggression.
- Individual units also have a unit-level AI that will keep them moving and patrolling.

The question of why the Norwegian navy would be patrolling US waters, the AI does take the concept of "alliances" very broadly. If Norway has an alliance or other defense/transit agreements with the US, then they will patrol there, and the US will help in Norway. This has been pointed out in other threads as being too common and we will be cutting back on it when regions are not at war - although joint operations among NATO countries or (when it existed) Warsaw pact countries were quite common.

... which in case of war just takes all its units (except maybe few sitting in garrisosns) and sending the blob at general direction of the enemy, separatly, without any resemblance of cohesion or anything?

The AI does co-ordinate and plan attacks, sometimes more than others, but the among other things it tries not to be repetitive or predictable. it does neither "blob" nor "separate", but may often be a combination of both.

Do you consider a good diplo AI for geopolitical game in 2014, which will not accept a deal of f.ex. mutual protection+free trade, but will accept free trade+mutual protection during the same negotiations?

This bug has been fixed a while back.


In general, we continue to work on and expand the AI, and Naval AI improvements have been highlighted for improvements during the Ultimate Early Access period, so we welcome player ideas and comments. One issue about all of the Supreme Ruler games is that the AI does so many things that most players don't even notice most of them, or mis-understand what the AI was trying to accomplish. This is both a problem, of course, but at the same is also a bit realistic, since in wartime one region often doesn't see or understand what the goals of the other are. Gameplay wise it's not always ideal though.

-- George / BattleGoat.

[Leo3ABP]

Originally posted by BattleGoat Studios:

Pulling this out of another thread that veered off topic, so that we can respond to some of the questions/comments:

While pathing around the shape of the real world has its challenges, for most routes the naval pathing takes the best routes based on the criteria it is designed to use. Such criteria include a desire to use supplied water areas, to stay out of hostile waters when possible, to generally stay close to coastlines, etc. There are also some technical reasons why very long or tricky paths are approximated, sometimes resulting in strange bottlenecks, because it would slow down the game to use more precise pathing. Remember that we allow thousands of units to be moving and acting simultaneously, and each unit is checked 1440 times per game day (once per "minute" of world time). The SR game engine has the most efficient and accurate pathing code of any game that features a map of this size, combined with the fact that the pathing code is "generic" (not hard-wired to the world, but allowing mods etc).

If you think it might be desirable to offer a game option for more accurate long-distance pathing at the cost of game performance, we can look at adding this as a game option for players.

All fluff aside, its for the 99% looks like being purely technical problem. First of all, doing pathing check every ingame second is definatly an overkill and a waste of cpu resourses. Considering that map is static and no unit ingame moves that fast to reach next tile in 1 second of ingame time, there is no reason to do so much polling for movement (or infact everything else. It can be either minute or even quarter/half of an hour or hour based, or in case with movement - tile based polling (i.e. unit gets polled for pathing after reaching new tile).
Also stating that SR has most efficient and accurate pathing code is extremely bold, especially in a current state of things.
It is desirable to have game optimization and less unnessesary usage of computer resources, then more accurate pathing can be added without performance loss.

Originally posted by BattleGoat Studios:

A couple of things here - first of all, the performance is not impacted by whether you see units moving or the movements are hidden by fog of war, in the Supreme Ruler games all units operate by the rules of the game, no AI cheating, so they keep moving/looking/acting whether you see them or not.

The reason the AI (and ministers with initiative) move units so much is that there's a number of things going on:

- Units are rotated in and out of reserve to keep the efficiency level of the units up. Units left in reserve will lose efficiency over time, and this takes time to recover once units are deployed.
- The AI is trained to be suspicious, so it will move recon and mobile units (infantry, tanks) around as both defensive protection and recon to watch for surprise aggression.
- Individual units also have a unit-level AI that will keep them moving and patrolling.

The question of why the Norwegian navy would be patrolling US waters, the AI does take the concept of "alliances" very broadly. If Norway has an alliance or other defense/transit agreements with the US, then they will patrol there, and the US will help in Norway. This has been pointed out in other threads as being too common and we will be cutting back on it when regions are not at war - although joint operations among NATO countries or (when it existed) Warsaw pact countries were quite common.

