Basic things Battle AI should know...

[JaM]

Strongly recommend reading this article:

http://l-clausewitz.livejournal.com/263863.html

Mass, Economy of Force and Maneuver explained. I know creating good AI is problematic. Thing is, results never will be fulfilling, as AI wont match Human (right now). Thing is, there is a way out of this, with much better results - instead of trying to create AI that will respond to any possibilities, why not just to create VI (Virtual Intelligence) that would simulate the human behavior well documented in similar conditions? Instead of having AI that reacts to any human move (sometimes even immediately when human makes the move, which results in strange behavior like formation shuffling we had in ETW), AI should choose the battle plan at the begining, and stick to it entire battle..

Every Military school will tell you that good plan that you stick with it, is much better than perfect plans you switch every few minutes.. Don't try to make AI fully independent, it will never work. Battle where AI knows to use refused flank, and Economy of force, would be much harder to play than battle where AI switches between plans every now and then after each move human player makes...

[warrior6]

because that would take a lot of time and would be just as exploitable as it is right now. how many plans can you really give the AI (given development time and the fact that they would have to do away with the current Battle AI) after a few battles, you will be easily able to recognize AI's "plan" before he engages and counter accordingly.

[Erwin Rommel]

Meh. I prefer the Three Strategies of Huang Shigong. Simple and easy to remember. Did I mention its only 3?

[JaM]

warrior6: thing is there is not much you can do against properly executed deferred flank and concentration on other... plus it is 100% better than standard CA's line tactics where they crash into your force in lines... this way it at least resembles the human behavior... If you read something aobut military tactics, you would know that there are not that much different battle strategies you could use in Roman Era due to available unit types. AI/VI should use historical tactics so game resembles the history and not some ultra modern tactics, so it would look like Star Wars... All i'm trying to say is that ROme Total War combat NEEDs to look like combat in Ancient Era. AI should not try to try some radical/untested or random strategies, but use systems that were used in reality... Plus, those examples on that web page, are quite hard to resist if done properly, average Human player would have quite a problem to resist such tactics no matter how many times you try...

and no, it wouldnt take more time, than trying to develop AI that will respond on human player behavior...it would be exactly the opposite.

[warrior6]

warrior6: thing is there is not much you can do against properly executed deferred flank and concentration on other... plus it is 100% better than standard CA's line tactics where they crash into your force in lines... this way it at least resembles the human behavior... If you read something aobut military tactics, you would know that there are not that much different battle strategies you could use in Roman Era due to available unit types. AI/VI should use historical tactics so game resembles the history and not some ultra modern tactics, so it would look like Star Wars... All i'm trying to say is that ROme Total War combat NEEDs to look like combat in Ancient Era. AI should not try to try some radical/untested or random strategies, but use systems that were used in reality... Plus, those examples on that web page, are quite hard to resist if done properly, average Human player would have quite a problem to resist such tactics no matter how many times you try...

and no, it wouldnt take more time, than trying to develop AI that will respond on human player behavior...it would be exactly the opposite.

its not as easy you think. comparing it to chess illustrates your not familiar with how AI works and where its at in this day and age. chess AI is easy enough to create AI that can give professional human players real problems. modern video game AI is nowhere near that level because of how much more complex it is.

they have also based the AI off historical maneuvers since empire total war. i remember videos showcasing the AI where they stated they based it off real tactics of the time. the problem is you cannot get it right off to a T. in the end the AI needs to be reactionary. the human can change his formation and position on the map very easily and the AI needs to adjust. even if it goes against its original plan from the get go which would no longer suit itself. also, if it makes the plan before the battle starts, how will account for your formation and also more importantly your position. if the AI decides a frontal assault, would it be wise to continue if i move my army to a hill in a purely defensive formation. if it decides a flanking type hammer and anvil maneuver, and i start the battle with good spearmen and muskets and on the flanks, would it still go through with it? would a real army go through with it?

in this sense, the AI must be reactionary. it needs to try to rearrange itself accordingly. the AI needs to remain reactionary because thats how humans are. if the situation changes distractically, your going to change your plan accordingly. to an extent its a lot like the mike tyson saying "everyone has a plan, until they get punched in the face"


on the other hand, implementation of completely historical tactics that were meant for an army of 40,000 troops to 3000 troops is also somewhat difficult.

