New Developer Blog: AI and Battles

[Bob the Insane]

Wow, I totally missed that! I hope that means we can now save our campaign battles like in the original MTW; that would be very happy.

Well he started off by saying this was a custom game and you could always save custom battles, so nothing new there...

I know that many out there have pretty exacting standards, but it would not have taken all that much to improve the battles in RTW. The mods do half of it already with speed, stat and morale tweaks. Simple things like getting the AI to maintain a formation and make intelligent decisions as to when to attack and when to retreat and keeping the commander safe would have made a very large impact into the game. (Note I mean simple in concept, not necessarily in code)

What I mean is that RTW would have been way better with a few little improvements in the battle AI so if they have made big steps in improving the battle AI, well I am pretty excited...

[Midnight]

I'll be very happy if the AI now uses its forces intelligently. Slower combat speeds also sound great.

[Martok]

Well he started off by saying this was a custom game and you could always save custom battles, so nothing new there...

Now don't you go ruining my optimism. We want saved replays for campaign battles! Let's hear it!

I know that many out there have pretty exacting standards, but it would not have taken all that much to improve the battles in RTW. The mods do half of it already with speed, stat and morale tweaks. Simple things like getting the AI to maintain a formation and make intelligent decisions as to when to attack and when to retreat and keeping the commander safe would have made a very large impact into the game.

Indeed. If Rome's AI had been able to at least follow some of the simpler aspects of battle, I think most people's opinion of it would have been much more favorable. I really hope they've fixed the AI in Medieval 2 so it doesn't make as many blatantly bone-headed mistakes.

[Puzz3D]

The magic number for unit size in the original battle engine was 60. When that was coupled with the 1 second combat cycle and a chance to kill that nominally ranged between 0.5% (NI vs NI) and 10% (ND vs ND), you got frontal combat resolution times ranging from 3 minutes to 30 seconds. You also got predictable results vs melee and missle combat factors within a reasonable degree of uncertainty. With a 20% advantage in melee combat factor (1 point) you could expect to win 6 out of 10 times, and with a 40% advantage (2 points) you could expect to win 9 out of 10 times.

MTW introduced variable unit size. Although it still ranges around the 60 man size, it makes balancing harder because of the fixed 20% step in melee combat factor (The steps in RTW are apparently 10% which is better). Combat factor has to be used to compensate for the disadvantage suffered by smaller units vs larger units due to multiple attacks on individual men, and it's not a linear relationship. I remember Tosa and I tried to balance 100 man yari ashigaru units in STW/MI against the rest of the 60 man units, and we weren't successful to the degree necessary to ensure balanced gameplay. Variable unit size also complicates balancing ranged units because the ranged unit's overall effectiveness on a target will vary depending on the size of the target unit.

Also, in RTW artillery apparently gets the same number of kills regardless of the size of the target unit which doesn't mesh well with the morale system, if it's similar to the system used in STW/MTW, which applied a morale penalty based on the percentage of men in the unit killed in a combat cycle.

I think it's important to use the original game, STW, as the standard of comparison when you talk about playbalance and gameplay. If MTW is used, you have the situation where, although improved, M2TW could still be inferior in terms of playbalance to CA's original standard, and if you use RTW as the basis, inferior in not only playbalance but overall gameplay (missing features in the engine) as well.

In terms of AI, it's certainly a big improvement to have the AI wait until it has used it's ammo and then charge forward as a whole army rather than use piecemeal attacks where a unit goes forward only to decide that was a mistake and turns around and walks back to its lines. If you don't have a strong rock, paper, scissors battle system, then making individual unit matchups isn't as important. However, the AI is still going to be predictable unless it varies its strategy for similar situations the way a human would. Original STW had an AI that varied its strategy over about 3 different choices for the same initial battle situation.

[Dutch_guy]

Now don't you go ruining my optimism. We want saved replays for campaign battles! Let's hear it!

Hear, Hear !

[Geoffrey S]

I can't for the life of me remember where it was, but there was a developer somewhere saying replays of campaign battles was being considered. Must add my voice to those hoping they'll be included.

Blog is interesting, but AI before release is always suspect; particularly in FPS, where a hyped-up AI turns out to be simple, predictable scripting. The explanation from Puzz3D on the magic numbers is interesting too, definitely a step forward for particularly multiplayer if results are consistant.

[screwtype]

Maybe screwtype and Puzz can shed light on what...

"As a little teaser I must mention unit sizes! For those that have played TW since STW you will know what the magic numbers are and should be really be happy to see them make a return in M2TW."