I was expecting to get some refference about real world joint operations, unfortunately we are talking about game here, and there are no concept or benefit of joint operations during peace time in it, so I suggest leaving all future refference to real world out of discussion. All fluff aside, it generally can be said that is AI more or less randomply distributes its unints according to alliances it has, which again gives it no merit and sets it further into "poor AI" department, and hurts small and poor countries gameplaywise.
As for AI "being suspicious", I would accept if AI was setting up units in some form of orginized facion across borders/chokepoints/defencive terrain as a responcse to a possible future threats from its neibourghs, but unfortunately thats not the case. For the most part AI is busy patrolling its domestic territory far from any borders or threats, thats again is especially noticeble if you play NATO or WP country. Sometime AI literally moves back and forth between two tiles.
Also, optimization wise (considering 1440 movement pollings per day for each unit) units that are mobilized for their effeciency increase dont have to move back and forth, buning fuel and hogging computer resources. If it is such nessesary feature to mobilize units back and forth to keep them effecient (which imo is a sloppy way to represent combat readyness optimization-wise), then at least those units should be stationary on military bases, not moving back and forth between two proverbial tiles, while hogging cpu resourses.

Originally posted by BattleGoat Studios:

The AI does co-ordinate and plan attacks, sometimes more than others, but the among other things it tries not to be repetitive or predictable. it does neither "blob" nor "separate", but may often be a combination of both.

Sorry, but I can only rephrase it as "AI just sends units randomly towards the enemy", which is unfortunally is the case ingame. Look for example at Hearts of Iron 3. Individual units there try to cooperate and act as a bigger army, maintaining front lines, attempting breakthroughs and encirclments, making orginized retreats etc, AI there while not perfect, does all this fine more or less. In SR AI just sends units in general direction of his enemy, without attempt for any orginized cohesion or orginized actions between separate units (unless u consider a blobbing an orginized actions).
Also there are cases where AI ends up being overly passive for no reason. In my last game as West Germany, Laos declared war on North Vietnam, and france intervened (as usual). France managed to repel NV blob and took initiative and started capturing territory, but at some point entire french army just stopped. There were no resource or unit shortage there or any other reason to squash NV which at that point had like 3-4 units left excluding garriosns. It took me sending 2 motorized Inf units there and shooting at NV garrison once to get entire french army moving and winning this war. Thats obviously an AI issue, since there was no reason for france to stop and wait my mot inf which added little to nothing to that war, except making a short attak and retreating.

Originally posted by BattleGoat Studios:

Do you consider a good diplo AI for geopolitical game in 2014, which will not accept a deal of f.ex. mutual protection+free trade, but will accept free trade+mutual protection during the same negotiations?

This bug has been fixed a while back.

[/quote]
In which iteration of the game? Coz its obviously still in SRCW. I was playing it a few days ago and was getting it. Infact all described bugs are still ingame and were experienced during last weeks when I tried to play SR again, to see whether anything has imporved at all. Apparently nothing did.
So while I might have sounded harsh in my previous posts, that was for the large part for this exact reason - BG claiming that certain bugs were fixed, while they actually were not, so please, understand me when I express my highest dissapointment in what BG says on forums.

Originally posted by BattleGoat Studios:

One issue about all of the Supreme Ruler games is that the AI does so many things that most players don't even notice most of them, or mis-understand what the AI was trying to accomplish. This is both a problem, of course, but at the same is also a bit realistic, since in wartime one region often doesn't see or understand what the goals of the other are. Gameplay wise it's not always ideal though.

Again, plese, dont come up with "real life" comparisons, we are talking about a game here, which as any game has lot of restrictions and mechanics tied to it. One of indications of good 4X/Geopolitical/Grand Strategy game is that while their mechanics are complex and can stay in the background for the casual player, they are still open and visible for players that want a deeper understanding how game works. Look at Hearts of Iron/Europa Universalis/Crusader Kings again, it is possible to play them without attemtping to comprehend their mechanics and just play them using a common sense, but they also do show exact numbers and % and whatnot for ppl who wants more indepth experience with their game mechanics.
So saying stuff like "oy we have a hidden complicated perfectly working mechanics which you plebs just cant see nor understand" when something does not work the player expects it to, is a very sloppy game design (or just an excuse for bugs) and again gives developer no merit.


Hopefully our future discussion will stay polite and constructive and there will be less flaming, bad feelings and ignores (chris Im looking at you <.<), thread and posts deletes etc. as soon valid criticism arises and questions start to get uncomfortable.

"The only way to selfperfection is to relize that you are not perfect"(c) just made this quote up

Have a nice day.

[Chris-BattleGoat]

I'll let George reply in more detail, but some things I noticed;

<...pathing check...map is static>
No it's not, transit rights can change at a moment's notice and have a big impact on how air or land units move from A to B. Since most naval travel is in international waters this is less of an impact but the status of the Suez or Panama could change at any time. Only the terrain is static, not the other factors.