[DaVinci]

its not as easy you think. comparing it to chess illustrates your not familiar with how AI works and where its at in this day and age. chess AI is easy enough to create AI that can give professional human players real problems. modern video game AI is nowhere near that level because of how much more complex it is.

they have also based the AI off historical maneuvers since empire total war. i remember videos showcasing the AI where they stated they based it off real tactics of the time. the problem is you cannot get it right off to a T. in the end the AI needs to be reactionary. the human can change his formation and position on the map very easily and the AI needs to adjust. even if it goes against its original plan from the get go which would no longer suit itself. also, if it makes the plan before the battle starts, how will account for your formation and also more importantly your position. if the AI decides a frontal assault, would it be wise to continue if i move my army to a hill in a purely defensive formation. if it decides a flanking type hammer and anvil maneuver, and i start the battle with good spearmen and muskets and on the flanks, would it still go through with it? would a real army go through with it?

in this sense, the AI must be reactionary. it needs to try to rearrange itself accordingly. the AI needs to remain reactionary because thats how humans are. if the situation changes distractically, your going to change your plan accordingly. to an extent its a lot like the mike tyson saying "everyone has a plan, until they get punched in the face"


on the other hand, implementation of completely historical tactics that were meant for an army of 40,000 troops to 3000 troops is also somewhat difficult.

Hm, i think it's possible to make a mix of historical acting BAI mixed with responsive AI-skills for diverse situations. It's mainly a matter of programming-time (and testing-time, of course) for that part of the game to provide that for the different fighting cultures, but just imo. very important and worth to invest lots of development-time. And btw., one basic part of historical acting BAI would be Formations_AI programming, which is not that hard nor needs super lots of time.
A great step forwards would be an AI that can decide to withdraw properly, ie. in such cases which you have described above (human player positions on a hill with superior deployment), AI withdraws, samples more forces and attacks later or is able to decide for a more save defence (just like a human player), instead of commiting suicide actions.

[DarthRuanek]

Yeah. Having a set battle plan would help. Responsiveness is obviously important, but not all-important. Human players rarely need to change their entire strategy after seeing an enemy cavalry group coming, for example.

It would help if the AI kept reserves behind the front lines, to respond to unexpected developments such as a sudden cavalry charge, a quick, unexpected rout, etc.

Another thing would be to teach the AI not to do certain things. Having three enemy groups in phalanx formation chasing my mounted general always made me facepalm.

[JaM]

its not as easy you think. comparing it to chess illustrates your not familiar with how AI works and where its at in this day and age. chess AI is easy enough to create AI that can give professional human players real problems. modern video game AI is nowhere near that level because of how much more complex it is.

they have also based the AI off historical maneuvers since empire total war. i remember videos showcasing the AI where they stated they based it off real tactics of the time. the problem is you cannot get it right off to a T. in the end the AI needs to be reactionary. the human can change his formation and position on the map very easily and the AI needs to adjust. even if it goes against its original plan from the get go which would no longer suit itself. also, if it makes the plan before the battle starts, how will account for your formation and also more importantly your position. if the AI decides a frontal assault, would it be wise to continue if i move my army to a hill in a purely defensive formation. if it decides a flanking type hammer and anvil maneuver, and i start the battle with good spearmen and muskets and on the flanks, would it still go through with it? would a real army go through with it?