...means.

Thanks for the vote of confidence AG but I'm really just an opinionated casual gamer. There are heaps of guys who know more about TW than me. I'm not even in the same league as Puzz, TW-wise - he's the real deal.

But since you asked - I really have no idea what is meant by this sentence, perhaps they mean a return to classical TW sizes, ie 60/80/100/120 men per unit or something.

[screwtype]

IIRC, in MTW, you had 100 for basic infantry and 60 for good infantry, cavalry was 40 and I think generals' units did not actually scale up, making them less important in larger scale campaigns.

I think in STW you only had a unit of 10 heavy cav for the general didn't you?

Mind you, those 10 heavy cav still came in pretty handy...

[screwtype]

It sounds promising but it's still a 'wait and see' for me. Another side of this AI behaviour may be being forced to chase horse archers who wont retreat round and round the map in a futile attempt to catch them.

Actually that happened sometimes in Shogun. If you only had foot units, you would chase the horse archers around the map getting peppered until your footsoldiers got exhausted, then the HA would charge and route your forces.

I thought it was quite realistic actually. That's what should happen if you try to run down HA with footsoldiers...

[SpencerH]

Actually that happened sometimes in Shogun. If you only had foot units, you would chase the horse archers around the map getting peppered until your footsoldiers got exhausted, then the HA would charge and route your forces.

I thought it was quite realistic actually. That's what should happen if you try to run down HA with footsoldiers...

I'm not discussing the tactics of a HA army v foot army. I'm taliking about after I've obliterated the AI standing army but one or two (usually) HA units refuse to retreat off the map and must be chased endlessly round and round. I've even 'lost' battles in this fashion.

I know it happened in Shogun, MTW, and RTW. Isnt it about time it was fixed. I'm not saying this will contribute to the problem BTW, I'm just pointing out the possibility. Given that the tone of the article suggests that the author was pleased and surprised by the AI behaviour I have to wonder if there hasnt been sufficient playtesting.

[econ21]

I know it happened in Shogun..

Actually, I don't think it did happen in Shogun. STW had what has been called "Benny Hill" code, whereby a unit that repeatedly tried to evade combat would eventually rout. It's not there in RTW, so you can win a (timed) defensive battle with only one cavalry unit if you have taken out the enemy cav.

Given that the tone of the article suggests that the author was pleased and surprised by the AI behaviour I have to wonder if there hasnt been sufficient playtesting.

Well, it often does take a while to find faults. But IMO this issue is pretty trivial in the grand scheme of things. Certainly not the key factor to decide if M2TW has better AI than RTW.

[Drake]

Did anyone else get caught in Rome, usually always with an Egyptian army that when you defeat 99% of it there's one chariot left who runs around for bloody ages and you end up chasing it around and around the battlezone.

I hope the battle AI is fixed up so that should this happen in MTW2 that this unit catches a grip and runs for the hills when faced by about 500 men.

[Puzz3D]

But since you asked - I really have no idea what is meant by this sentence, perhaps they mean a return to classical TW sizes, ie 60/80/100/120 men per unit or something.

I think Palamedes is speaking from a multiplayer perspective when he talks about magic numbers for unit sizes.

In STW multiplayer, the default size was 60 men and the consensus was that that offered the best balance between fighting and maneuverability, although, cavalry was considered to be a bit cumbersome at that size because the maps were small and if one man in the cav unit caught the edge of an enemy unit the whole cav unit got drawn into a fight with that unit.

In MTW, the default was cav = 40 men, most non-spear inf = 60 men and spears = 100 men. The 40 man cav helps maneuverability and the 100 man spears helps with the rank bonus, but overall it's still based around the 60 man size. We can see in Samurai Wars for MTW/VI that maneuvering 60 man cav units isn't as much of a problem on large maps, and that 60 man size means the cavalry retains its usefulness for more encounters than do 40 man units. Also, after STW, you could disengage cavalry from a fight.

In RTW, the default seemed to be based around a 40 man unit size, and there was no way to get it to 60 because the next size was x2. The perception by many veterans at the time was that the default size of 40 was causing the fast routing in multiplayer and that 60 was preferable while 80 would reduce maneuverabliity too much. Gil Jaysmith posted that the solution to fast routing was to play at huge unit size which made him seem oblivious to the importance of maneuverability to the gameplay. Huge setting can also reduce multiplayer fps a lot.