<...In which iteration of the game? Coz its obviously still in SRCW>
The bug where offering different treaties in a different order created different results has been fixed in SRCW for months. It was fixed before SR1936 went into early access and there have been no reports since then. If you've seen it since then, you haven't started a thread about it. If you know of a way to cause it return different results, please start a thread and explain how to recreate the issue.

<... plese, dont come up with "real life" comparisons...>
While you judge real life comparisons not to be valid, we disagree and see them as quite valid. You can expect that we will continue to cite them.

[Leo3ABP]

Originally posted by chrisahl:

I'll let George reply in more detail, but some things I noticed;

<...pathing check...map is static>
No it's not, transit rights can change at a moment's notice and have a big impact on how air or land units move from A to B. Since most naval travel is in international waters this is less of an impact but the status of the Suez or Panama could change at any time. Only the terrain is static, not the other factors.

Suez and Panama can be more or less qualified for "dynamic" changes coz of being chokepoints, but were are talking 2 (two!) cases of what technically is "terrain becoming impassable". It does not need 1440 pathing polls to do pathing around as soon as those tiles get impassable status nor is there any extraordinary pathfinding algoritms requred to overcome this, every strategy game does this nowadays, sorry we nolonger in 1989.
As for the transit rights in territorial waters:
first of all sloppy AI pathfinding and ships bumping into land is observed in both territories with and without transit rights, and is tied to a shore line rather than anything else.
Secondly, and you said it yourself, most of shipping occures in international waters and transit status has very little impact.
Last but not least, we are talking about a tile based game with fairly static basic rules for tiles (even taking into account transit rights) which are first and foremost whether tile is passable or not, game does not have to take into account things like object dimentions, speed, physics etc, so sorry when Im not buying when I hear that pathing in SR is extremely comlicated an sophisticated matter, coz it is not. 2D tiles and lack of any other basic properties for objects traversing them is probalby easiest enviroment to make pathfindin in.

Originally posted by chrisahl:

<...In which iteration of the game? Coz its obviously still in SRCW>
The bug where offering different treaties in a different order created different results has been fixed in SRCW for months. It was fixed before SR1936 went into early access and there have been no reports since then. If you've seen it since then, you haven't started a thread about it. If you know of a way to cause it return different results, please start a thread and explain how to recreate the issue.

I guess lack of reports could also mean that with small playerbase as it is (and even less ppl actually bothering to follow game on forums/send reports), nobody bothered to report or play your game at that point after latest patch/patches.
Sarcasm aside, I think I was clear enough on the issue in previous posts, that if you bothered to read them ofc. As of 3 days ago the issue with different order of deals netting different results stands.
For example AI will almost never accept formal alliance, but if you offer all parts of alliance separately (i.e. mutual defence, tranist, etc) and put formal alliance in the end, AI will accept.
Another example can be is if you put free trade, free labour and criminal extradition and then place mutual defence or roket defence in the end, AI will not accept, put mutual defence/roket defence in the beggining of list of deals and AI will accept the offer. Am I clear enough now?

Originally posted by chrisahl:

<... plese, dont come up with "real life" comparisons...>
While you judge real life comparisons not to be valid, we disagree and see them as quite valid. You can expect that we will continue to cite them.

Considering how limited is SR (or any game infact) to real life, that staement cant be taken seriously. I would partly agree if we were talking about certain intended game mechanics that attempt to simulate real world, but in case with "joint operations" its just an excuse for AI randomly distributing units over its allies. Again, if there was a "joint ops" mechanic ingame, where player could sent part of his military forces to another country for joint ops excersise, which would net things like improved/redced relations with certain countries, additional bonuses to unit effeciency/experience etc, I would only be happy. But excusing the poor AI by making up a feature that not exist in the game and claiming it is realistic - thats just childish and looks like fanboyism more than anything else. Needless to say thats not the way an adult and professional developer should be behaving in my opinion.

[Chris-BattleGoat]

<...but if you offer all parts of alliance separately (i.e. mutual defence, tranist, etc) and put formal alliance in the end, AI will accept.>

No, it won't. We fixed this. If you have proof of this still being broken, zip up the savegame and email it over. [e-mail]. Same for the order of the other treaties. As I said, this has been fixed for months.

<Sarcasm aside>

That would be nice since you suggested future discussion stay polite and constructive.

[Leo3ABP]

Originally posted by chrisahl:

<...but if you offer all parts of alliance separately (i.e. mutual defence, tranist, etc) and put formal alliance in the end, AI will accept.>

No, it won't. We fixed this. If you have proof of this still being broken, zip up the savegame and email it over. [e-mail]. Same for the order of the other treaties. As I said, this has been fixed for months.