in this sense, the AI must be reactionary. it needs to try to rearrange itself accordingly. the AI needs to remain reactionary because thats how humans are. if the situation changes distractically, your going to change your plan accordingly. to an extent its a lot like the mike tyson saying "everyone has a plan, until they get punched in the face"


on the other hand, implementation of completely historical tactics that were meant for an army of 40,000 troops to 3000 troops is also somewhat difficult.

oh, believe me i know how it works, i just think its time for different approach, and instead of trying to make AI to "think" i'm saying CA would be better off to teach AI to use tactics really used instead of trying to come with something unpredictable. I recommend you reading some articles when ETW was about to be released - they were talking about advanced battle AI that will try to outsmart the human player - and instead we had Battle AI that outsmarted nobody,because it reacted even to small moves, which resulted in formation shuffling... and i doubt anybody found vanilla ETW battle AI anyhow challenging... (most players just bombarded AI when it started to swap left units with those on right wings, so they took horrible oblique shot casaulties... battle over)

[JaM]

btw, how do you think are Chess program algorithms made? they are NOT made to respond to human player, but they are programed to have a plan and try to win based on it. Battle AI needs to be active to be a challenge, it cannot be reactive!

[Greve Af Göteborg]

The AI should have a preliminary plan at the start of the battle that the AI will try to achieve. But the AI should also be flexible in changing parts of the plan as the battle progresses.

[JaM]

Yes, i'm not telling it should not react. I'm just pointing out the need for grand plan at the beginning, and need for AI to work as an whole army, and not like a bunch of independent units we had in ETW and NTW...

[DaVinci]

Finally a real important theme in this forum section, good move, JaM

And yes, i agree. An active BAI with (authentic) start-plans per possible situation probably would be revolutionary for TW battles, while the programming-challenge would be indeed to implement properly the reaction-item per (changed) situation, ie. to surprise/challenge the human player still, when the human player tries to trick out the AI.

[Bethencourt]

This is very much this idea:

http://www.twcenter.net/forums/showt...9#post11772749

Maybe if you give different AI to different generals admirals you can get more varied battles.

In campaign it is done for different factions.

Or maybe create 50 different plans of battle and add randomly one to the AI before each battle. I think this is smarter than giving each unit an individual reaction to my units movement and actions. To avoid the pack fenomenon that you could see in previous games, units going without any formation from left to right and back again untill the end of the battle. So given a battle plan the units can react when acomplishing their part of the plan. Now it seems they go headless.

And it is realistic, all generals have some plan to execute in battle, no matter later it does not get perfectly done .

And going further on it, the AI could evaluate the superiority/ inferiority situation so the selection of the plan would be done betwen a list of offensive plans or deffensive plans depending on the situation. It doesn't mean that it always has to be deffensive when in inferiority(in quality or numbers), or vice versa, some general personality factor could be used.

[JaM]

DarthRuanek: not just what not to do, but how effectively using reserves - as pointed in that link, every general should try to gain local superiority instead of parity, because that is how battles are won.. and you can count on it, when i say that human player would have quite hard time to effectively handle such use of force - imagine AI that would assess human player deployment, then decide that it will refuse one flank, concentrate on another and keep reserves close, hidden in woods. then it would allow you to attack its refused wing, and simultaneously with your attack counterattacked with superior force at the other wing,together with reserves... btw many generals used this tactics to present weak position to enemy to draw him to attack it - most famous were Ceasar and Napoleon btw (battle of Austerlitz is perfect example...)

[CptAustus]

The AI should try to build "pre-determined" armies, so it can more easily employ strategies such as "Ok, hastati are closing in, now send the principes to the flanks and my shiny triarii behind all of them, just in case" or "oh wait, the player has cavalry, I should probably keep those two spears and two light cav to defend my back instead of fail at out flanking". That would be much cooler than your average AI throwing all they have against your first line of disposable units.

[Rasic]

Every Military school will tell you that good plan that you stick with it, is much better than perfect plans you switch every few minutes.. Don't try to make AI fully independent, it will never work. Battle where AI knows to use refused flank, and Economy of force, would be much harder to play than battle where AI switches between plans every now and then after each move human player makes...

