New Developer Blog: AI and Battles

Quote:

---

Originally Posted by Puzz3D

I'm an electrical engineer.

---

That was a nice read, good to know you a little better. Monte Carlo simulations imply you know a thing or two about statistical physics and explains why you're strong with numbers. I'm sure you could re-test the RomeTW or MTW2 engine if you wanted to and find those numbers again.

Even with a game like STW, with only a limited number of units, balancing turns out to be difficult. But an army is only broken if it doesn't have a counterarmy. In RTW, the all-cav place-them-on-top-of-each-other army seems invincible, but I'm sure that that army will no longer be possible in MTW2. What other options of unbalance are there?
- spears/swords/cav/archers too strong
- 2 out of 4 too strong (MTW, 8 men-at-arms and 8 knights)
- none of them too strong (STW, though monks can be nasty)
I don't know MTW2 will turn out in advance, but I hope every army has a counterarmy.

More important is more clever AI however. Perfect balance may lead to chess and a discovery of new numbers, but perfect AI makes you playing the campaing game for a long time. Multiplayer would be better with balance, singleplayer with AI.

I would aim to perfect one before trying to perfect the other.

Quote:

---

Originally Posted by SpencerH

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).

---

Hmm, I must say I've never experienced this problem. That has certainly never happened to me in Shogun - my Yari cav always chases down the enemy, LOL - and I can't recall it ever occurring in MTW or RTW either. Spear cav can usually catch HA pretty easily in my experience. Have you tried, say, not clicking on the HA, but clicking on the ground past them, or to one side of them and then wheeling round to get them before they start retreating?

I wont say that it's exceedingly common in RTW but it's happened enough times to be of note. A typical situation for me will be a fairly open battlefield with AI HA or light cav moving forward against my flanks followed by a line of AI inf units. The inf battle usually requires almost all of my attention once they are engaged because of the ridiculous battle speeds. Therefore, my counter to the AI cav is to simply to 'click attack' with my cav and get back to the inf battle.

Fast forward to the end of the battle ...

Most enemy units have routed and have left the battle field but 1 unit (usually) of winded AI cav remains pursued by whatever winded cav I have left. I have watched them chase such a unit round the battle square 3 times. Adding additional units to the chase usually does nothing since they will simply join in the game of 'follow the leader'. Perhaps pausing and clicking on the ground might do it but given the lag that occurs after orders, I doubt it and if they didnt catch them at that point they'd be even further behind with the lag after new orders.

Hi Puzz,

Great post and thanks for the indepth explanation of how you have been involved over the years.

I gotta say, you do scare me...but in a good way :laugh4:

I really don't think MTWII will be as pure and as well done as Shogun when it comes to AI and balancing, but I do hope that they put enough thought and balancing into this version for it to be able to stand as a good game by itself.

Comparison's are always dangerous and rarely fair but I get the impression that MTWII will stand a head taller than everything that has gone before...but still look up to the oldest brother of them all...Brother Shogun, "the number one son, in the family".

:2thumbsup:

The chasing of skirmisher cavalry round the 4 corners of the battle maps in RTW was one of the most tedious aspects of the game. For that reason alone I always hated fighting against those factions which had loads of horse archer units, e.g. Persia.

In Shogun there wasn't this problem since horse archers weren't the fastest cavalry. Yari's could and would catch them making them not the irritating omnipotent nuisance that they are in Rome. I can only hope that M2TW, with its newly introduced "impassable terrain", will prevent similar ridiculous escapades chasing skirmishing horse units round the 4 corners of the battle maps...

Quote:

---

Originally Posted by AussieGiant

Comparison's are always dangerous and rarely fair but I get the impression that MTWII will stand a head taller than everything that has gone before...but still look up to the oldest brother of them all...Brother Shogun, "the number one son, in the family".

---

Really? Consider this:

The melee battle mechanic has been changed between the older STW/MTW battle engine and the new RTW/BI battle engine. In the old engine, multiple attacks occured within a combat cycle with death resolved at the end of the cycle. Each man could strike once and parry multiple times within a single combat cycle. A man had 8 sides, so could in theory be attacked 8 times with combat bonuses applied if the strike came from the side or back. Since each man got to strike before death was determined, two men could kill each other, and I've seen this happen. This is important to playbalance because the outcome of combat is not biased by an arbitrary determination of who strikes first. I've tested 60 man units many times in STW, and it's extremely rare for equally matched units to come back to win once getting behind by a 15% deficit in men which is 10 men.