I can understand wanting to cover a wider range of unit size than the factor of 2 that 60/80/100/120 provides. However, why were the steps made larger when all you have to do is increase the number of steps? Increasing the size of the steps demonstrates an insensitivity toward providing multiplayers the kind of control over game parameters that they need to improve playbalance. You can also see this in the on/off nature of the morale, fatigue and ammo settings, and further in the condensing of what were once 3 separately controlled parameters into a single on/off setting. You also see this in the removal of separate money settings for attacker and defender, and the reduction of the money settings to discrete values in RTW v1.0. It wouldn't matter if CA balanced the game well to begin with, but they didn't. Multiplayers were left not only with an unbalanced game, but no way to improve it either. The money setting was changed back to a continuously variable amount in RTW v1.2, and M2TW apparently restores separate money setting for each team. However, the game mechanic simplifications in the new battle engine are apparently still there. Magic unit sizes aren't going to fix that, and those numbers would change anyway with the new engine. What's important is having the units and armies fight long enough for flanking tactics to be employed, and for those flanking tactics to be effective unless the idea is to have a shoot and rush type of gameplay.

[SpencerH]

Actually, I don't think it did happen in Shogun. STW had what has been called "Benny Hill" code, whereby a unit that repeatedly tried to evade combat would eventually rout. It's not there in RTW, so you can win a (timed) defensive battle with only one cavalry unit if you have taken out the enemy cav.


Well, it often does take a while to find faults. But IMO this issue is pretty trivial in the grand scheme of things. Certainly not the key factor to decide if M2TW has better AI than RTW.

You could be timed out while chasing cav in Shogun too. Actually as I was writing this I was thinking of chasing mounted crossbowmen and being timed out. Hmmmm, that doesnt fit!

In any case, I dont think it's a big deal either. I'm just pointing out that without other AI improvements to prevent the 'infinite chase sequence' this seemingly positive change could have unforseen repercussions.

[screwtype]

Actually, I don't think it did happen in Shogun. STW had what has been called "Benny Hill" code, whereby a unit that repeatedly tried to evade combat would eventually rout. It's not there in RTW, so you can win a (timed) defensive battle with only one cavalry unit if you have taken out the enemy cav.

Yes, but that was only for the human player as I recall. The AI could avoid combat as long as it liked and not rout.

[screwtype]

I'm not discussing the tactics of a HA army v foot army. I'm taliking about after I've obliterated the AI standing army but one or two (usually) HA units refuse to retreat off the map and must be chased endlessly round and round. I've even 'lost' battles in this fashion.

I know it happened in Shogun, MTW, and RTW. Isnt it about time it was fixed. I'm not saying this will contribute to the problem BTW, I'm just pointing out the possibility. Given that the tone of the article suggests that the author was pleased and surprised by the AI behaviour I have to wonder if there hasnt been sufficient playtesting.

Yeah, that could be a problem, but the solution was to have a unit of Yari cav of your own to deal with those pesky horsemen.

You're right it could be annoying, but since it's something that could be avoided with proper preparation, I don't see that it was such a big deal. To me, the occasional mismatch of armies just added to variety and the imponderables of battle, but it's different strokes for different folks I guess.

[screwtype]

In STW multiplayer, the default size was 60 men and the consensus was that that offered the best balance between fighting and maneuverability

Well regardless of what he meant, I agree that the 60-man unit size seemed much more appropriate than the 40 man default in RTW. Also those silly odd numbers for cav were rather annoying in RTW. So I'd welcome a return to the 60 man default size if that's what occurring.

I never played RTW much above the default size, because quite frankly I didn't like the game much and by the time I got a system capable of handling larger sizes, I had lost interest. So I don't remember what the actual larger sizes were, only that there was no 60-man size you could play which was annoying. But I liked the 20-man increment in unit sizes we had in STW and I saw no reason as you say to have larger increments in RTW.

RTW in general was remarkably lacking in settings for a game of that sophistication. Compare it, for example, to Imperialism II, a modest little strategy game that nevertheless has a very large number of difficulty settings from anywhere between 100 or less to 400. Or to the Panzer General series which again has a great many different settings the gamer can tweak to his desires. Give us more choices please CA!

[Bob the Insane]

One of the nice things about playing RTW on huge was that it actually made the republic era cohorts the correct size, 160 men (two centries stood next to each other).

Also the phallanx units looked better with large numbers and the masses of men simply looked right on the ancient battlefield...

And unlike MTW it did not take twice as long to build huge units. I remember trying huge in MTW but I did not like the doubled recruitment times nor how unweldly everything got on the battlefield.

I wonder how MTW2 will play?

[|Heerbann|_Di3Hard]

One of the nice things about playing RTW on huge was that it actually made the republic era cohorts the correct size, 160 men (two centries stood next to each other).