They also will tell you that you should have a contingency plan for all predicted enemy actions. It's the basis of an operations order. The problem comes in when the AI can't react to changes on a larger tactical level or move with any sense of direction, only small level reactionary moves like you said.

It seems to me right now that the AI just tries to pair up two units it considers good matches. It sometimes works and they can kill a group of skirmisher units with a nice cavalry flank when you aren't paying attention or leave them unguarded, but generally speaking they can't adapt on a higher level to see more wide flanking maneuvers that involve several units.

I agree completely that the AI should have more of a set plan for overall, but I think that is achieved by doing away with unit pairing on an individual basis.

The "legion system" if applied to the other factions would mean that you will generally see armies who are built with a purpose in mind instead of the AI combining multiple units into a random mixed stack. If the AI builds an army to be specifically heavy infantry focused than it should use that specific advantage. If it builds one that is heavy skirmisher focused than it should attempt to harass and bleed the player as much as possible before closing in.

[JaM]

Actually changing plans during battle was almost never successful, most of the time it just confused own units -we are talking about ancient era without advanced comunications modern armies have - most of the time, it took lots of time to pass the orders, so practically commanders relied on subordinates to follow the master plan for battle.

[Bethencourt]

Terrain is important but in previous games solution and in the proposed one. It seems that in previous one the terrains was not taken into accout by the AI. I remember now so many Empire battles played in Milan where there are good high positions and the AI cannot react properly.


Also it seems that the plan should be given to groups of units so they can perform the movements orderly, not in chaos, even the bakwards movements when the formation consideres that there is need of it. So if a unit is in the right position of a line, if the formtion is ordered to go back that unit just turns back and starts walking in the left side of the formation, and does not try to go again to the right position it had in the previous movement, so units in the line do not make an X movement to recover initial position. This simple change would mean a lot. Maybe it is already solved in Shogun II?

[DaVinci]

Terrain is important but in previous games solution and in the proposed one. It seems that in previous one the terrains was not taken into accout by the AI. I remember now so many Empire battles played in Milan where there are good high positions and the AI cannot react properly.


Also it seems that the plan should be given to groups of units so they can perform the movements orderly, not in chaos, even the bakwards movements when the formation consideres that there is need of it. So if a unit is in the right position of a line, if the formtion is ordered to go back that unit just turns back and starts walking in the left side of the formation, and does not try to go again to the right position it had in the previous movement, so units in the line do not make an X movement to recover initial position. This simple change would mean a lot. Maybe it is already solved in Shogun II?

Totally right, terrain is one these "Basic things Battle AI should know...", one priority. As well in NTW, AI regulary tries to bomb away hills, ie. positions its artillery behind hills instead on them, that looks pretty funny or rather embarrassing for the AI, and you think "oh, this poor AI, how could've gone this through the QM of CA's testing department", same is valid for infantry there. Related to Rome II, it just concerns all kinds of range units, not so much as with gun powder TW titles, but still valid.

[Humble Warrior]

Terrain is important but in previous games solution and in the proposed one. It seems that in previous one the terrains was not taken into accout by the AI. I remember now so many Empire battles played in Milan where there are good high positions and the AI cannot react properly.

Yes. While STW2 BAI is now getting better, it still fails when it comes to terrain. Yes, the programmer understands hills and possibly forests, that`s it. But he hasn`t bothered about getting the AI to use the lie of the land and LOS. For example, it never makes use of a valley situation if one presents itself, or hides behind a hill to force the enemy closer. I hid my melee men just behind a hill to force his better rifle men to come right up then I charged and got them in melee.

On the other hand, i`ll admit that if people insist on using the `debug` camera, then no AI can be smart enough to beat him, not even a Human can. So I hope the people complaining about bad AI are using realistic and even restricted views.