Can this be done? Im sure with the correct tools and the competant people a game like that can be achieved, it would be nice :)
I think battle AI will always be somewhat of an issue, but EU does show that campaign AI can be better at least. That game is also not turn based though, and you need turns if you want total war type battles, unless you have it real time but pause the campaign for the battles. I think the turn based system does hurt the AI though because it makes it very easy for the human to launch sneak attacks and surgical strikes that the AI could counter if it was in slow real time like EU, but cannot because of the turn based system. ETW and NTW did try to help with this issue a bit by making the zone of control around armies way bigger so they could reinforce from further away, and it does help in some cases, but it's not enough (particularly when the battle AI is so bad that you can easily defeat a force 3 times the size of yours).
Last edited by WinsingtonIII; 05-13-2010 at 00:39.
from Megas Methuselah, for some information on Greek colonies in Iberia.
I think battle AI will always be somewhat of an issue, but EU does show that campaign AI can be better at least. That game is also not turn based though, and you need turns if you want total war type battles, unless you have it real time but pause the campaign for the battles. I think the turn based system does hurt the AI though because it makes it very easy for the human to launch sneak attacks and surgical strikes that the AI could counter if it was in slow real time like EU, but cannot because of the turn based system. ETW and NTW did try to help with this issue a bit by making the zone of control around armies way bigger so they could reinforce from further away, and it does help in some cases, but it's not enough (particularly when the battle AI is so bad that you can easily defeat a force 3 times the size of yours).
I'm fairly certain a much better BattleAI is possible and isn't even that impossible to implement by CA. They just need to add a lot more variable parameters to account for more diverse battle situations. Right now, several of the biggest issues that I encountered in the battleAI (none of which are that hard to implement) are:
- RTW does not know how to skirmish using it's entire army. There's always that single unit that stands still. Attack that unit and even missile cavalry will join the frain. A clever AI will take into account the cost of sending it's missile cavalry into the fray (i.e. calculate what the likely damage in charging and melee is compared to the damage it could do if it just sacrificed that unit and continued to skirmish). The mathematical ideas necessary behind such reasoning were extensively developed some 30 years ago. I guess it's about time to implement them.
- RTW hardly takes into account fatigue. More often than not armies are winded or worse when they engage me. It's not that hard to tell an AI to rest it's army before entering the dead zone of my missiles.
- RTW doesn't take into account the likeliness of routing units, especially not her own routing units. It's not that hard to tell an AI it should prefer any other unit to charge but that unit that is already badly shaken due to exhaustion and incurred losses. These units should obviously be held somewhat in reserve and/or rested.
- RTW doesn't take into account the objectives of a battle. It merely tries to win. It's not hard (nor the result of pure genius) to consider the importance of the battle based on the position on the battle map. Invading armies should be hurt, regardless your own costs. While the invading armies of an AI should be more careful to their own losses, rather than pursuing victory. The only situation where RTW takes these considerations into account is when it is being besieged. It will often attempt suicide sallies, which is ok. (Though if such a sally fails, the AI should attempt to gain a draw to gain time, which it doesn't.)
- The formations I encounter are quite fine in their own right (I'm using MarcusCamillius', Darth's etc formations*, so I can't distinguish between these and basic RTW). But it doesn't requires a genial pair of brains to realise these formations should be formed according to both the terrain and the enemy army composition. At the moment, these are sadly lacking. The AI will attempt to form something of a 'best' formation imaginable, regardless of enemy opposition or terrain. Again: these could be easily fixed by a AI-program that calculates the expected losses vs kills with regard to the enemy army composition and the terrain and give each tactic a score. You could still allow the AI to start with a historical formation and initiate a formation-improving-program as the game starts. This way, the AI would (apart from fight) continuously seek to improve it's formation before attacking you.
- At this moment, most individual units hardly behave according to a plan. They react far too much on individual incentives. Hit a unit with missile fire, and the AI will react with one or two units. It should either react with her entire army (sending in everyone is undoubtly more efficient than sending in two units against an entire army) or it shouldn't react at all (hoping you'll soon run out of missiles before killing too many opponents, which makes your missile units worthless).
Of course, writing a battle AI on the level of a unit is significantly easier than writing one on the level of an army.
* No bad word about the modders who developed these formations though, as they undoubtly did the best they could given the hardcoded constraints they faced.
Ancient TWs and the like would be great. I for one would love a new Rome or Shogun. But what would be done with this new naval engine that they have so lovingly nurtured over the ETW and NTW titles? My lack of historical understanding tells me that ancient naval warfare might be somewhat bland.
Andy, I would be surprised and shocked if I found out you have not applied to work for CA, since you clearly known precisely what CA needs.
On a more serious note, please put aside the anthropocentric views. They aren't really handy when looking objectively at an issue such as AI. You bring up some issues in your post that aren't relevant to battle AI as much as they are to strategic AI. What you take for granted is one thing, what computers are created to do is a completely different one. Battle AI can do many things you'd expect it to do. It's quite more tangible and flexible in the hands of a programmer. With regards to strategic AI, the only ones winning in this field are the chess programmers. And even there one still sees a majority of the work put into pure tactical calculations. Again, to recap, don't expect much from a big-picture strategic standpoint. Battle AI can and has been improved heavily over the decades. Strategic AI can only act so much as a human does. Strategic games of the hybrid genre such as Rome are a very long way off from reaching your or my level, let alone Washington, Napoleon, or Eisenhower. Any higher expectations, I would say, border on ludicrous.