Also the phallanx units looked better with large numbers and the masses of men simply looked right on the ancient battlefield...

I agree with you. In my opinion they developed RTW with huge units size, but while testing they recognized, that the graphic engine cannot handle enough men. So they shrinked the size to 20 men units :( Therefore huge unit size looks better than normal.

[Jambo]

lol Puzz, the yari ashigaru balancing discussions... man those were a real headache.

one minute , next minute

[L'Impresario]

An annoying thing with RTW default (or even large) settings is that you can end up having a handful of men still in the fight if they manage to avoid annihilation by any cav giving chase. In earlier games of the series that was the exception for logical morale settings. Sure, you could end up with that odd cav or 10 inf men, but not so often as to hurt any suspension of disbelief.

[AussieGiant]

Thanks for the vote of confidence AG but I'm really just an opinionated casual gamer. There are heaps of guys who know more about TW than me. I'm not even in the same league as Puzz, TW-wise - he's the real deal.

But since you asked - I really have no idea what is meant by this sentence, perhaps they mean a return to classical TW sizes, ie 60/80/100/120 men per unit or something.

Hey screwtype,

Well, your opinions seems well thought out and accurate to me.

Certainly Puzz is "Yoda like" when it comes to the whole topic, but your always on the mark, or very near it.

I've been off-line a bit so, it seems Puzz has given us the explanation on this page of the thread.

[screwtype]

Hey screwtype,

Well, your opinions seems well thought out and accurate to me.

Certainly Puzz is "Yoda like" when it comes to the whole topic, but your always on the mark, or very near it.

Modesty prevents me from an outright endorsement of your views, so let's just say you are obviously an individual of superior tastes

[Puzz3D]

An annoying thing with RTW default (or even large) settings is that you can end up having a handful of men still in the fight if they manage to avoid annihilation by any cav giving chase. In earlier games of the series that was the exception for logical morale settings. Sure, you could end up with that odd cav or 10 inf men, but not so often as to hurt any suspension of disbelief.

In MTW, there is a global morale penalty for level of decimation of the army. It helps prevent a battle from being dragged out by a handful of units which have suffered heavy casualties. I don't know the level of decimation at which this penalty is incurred. I don't think this penalty was in STW because all units routed towards the original entry point on the map so it's much easier to chase off a defeated army. MTW introduced routing away from the threat which makes it a lot harder to mop up after a battle, and I think that's why the level of decimation penalty was introduced.

[AussieGiant]

Modesty prevents me from an outright endorsement of your views, so let's just say you are obviously an individual of superior tastes

LOL. Classic comment.

Puzz, just how much time have you spent on this whole "Thing". By "thing" I mean, how do you know so much?

And are you a professional gamer?

[Puzz3D]

Puzz, just how much time have you spent on this whole "Thing". By "thing" I mean, how do you know so much?

And are you a professional gamer?

I'm an electrical engineer. As part of my job, I did Monte Carlo simulations on scintillation detectors coupled to phototubes through light guides for a couple of years which involves using computer generated random numbers and mathematical models to simulate the light distribution from an array of scintillation detectors onto an array of phototubes, and then use a maximum liklihood algorithm on the phototube outputs to determine from which detector the light originated. I'm also a chess player with a 2000 USCF rating. I went from playing competitive chess for about 25 years to playing Shogun Total War online in Oct 2000. I was intrigued by STW's use of a statistical model to simulate combat, the emphasis the game put on tactics, coordinating units of various types and protecting your daimyo ala chess, and the prospect of playing as a team which chess lacks. I was amazed at how well the system worked, and LongJohn should be justly proud of developing the battle system. I've also been on four CA beta teams, and don't ask me about that because CA NDA's never expire.

Shogun was incredibly well balanced, but there was a bug in the cavalry charge which I only recently became aware of when developing the Samurai Wars unit stats for Barocca's STWmod for MTW/VI when ShingenKrypta and I went back and ran some online tests in original STW. There is an unofficial guide on IGN published back in 2000 with a chart Shogun Guide that clearly shows that Heavy Cav was supposed to defeat Warrior Monks. Due to the bugged cav charge which made cav behave more like mounted infantry, HC wasn't very effective against WM, and this lead to the infamous monk rush which was the only really serious problem with the online game. Expert players have claimed they could regularly defeat the monk rush. Unfortunately, although original STW had the capability to save replays using a commandline option, CA didn't reveal how to do it at the time, so no battles from those days exist today.