<Sarcasm aside>

That would be nice since you suggested future discussion stay polite and constructive.

Sent you an email with save file along with following description:

Save file contains a freshly started game as West Germany

Contact Norway and offer Formal Alliance - it gets rejected (indicator red in diplo screen)
Do same offer, but add a non agression pact (which is INCLUDED into formal alliance treaty) prior to offering Forma Alliance - it gets accepted.

Portugal, same as norway, but you also need to add full transit rights (which also INCLUDED into Formal Alliance) prior to offering Formal Alliance.

Denmark, South Korea, Australia net the same result.


While I didnt managed to replicate an exact cases (EDIT Or more likely havent spent enough time trying /EDIT) where order of offers which are not part of Formal Alliance change the outcome of deal in this save, Im fairly certain that I encountered this diploAI behavior on my other recent save where I progressed a few years into the game. I am unwilling to dedicate unreasonable amount of my free time with a new game just to prove you that bug exist, and my older save has most alliances up and running and Im unsure how helpful it will be for our current problem, nor do I willing to invest lot of time into playing it either simply for the sake of proving something on internet forum. Anyhow, I think the issue with Formal Alliances I describe above is already enough to receive attention. Please notify me via steam forum if you want that older save as well.


On a personal note, Im not expecting SRCW to recieve any patching now. Im not naive and realize that BG wont bother with patching their old product that provide no income anyway, but at least be honest with yourself and admit that SRCW was never polished and had plenty of issues left when you abandoned it and moved over to SR36 and now rushed to SRU before HoI4 will ruin any chance of sales for you.

[BattleGoat Studios]

Regarding the diplomatic trading problem discussed, Chris misspoke a bit - the issue was fixed in SR1936, not SRCW. It does exist in all versions of SRCW. It does not exist in SR1936 or in Ultimate.

Regarding your comment about SRCW being abandoned, we provided free updates for it TWO YEARS after it's initial release. The diplomatic trading problem was considering low priority because it was player triggered - it was more of an exploit than a bug. If you didn't want to experience the "bug", simply don't add elements that you already know were part of the original package of treaties. Because of that, we prioritized other feature requests and new content for the SRCW updates that we pushed.

Regarding your feedback on the elements such as pathing, there have been detailed discussions of some of the details involved in threads on our battlegoat forum ( www.bgforums.com ), so I would direct you to there if you'd like more background on why we made certain choices in how the AI works. If you know of another game that has maps of this size (ours is 1850 by 750 unique map locations) that can do real-time pathing I'd be happy to take a look at the comparison.

-- George / BattleGoat.

[Leo3ABP]

Originally posted by BattleGoat Studios:

Regarding the diplomatic trading problem discussed, Chris misspoke a bit - the issue was fixed in SR1936, not SRCW. It does exist in all versions of SRCW. It does not exist in SR1936 or in Ultimate.

Regarding your comment about SRCW being abandoned, we provided free updates for it TWO YEARS after it's initial release. The diplomatic trading problem was considering low priority because it was player triggered - it was more of an exploit than a bug. If you didn't want to experience the "bug", simply don't add elements that you already know were part of the original package of treaties. Because of that, we prioritized other feature requests and new content for the SRCW updates that we pushed.

Regarding your feedback on the elements such as pathing, there have been detailed discussions of some of the details involved in threads on our battlegoat forum ( www.bgforums.com ), so I would direct you to there if you'd like more background on why we made certain choices in how the AI works. If you know of another game that has maps of this size (ours is 1850 by 750 unique map locations) that can do real-time pathing I'd be happy to take a look at the comparison.

-- George / BattleGoat.

TWO YEARS is not that much as you make it look, Blizzard still providing support to Starcraft 1 even today, and they not only limited to bugfixing. Few years ago they released a BALANCING patch!!! Balancing the game which was over 10 years old at that point. While very few companies provide such long support for their products, still you are not correct by pointing 2 years support time as something extraordinary and worth of being proud of. IMO it should be normal practice to provide support and bugfixing to software untill it polished and bug free, which unfortunately wasnt the case with SRCW.

While 1387500 tiles may look impressive, the truth is if there was 10 times more tiles, or 10 times less, nothing would change. Again I must say that programming pathfinding in 2D enviroment where tiles basically split into passable/not passable and then divided between 2 types of units (land and naval, Im not taking planes here for obvious reasons) is not an impossible or extremely difficult task. For land units main pathing obstacle being rivers, which are just a fraction of 1387500 tiles mentioned above, the rest is passable (unless water). Even simplier for navy for obvious reasons.