This mechanic was change in RTW. You can see by observation that the strike and counterstrike are now taking place in two different combat cycles. The man who strikes first is getting an advantage. I'd like to know how first strike is being determined because it makes a difference. I have observed the same man getting to strike twice before his oppenent strikes back, so maybe there's been an attempt to mitigate the first strike advantage. However, I'm not sure the double strike is necessarily going to the man who was struck first. Even if it's a randomized effect, it's going to contribute to uncertainty in the combat results.

I just ran two tests in RTW of three 122 cohorts fighting three 122 man cohorts. These units are perfectly balanced against each other, but the combat resulted in a very lopsided win for one side. For example, the second test resulted in 291 kills for one side vs 128 kills for the other out of a total os 367 men on each side. This wasn't the result of killing routers. Does the concept of balancing have any meaning when the uncertainty is that high? Does it make sense to spend time and energy trying to master such a game?

Multiple attacks are retained by not allowing a man who is engaged to turn and face a second attack from the side or rear. I'm not sure if I've seen triple attacks, but I may have. It seems to me that the old engine was a better simulation of combat, and it had several features that enhanced that further which have been lost in the new engine. The lower level of uncertainty in the combat model rewarded a well executed battleplan, and there was more depth to the tactics because small advantages weren't swamped out by the random factors. You could actually accumulate small tactical advantages over the course of a battle, and it paid off. The accumulation of small advantages is a basic concept in the game of chess. There is concern now that the finishing moves are going to further denegrate the combat model as a simulation of hand-to-hand fighting.

Officially Considered Puzz...

...thanks...I think.

Oh well, we will see. That's about the only true statement left to make.

The kill moves do seem to imply a need for some large modification of how individual combat is handled.

Do you get to insert your concerns into CA's development at all now?

I used to send the Rome developers the odd mail every now and then during the beta testing process. These were often ideas to improve or enhance the game and granted probably outwith the scope of patch, but as to whether they got read or considered for future projects is a different matter.

In the past, during LongJohn's involvement with TW, there seemed to be more interaction between parties and discussison on game mechanics were commonplace. Because the TW franchise wasn't as commercial and "big business" as it is now, maybe the community had a bigger influence on certain things? - but that's just speculation of course. :D

Quote:

---

Originally Posted by AussieGiant

Do you get to insert your concerns into CA's development at all now?

---

No except what I post on the forums. I've never emailed CA because I know several people who have tried this and they never got a response.

Quote:

---

Originally Posted by Puzz3D

No except what I post on the forums. I've never emailed CA because I know several people who have tried this and they never got a response.

---

Fair enough.

I think they do read a lot here. This forum seems to be to most robust and informed, and that is always of value when companies are trying to maintain their business success.

Quote:

---

Originally Posted by AussieGiant

Fair enough.

I think they do read a lot here. This forum seems to be to most robust and informed, and that is always of value when companies are trying to maintain their business success.

---

Hopefully.But why they don't speak with us directly. E.g. I played more than 4000 (or 5000? ;) ) MTW Vi battles in multiplayer and i guess I know the game better than some MTW2 developer ;)

Quote:

---

Originally Posted by Puzz3D

A man had 8 sides, so could in theory be attacked 8 times with combat bonuses applied if the strike came from the side or back. Since each man got to strike before death was determined, two men could kill each other, and I've seen this happen.

....

---

I do remember reading (in this forum maybe) that there was a limitation in the MTW being so that any individuals could only be engaged by 3 others at most at any one time. This was one of the reason behind the superman kings... They had loads of hit point, good stats and even when completely surrounded where still only fighting three men at once. I agree withthe rest though...

I too would love to know how the battle enigne functions now...

Quote:

---

Originally Posted by Puzz3D

I just ran two tests in RTW of three 122 cohorts fighting three 122 man cohorts. These units are perfectly balanced against each other, but the combat resulted in a very lopsided win for one side. For example, the second test resulted in 291 kills for one side vs 128 kills for the other out of a total os 367 men on each side. This wasn't the result of killing routers. Does the concept of balancing have any meaning when the uncertainty is that high? Does it make sense to spend time and energy trying to master such a game?

---

With permission, Puzz, I'm sure you are not biased, but did you mouse over the 3 cohorts who suffered 291 kills to make sure they contained (367-291)/3 = 25.3 men on average versus 80 on the other side? If you simply used the battle results page (which pops up once the enemy army routs) the 291 number may have been polluted by simulated routing kills even though you didn't kill any routers (assuming we're in custom battle). Still though, the results are staggeringly close to random results. Was the difficulty at normal?