Back then I remember many players asking, "Why did my units run away?". Very little was known about the morale system and CA wasn't talking. We did know the unit stats from the Official Strategy Guide and a couple of online guides, but not the projectile stats. None of these parameters were in external text files in original STW like they are now. Since I had two computers on a LAN, I started designing tests to figure out the morale system. I determined the numerical transistion points of impetuous, steady, uncertain, wavering and routing. Then I was able to geometrically determine what constituted a flank, how close a unit had to be to inflict a flanking morale penalty, the size of that penalty and where you had to position a unit so that covering a flank cancelled the penalty. I discovered the cause of the Benny Hill effect, and this was confirmed by LongJohn. I also determined the size of the outnumbering morale penalty, that only the number of enemy banners within a certain radius contributed to it, that the size of the penalty diminished with the total number, that it had a maximum cutoff value of 3 banners and that it didn't matter if the enemy units were hidden. I measured the size and range of the morale effect of friendly units rout, the supporting effect of the general and the effect of the general dying (-8 morale) or leaving the field (no effect on morale). I measured the morale effect of being on higher ground, the morale penalty for being disrupted, the penalties from being targetted by archers and guns, and observed the hysteresis in the morale system. I went on to measure fatigue rates for walking, running and fighting and different weather conditions vs armor, the loss of accuracy for archers in light, medium and heavy rain and the misfire rate of guns in light, medium and heavy rain. Kraellin's map, Ranges, was instrumental in being able to repeat the tests at precise ranges. Later after MI was out and the parameters were in external files, I measured the effectiveness of archers and guns vs armor rating and power rating, and the ratio of kills in units adjacent to the targetted unit. I determined the ballistics of guns, and around this time LongJohn confirmed a lot of my measurement and revealed the chance to kill algorithm. I also discovered that the cause of the strange gameplay in WE/MI multiplayer was a +12 morale being added to all units. This was later confirmed by CA, and they removed it in the v1.01 patch to WE/MI. I believe that our interest in the game mechanics influenced LongJohn to write those two fine chapters about teh game mechanics in tehe Official MTW Strategy Guide. However with RTW, CA has closed the door on that.

I know some players discovered things about the game and kept it to themselves to use to advantage in multiplayer games. Since I was a chess player and that game is played with both players knowing all the rules of play, I always posted at the org everything I discovered about the game. I don't ascribe to the idea of keeping information about the game mechanics secret for personal advantage in a multiplayer battle. From my experience playing chess, I know that mastering tactics is the lowest level of playing the game. Once players have mastered the tactics, the gameplay rises to a strategic level and that's when it becomes really interesting especially in a game such as STW where teamplay is possible. The plethora of units and factions and upgrades (STW only had honor upgrades) places too much emphasis on buying the best units with the best upgrades which robs the multiplayer gameplay not only of its strategic potential, but even it's tactical potential. The fast gamespeed, lack of options and loss of battle engine features in the current incarnation of Total War takes the game further down that road and away from the outstanding accomplishment in terms of gameplay that original STW achieved.

Currently I don't do any testing on RTW/BI because I don't think the new engine warrants the time and effort to do those tests.

[Nobunaga]

[hoom]

they recognized, that the graphic engine cannot handle enough men

The graphic engine has no issues with enough men.
Most peoples PCs have issues with handling enough men (but it doesn't take that much PC to handle Huge size, I handled it with a 2.2ghz single core with 1GB ram & a 9800np GPU, though it chugged on the bigger battles)

PS: Puzz = legend.

[Bob the Insane]

In MTW, there is a global morale penalty for level of decimation of the army.

I think they have this in RTW, at least during a custom battle that I had as good as won I noted one still intact enemy unit (most of the rest where routing off the feild at this point) had a morale level of Steady with the note "dismayed at the loss of the battle".

Now the note usually refers to whatever morale bonus or penalty the unit is experiencing at the time and this note seems to indicate some form of global "we lost" morale penalty...

[SpencerH]

Yeah, that could be a problem, but the solution was to have a unit of Yari cav of your own to deal with those pesky horsemen.

You're right it could be annoying, but since it's something that could be avoided with proper preparation, I don't see that it was such a big deal. To me, the occasional mismatch of armies just added to variety and the imponderables of battle, but it's different strokes for different folks I guess.

I'm not sure if its my writing or your reading, but I am talking about my (usually) winded cavalry chasing winded AI cav round and round the map edges. It is not something that can be "avoided with proper preparation" (although it can be dealt with in the battle with a blocking force to cause the AI cav to slow and be caught of course).