First game that springs to mind would be again Hearts of Iron 3, though there are "only" over 10000 tiles there iirc. While it doesnt look as impressive as 1387500, its not the amount of tiles that make huge difference, as i described above and will further describe below.

Some food for thought: To optimize further, pathfinding can first poll to check whether unit and intended destination are connected with landmass (in case of island/different continents). Instead of drawing a path and wasting cpu resources, parts of map can be defined into zones, and zones can be defined whether they connect with other zones. That would be most effecient for iland heavy regions ofc, but will also help when transporting troops between theaters. So upon order, pathfindingAi first checks if its current zone is connected with destination, and only then start polling a route there, freeing up some unnessesary calculations.
Another, and imo one of most important, thing to optimize would be to get rid of per-ingame-second polling. Its definatly an overkill and just eats up unnessesary resourses, especially considering how many units there are on map at times.

Anyhow, I find it strange to explain such obvious things to an experienced developer who has multiple commercial games behind his belt, and I do feel like doing your work for free here.


Thanks for inviting me to BG forums, but I think I will pass. Im not much of a forum person, and only rarely enter into forum argument, coz tbh i consider it a waste of my time for the most part. As somebody wrote once: "winning argument on internet forum is like winning in paraolimpycs - you won but you are still retarded" ( while retarded might be not best word, I kept it for a sake of quotation), though as you could see I do have rare urges sometimes to participate in internet discussion.
Also I think we have talked over most of the things at least i intended, and I dont see much left to talk about. Over the course of our conversations, i might have sounded harsh sometimes, but for the most part it was because I care for the game (there are very few geopolitical games out there, even less good ones) and want it to be better instead of just getting halfassed replies that either misinforming, aimed for a "wide user audience" or have little to do with the subject. So no hard feelings.
I really hope SRU will be a good game and you will polish it, unlike previous titles, and Chris will spend less time arguing on forums and more time writing tooltips, and I wish you luck in that endevour. But may curse of bancupcy befall on you if you make SRU in similiar state SRCW is :)

Anyhow, good luck and remember, I will be watching :)

[Storm_Chasee]

Since this thread is on AI behavior and this game look like it was derived from SR1936, let me describe a couple behaviors that made me gag.

1. France not putting any ground units to defend the Maginot line right before WW2 starts. It's one thing for the AI to do that when I'm playing Germany on a very easy difficulty level, but not when I'm playing US or Brazil playing an easy difficulty. I played 4 or 5 games (I had the FOW off so I could see what the AI was doing.) and not once did France man the Maginot Line or survive to 1940.

2. Turning units into merchant marine and sailing into known well defended ports. I blew away about 1,000 German and Italian battalians when playing the US in one game. The AI shouldn't be doing that at even easy difficulty level. AI was also doing that to my units too even after I had military priority of where it was sending them to 'none'. On any intiiative the AI shouldn't send merchant marine to a zone with no priority. I wound up having to set the intiative to none to keep the AI from things I definately did not want to do.

One thought is to have a limited amount of merchant that a player or AI could use at any time that could be increased by building more.

Also AI appears totally unwilling to scrap old equipment particularly planes. Again playing as the US it was frusterating that UK wouldn't scrap the old garbage planes to use the resources to build new ones. Also having all those planes were in my way so I couldn't bring over more air power and use it effectively against the Germans.

This may be a function of difficulty level but the AI's navies sat and did nothing.

[Strategy Informer]

I feel the game AI is totally broken beyond repair. Look I want to love the games, I own them all. Heck i'll most likely pick this up as well. But explain to me how someone can win the space race and eliminate the US as playing Ethopia, with very marginal strategy.. (Yes I am reffering to Kelar). This shows and proves that the AI is incapabile of reacting to the player on a strategic level. As well as underlying that the game machanics are complety broken. There is no way that this should be possible in the simulation...

36 has these issues as well.. Itialy destroying russia and such. Asymetry is a good thing you know... if your going to do a ballance everyone to be able to do anything, it should be a scenario specificly designed to do that, other wise you get these situations that are absurd.

Sure its fun to eat the world, and some people love taking there nation and taking over the map, thats awesome, but are we playing Risk?

I have to agree with the older post, SRCW was well.... how do I put it... Made Baby Jesus cry from a strategic standpoint. I'd like a chalange not some artificial buffing of the AI... heck I'd just like to see something resembling realism.

Yes please explain the decisions that allow a game that is supposed to have such death enable Ethipoia to win a space race, let alone a succesfu invasion of the US in the 70's...
Explain to me why the AI durring a sucesful invasion will just go brain dead and sit.
Explain to me how SWU will be better?