Quote:

---

It seems to me that the old engine was a better simulation of combat, and it had several features that enhanced that further which have been lost in the new engine. The lower level of uncertainty in the combat model rewarded a well executed battleplan, and there was more depth to the tactics because small advantages weren't swamped out by the random factors. You could actually accumulate small tactical advantages over the course of a battle, and it paid off. The accumulation of small advantages is a basic concept in the game of chess.

---

I wholeheartedly agree with this and think the increased random factor is due to the following:
- separate attacker and defender attack rolls, going first matters a lot
- fewer number of attack rolls
- higher chance to hit for units in general, to compensate for the fewer attack rolls
- higher chance to hit for lower tech units as the attack - defense difference doesn't matter as much anymore, (let alone things like height difference, spear vs cav bonuses, etc.)

The new engine will change the following:
- higher number of attack rolls
- lower chance to hit due to higher defense values overall

Quote:

---

There is concern now that the finishing moves are going to further denegrate the combat model as a simulation of hand-to-hand fighting.

---

I therefore do not share this concern.

Quote:

---

Originally Posted by Bob the Insane

I do remember reading (in this forum maybe) that there was a limitation in the MTW being so that any individuals could only be engaged by 3 others at most at any one time. This was one of the reason behind the superman kings... They had loads of hit point, good stats and even when completely surrounded where still only fighting three men at once. I agree with the rest though...

---

I can't find the post, but CA did clarify that in MTW more than 2 men can strike at a single man. They said it's only limited by the number of men that can fit around the enemy man. Each sprite has 8 sides. In practice with two units fighting, you don't often see more than two men attacking one because the men tend to stay in formation even when they are in engage-at-will. I might be wrong, but as I recall there was a better swarming effect in STW.

Generals in STW and MTW multiplayer had 6 hitpoints, and more than that in SP depending on their health. The superman kings were also affected by battlefield upgrades which increased their combat power making them even more difficult to dispatch especially since they were usually a strong unit to begin with. In MTW/VI v2.01, LongJohn removed battlefield upgrades from multiplayer because we successfully argued that they could swamp out the anti-cav bonus and players were using a trick of taking good units such as chiv knight cavalry at valor = 0, and then getting free valor upgrades, which would otherwise be very expensive, by using those valor 0 units against weaker units. It was reported by multiplayers that 3 man cav units were beating 30+ man anti-cav infantry units towards the end of a battle. Now we see that battle field upgrades have made a return in RTW/BI multiplayer. What happened to the agreement between CA and the multiplayers that battlefield upgrades were not a good thing to have in multiplayer?

Quote:

---

Originally Posted by sunsmountain

With permission, Puzz, I'm sure you are not biased, but did you mouse over the 3 cohorts who suffered 291 kills to make sure they contained (367-291)/3 = 25.3 men on average versus 80 on the other side? If you simply used the battle results page (which pops up once the enemy army routs) the 291 number may have been polluted by simulated routing kills even though you didn't kill any routers (assuming we're in custom battle). Still though, the results are staggeringly close to random results. Was the difficulty at normal?

---

Normal difficulty in custom battle which isn't the best since the test should be carried out on LAN or online battle to eliminate the AI. I didn't end the battle prematurely so I don't think I got any simulated routing kills, but there is a possiblity of an unknown effect since the result was in the AI's favor both times. I actually didn't run the test to look for that. I only ran it to look for triple attacks.

Quote:

---

Originally Posted by sunsmountain

I therefore do not share this concern.

---

I just recently saw a post by someone at .com raising a concern about the game allowing more than 2 on 1 fights and how finishing moves might affect the combat. It has been mentioned here before that a man is probably invincible while he carries out the finishing move animation. If so, this would clearly be another artificial effect introduced into the combat model.

None of the stuff I've brought up in this thread will be a concern to players who don't care about reasonably predictable combat results which is apparently the vast majority of players. It just doesn't really matter anymore. Right now I'm no longer willing to pay $50 for gameplay that I will probably find to be irritating, and I certainly can't recommend that the 10 members of my clan should purchase the game so that we can all play online again.

I think you should cool down Puzz3D. I respect you and your opinions, but you don't need to get mad over it.

I don't care how they do it, but I do care that it is a game that can be understood. And, for what I know, RTW is "really" predictable (Sometimes it can be unclear, but then, it is experience that differs), and I guess M2: TW will be the same.

Quote:

---

Originally Posted by x-dANGEr

I don't care how they do it, but I do care that it is a game that can be understood. And, for what I know, RTW is "really" predictable (Sometimes it can be unclear, but then, it is experience that differs), and I guess M2: TW will be the same.

---

One side winning a perfectly balanced encounter by over 2 to 1 is predictable? The result could easily flip the other way. This is called chaos, and it works against thoughtful gameplay.