Andy, I would be surprised and shocked if I found out you have not applied to work for CA, since you clearly known precisely what CA needs.
I have to disappoint and shock you. ;)
On a more serious note, please put aside the anthropocentric views. They aren't really handy when looking objectively at an issue such as AI. You bring up some issues in your post that aren't relevant to battle AI as much as they are to strategic AI. What you take for granted is one thing, what computers are created to do is a completely different one. Battle AI can do many things you'd expect it to do. It's quite more tangible and flexible in the hands of a programmer. With regards to strategic AI, the only ones winning in this field are the chess programmers. And even there one still sees a majority of the work put into pure tactical calculations.
I don't really agree with you here. Give a computer time and CPU and it will quite easily find the best solution. It just needs to check all possible outcomes for several moves in a row. Higher difficulty levels on chess computers (at least on the one I'm experienced with), just means more turns the AI will take into account. It calculates for every possible series of moves the losses and takes, and simply follows the one with the highest outcome. It could be refined (meaning: checking which one of parallel outcomes is more likely to be achieved), but then we're no longer talking about the kind of chess-computer that is to be sold to any greater public.
Again, to recap, don't expect much from a big-picture strategic standpoint. Battle AI can and has been improved heavily over the decades. Strategic AI can only act so much as a human does. Strategic games of the hybrid genre such as Rome are a very long way off from reaching your or my level, let alone Washington, Napoleon, or Eisenhower. Any higher expectations, I would say, border on ludicrous.
I fail to see the difference between what's needed for a better strategic and a better battle-AI. Both use essentially the same algorithms, with the battle-AI the most complex ones (because she faces binomial and path-finding problems that need to be solved in real time).
In the underlying text I'm basicly arguing the differences between a better BAI and the current one are - for the larger part - quite easy to implement. We already have the needed aspects of a decent BAI, they just need a tiny bit more input to be remarkable more effective.
Spoiler Alert, click show to read:
The issue at hand is basicly a shortage of parameters the AI takes into account. If it did that, it's behaviour would improve significantly. Removing fatigue already helps the AI greatly. CA knows exactly by what formulae these battles are resolved. She knows the impact of lethality, weapon type, fatigue and morale and she has the formulae in which these numbers are put. The only thing she needs to do is to put all and not just some of the variables in the algorithms that make AI-spearmen attack my cavalry. They do that already simply because the BAI already calculates the efficiency of units in the battle (read: the BAI is already able to read my units. They just don't use all available information). Right now it's like the BAI doesn't know all the acceptable moves of chess, and therefore won't consider several moves herself. Even if these extra parameters would make the calculations to slow, they could easily be done before the start of the battle (i.e. create a matrix that reflects the lethality and efficiency of the units on the terrain when facing each unit, 20units*20units=400 numbers) and correct during the battle for terrain, morale and fatigue.
About the strategic considerations: Think about my remarks about the objectives of a battle as a suggestion for a number that tells the AI which ratio losses-kills are acceptable to her situation. Such a ratio would obviously be calculated on the strategic map (where the AI can decide what forces could be brought in should she loose). Import that single ratio into the BAI, and compare it to any plan the BAI wants to resolve. I.e. calculate the expected losses-kills based on position, morale, stats,... and compare these to the ratios that should be achieved. Right now the number the BAI uses to decide whether to fight or to retreat is a static number. Make it a variable based on the outcome of the CAI, and she would perform much better.
The biggest difficulty in such a 'battle plan' (as opposed to a more unit-based approach) is the number of units on the battlefield (as this number pretty much determines the number of available and reasonable plans). Given that this number of units hasn't increased that much since MTW1 (and further increasing that number isn't primordial either), given that heuristics to combine units can easily be developped (I believe the AI already calculates the movement-patterns for several units instead of each unit individually when lining up), I'm honestly convinced this could be realized.
I'm fairly certain a much better BattleAI is possible and isn't even that impossible to implement by CA. They just need to add a lot more variable parameters to account for more diverse battle situations.
Well, I'm sure it could be improved, but I'm not sure it will ever be up to the standards most people want.
from Megas Methuselah, for some information on Greek colonies in Iberia.
Did you notice this statement as well? It's from the article they link to on ign.com
"Though in possession of full details, our lives wouldn't be worth living if we divulged what we know, but let's just say that the new PC-only game eschews the gunpowder weaponry of recent Total War titles and gets back to basics"
So no 1800's... My guess (and most people's guess I think) is Rome 2. RTW was far too successful for them to not revisit the era. But, I guess you never know. Here's to hoping that CA's higher regard for historical accuracy that was shown in Napoleon will remain for Rome 2 (and ideally become a much higher regard).
from Megas Methuselah, for some information on Greek colonies in Iberia.
I too hope for a second Rome. Although I should probably be more concerned by the high pace at which they release new games, rather than excited about the sequal of a great game now about five years old.
I actually don't care what time period the next TW is about as long as it is as moddable as RTW. From RTW onwards the modding part of the game have gone more and more difficult.
Re: The next game in the total war series?
Originally Posted by lionhard
from Megas Methuselah, for some information on Greek colonies in Iberia.
Reply With Quote
from plutoboyz
Silence is beautiful
Bookmarks