Quote:

---

Originally Posted by Puzz3D

One side winning a perfectly balanced encounter by over 2 to 1 is predictable? The result could easily flip the other way. This is called chaos, and it works against thoughtful gameplay.

---

I wouldnt mind an occasional lopsided result such as you describe, say 1-20 battles. After all, its not chess we're playing.

Quote:

---

Originally Posted by SpencerH

I wouldnt mind an occasional lopsided result such as you describe, say 1-20 battles. After all, its not chess we're playing.

---

You can still get that lopsided result on your own, however, simply by outmaneuvering and outthinking the enemy--which is originally what Total War battles were all about (and still should be, IMO).

If two units of 60 men are perfectly identical in every way--same type, same valour, same weapon/armor upgrades, they're facing each other on the same terrain (i.e., not on a hill), etc.--then they should lose men at roughly the same rate.

The only thing that would really change that is if the "random"-number added to attack/defense rolls ended up favoring one unit over the other. Now of course this can and will happen from time to time (since statistics allows for clustering of unlikely events), and that's perfectly fine. But the vast majority of the time, two units (or groups of units) that are identical to each other and attack each other head on should suffer losses at approximately the same rate.

Eventually one side will win (because of the random dice rolls), but it's usually not going to be by a huge margin--nor should it be. I'm not a numbers guy like Puzz, but I still understand the basic concept behind what he's saying.

Well, basically all my experience comes from MP playing, and that's my reference when talking about battles, their speed and balance.

And so.

Quote:

---

One side winning a perfectly balanced encounter by over 2 to 1 is predictable? The result could easily flip the other way. This is called chaos, and it works against thoughtful gameplay.

---

It is predictable that a side that flanks at the right time in the right spot through the melee encounter will win. Or a side that manages to keep his men off dart-fire, or missile fire, or even safe till the last moment, or a man who retreats his whole infantry line before contact because the morale level of his men is low, and then strikes back with an organised charge/flank setup..

@Martok: If you put 2 identical units against each other with identical circumstances, you won't get much of a difference in their men rate at the end of the battle. Though, if you enter moves such as flanking, "messing" (Messing that unit's formation), you might get what you're talking about. And, that is normal.

Of course tactical movement, terrain, etc should be able to effect the battle outcome between two identical units. I'm simply stating that a battle is not a chess match and that there should be an 'imponderable' occasionally included in the battle calcs. Clustering of random numbers may do it but given the large number of 'rolls' required to perform a unit-unit battle I doubt it (and I havent seen it happen). It would make it interesting if just once in a blue moon a heavy cav unit got repulsed by peasants.

Perhaps but then we'd have people on the forums saying "WHY DID MY GOTHIC KNIGHTS GET DESTROYED BY PEASANTS? CA WHY? DIE IN A FIRE!!!"

After the charge - why not? Knights can (relatively) easily get pulled off their horses. Maybe then horses then might be used to charge, then regroup, and recharge again!

And from that, why nt have that as the cavalry unit's standard "attack", and for them to remain in melee requires the alt btton to be depressed.

~:smoking:

I think a way to do it would be to have the possibility for a unit morale upgrade or downgrade once a particular unit v unit combat has commenced.

Quote:

---

Originally Posted by SpencerH

I'm simply stating that a battle is not a chess match and that there should be an 'imponderable' occasionally included in the battle calcs. Clustering of random numbers may do it but given the large number of 'rolls' required to perform a unit-unit battle I doubt it (and I havent seen it happen).

---

I've seen it happen, and it's repeatable in tests. With a 2 combat point advantage (44% stronger) in STW using 60 man units, a unit will loose 1 out of 10 fights. With a 1 combat point advantage (20% stronger) a unit will loose 4 out of 10.

When I say that I havent seen it happen I was refering to a totally unpredictable effect ie regular peasants beating knights (for example) on an otherwise even playing field. A unit with a 44% advantage losing 1/10 and a A unit with a 20% advantage losing 4/10 seems a little skewed toward the weaker but does not seem to include an 'imponderable' factor.

Spencer, that does happen you know.. I remember awkward times when my bunch of catas got beaten by 1 Praetorian Cavalry unit..

But I assume that was in a campaign battle where the cata's could have been low morale from loss of general, winded, charging uphill against a fresh high morale Praetorian Cavalry unit, not in an 'even playing field' custom battle test.

In any case, the fact that Puzz gets lopsided battle results in a custom battle test suggests that CA have included an 'imponderable factor' in the battle calcs and as you have suggested there may be a significant effect on unit v unit combat from the inclusion of a 'random number'.

All I'm saying is that if Puzz did his 3 v 3 custom battle 1000 times I would expect to see kills results that followed a normal distribution with a tight SD but what I'd also like to see is some 'outlier' results and even some 'extreme outliers' (such as the results he described) to take into account the weird 'stuff' that happens in battle.

EDIT and what I'd like to know is that the outliers occurred on purpose and weren't just another bug.

Quote:

---

Originally Posted by SpencerH

In any case, the fact that Puzz gets lopsided battle results in a custom battle test suggests that CA have included an 'imponderable factor' in the battle calcs and as you have suggested there may be a significant effect on unit v unit combat from the inclusion of a 'random number'.

---

I don't think it suggests that. It suggests that the uncertainty in the combat model is high. This means that tactical advantages have to have strong enough effect to overcome that uncertainty. This leads to the kind of black and white tactical gameplay that's in RTW where for instance, striking at a man's back is practically a sure kill, or routers can be mowed down like a lawnmower going over grass or shields provide an inordinate amount of protection from arrows to a unit that otherwise gets obliterated by arrows.

You can't make intelligent tactical decisions if what should be good decisions are thrown back in your face as mistakes. It isn't as though you can recover from a major mistake in a fast moving battle. Having a cheap enemy peasant unit beat your expensive knight unit is going to cost you the battle in MP. If you make it a rare event, it's blatantly unfair to one player. If you make it a likely event to remove the unfairness, the battles will devolve into a game chance.

We know that compared to STW the anti-cav bonus was reduced in RTW. That isn't logical in a game that has an increased level of uncertainty. I doubt the game is being "designed" to provide a certain level of statistical uncertainly. It's as though that's just falling out by happenstance as a result of some other decision making process such as, "We have to speed up the battles to appeal to a wider audience.". I've certainly seen plenty of interviews and posts by CA employees justifying the "new" gameplay with the argument that's it's being done to appeal to a wider audience. I would say that wider audience is oblivious to statistics, and that allows poor design in this area of the game to be satisfactory.

Just look at the load/save issue to see what this company tried to put over on the customer. Remember they said it was "designed" to work that way. The only reason it was changed is because of the bad publicity that resulted once it was discovered that important parameters were not being saved in the savegame file. CA can make the game any way they want, but I hope it's clear to people that the Total War series is no longer a serious strategy/tactical game.

M:TW was supposed to have all the good things you have been talking about, right? It was supposed to be predictable and provide a deep multiplayer experience.
In some custom battle tests I found M:TW to be just as random as you make R:TW out to be. Pitting Chivalric Sergeants against Saracen Infantry (two units with equal stats) produced massively one-sided results: four out of five times the CSs won, with about sixty men left in the unit while the Saracens had only 25 or so left. The one time the Saracens won, they had only slightly more men than the CSs (whose general happened to die).
Making things even more equal out of personal curiosity, I pitted two units of CSs against one another. Results varied, but what almost always happened was that one of the units would take many more casualties than the other. Charge bonuses seemed to be of low consequence (the charge being stopped the second of impact due to high defence rating).
So if even units with high defense and low attack produce results this skewed in M:TW, perhaps R:TW isn't so far off? The way I see it is, it's not a given which of two units will lose. But once it has begun to lose, penalties accrue due to being outnumbered (if only slightly, and very locally), and in the long run, these give results that are not consistent with units being exactly equal from the start out.
Or are they consistent? Isn't it, in a way, realistic for a unit that has begun to lose, to continue losing, and losing badly in the end? It's the same we see in women's tennis: Even between roughly equal rivals, it's rarely more than two sets, and these are usually quite clear. Even if it's three sets, tie-breaks are rare.

Historical realism is another venue to explain these findings: It wasn't uncommon for one army to be routed and decimated by a smaller one or by an equally strong one, with no historian being able to identify the cause. I think it's just human to misinterpret the general situation and develop a panic that spreads through the ranks.
Warfare in medieval times was a VERY chancy and risky affair. That is the reason there were so few, yes: FEW, big battles during the whole period. Most warfare was economic and political: Damage the other one's assets (land, peasants, castles) enough so he can't sustain his armies or at least can't keep his allies from frowning on him. Rare was the time it came to a big, decisive battle, and these were even more rarely decided by the generals' decisions. The way some battles played out, it seems one side had acknowledged from the start that they had no chance, and after some fighting quit the field. It's not really human to create a great carnage and stay in the fight, still hoping to win, after half of both armies are dead. There were exceptions, like the battle of Bosworth in the War of the Roses, but these were, again, rare.