Throwing out a question concerning AI

[Llyranor]

RTS are meant to be mainstream, of course they'll cater to the playerbase who wants at least some frenetic action in their experience.

 

Play wargames. The AI there is pretty good oftentimes, even in real-time engines.

[Llyranor]

Imagine, if you will, a game taking place entirely in real-time. Let's pretend it's a RTS. You have your units and stuff.

 

Let's implement command and control on those units, and implement some sort of chain of command. So, you have your HQ, which has its own set of sub-HQ's, who each have their own subunits, and so on and so forth.

 

In terms of how you issue commands, you can essentially select any unit along the chain and give it to them. If you give it to a unit, the order applies to all of its subunits (unless you've detached some of them and put them directly under 'your' control - meaning the lead HQ). The AI takes into consideration what units it has under its commands and then formulates a plan according to that. For example, if it has a good amount of units, it'll set aside some reserve troops, and set up multiple fronts, perhaps send off a few to cover the flanks, and so on.

 

In terms of how much you have to click, it's pretty quick. There's no micromanaging every single thing the unit/s do/es. Just select the general pathway, give a bit of info on how you want to carry out the order (shortest route, fastest route, covered route, etc), and the AI will carry it out to the best of its abilities (again, utilizing its resources/units appropriatetely for covering flanks, artillery back-up on-call, etc).

 

Here's a, say, example of what the command menu would look like. Simple, efficient, and effective. You set what you want, give the general waypoints, and off you go.

 

Now, what's the point of all this? Now, again, this is an issue of command and control. Let's say you don't absolute perfect control over every single unit at any time you want (which, perhaps, is the reason why RTS may be clickfests in the first place). Let's implement something called order delays. For every order you give, there is a time delay before it gets executed. This is pretty much the simulation of the order going down the chain and being planned out. After all, getting hundreds of units from one position to the next isn't necessarily a walk in the park. It takes time. Each specific unit has a certain 'time delay' associated with it, and when you issue an order to it, it applies to all its subunits.

 

Let's also say each HQ/unit has a certain capacity load. They can only command so many units at once (which may apply if you attach more units to them than their initial subunits), and the more units you add to their command, the longer the order delay may take (representating, again, the amount of resources it takes to command larger armies).

 

Let's use this example. Here you see the max capacity of the unit at 16, and that it currently has 4 units under its *DIRECT* command (meaning that each of those 4 units could have their own subunits, but it wouldn't affect the load of THIS unit). It shows the unit delay of this specific unit, and also the total time delay for the whole force under this unit's command (going throughout the whole chain of command).

 

You can, also, in specific situations, put units directly under 'your' command. In which case the unit is removed from its previous HQ's load and put under the lead HQ's. Since there's less of a chain of command, the time delay is less.

 

I went too much into detail, but what's the point of all this? Essentially, you can't clickfest your way to victory. It's not even remotely possible. First off, the time delay means that you can click as quickly as you can, but it won't have any effect until, say, 30-60 min (in-game time) from now. Second, each time you click more and issue new orders, there's a NEW time delay, which means the less you CHANGE your initial plan, the less you screw up your logistics. This is where superior foresight and planning comes into play. The player who can better predict the enemy's move and creates a plan according to that is the one who has the edge. It doesn't matter who clicks faster. Given that no plan survives contact with the enemy, hehehehe. This is essentially a contest of actual decision-making, not clicking.

 

Adding the capacity load system to the order delay system is pretty nice in the sense that micromanaging all your units could in fact be detrimental, since the lead HQ only has so much capacity load. You are essentially, the commanding officer. You deleguate orders. You do not specifically order every single soldier order. It happens at a higher level.

 

Oh, and I guess you could pause the game and issue commands if you wanted to, too.

 

There we go. Pure strategy/tactics (depending on the scope of the game). No clickfesting. Period. All in real-time. Heck, you couldn't even do this in TB (unless it were done very awkwardly). The whole point is that RT vs TB is moot. It all depends on game design.

 

Will we ever see this in a maintream RTS, though? Hahahahaha.

Edited January 27, 2007 by Llyranor

[kormesios]

Kasparov's issue was that Deep Blue wasn't designed to be a good chess player, but rather it was designed specifically to be Kasparov. 

 

Sorry for jumping in late, but I thought this was worth emphasizing.

 

The truth is, despite the huge computational powers computers have, it's still more or less a wash whether a computer will outplay a human grandmaster. And this, despite the fact that chess has well worked out strategies and millions of words describing them for programmers to draw on, so programmers know roughly what every human approach is.

 

Humans are just very good at some things, like pattern recognition and sensing "good enough" solutions, that computers are bad at.

 

Two other points, mentioned already, deserve more detail:

 

(1) The goal is to make an AI that's fun to play against, not one that always wins. Humans often find strategies after a game is published that gives them a significant advantage at one thing or another, then exploit it mercilessly. If a computer algorithm found some strategy which allowed it to win consistently, programmers would immediately prohibit it from using it.

 

(2) With Deep Blue, programmers adapted it to Kasparov. But usually it goes the other way--the player can adapt, the game AI sticks to the same set of algorithms. Once you find a weakness, even if it's subtle, it's always there. An idiot AI may do a frontal assault all the time, and if you know that it's easy to stop. A "clever" AI may analyse the situation and do a flanking assault if appropriate. This is better, but a good player will eventually learn when the AI decides which it's going to do, then trick the AI into trying the wrong one, counter attack, and beat it like a red-headed stepchild.

[Morgoth]

The best RTS AI is still the one that always lets me win!

[metadigital]

...

I went too much into detail, but what's the point of all this?

...

There we go. Pure strategy/tactics (depending on the scope of the game). No clickfesting. Period. All in real-time. Heck, you couldn't even do this in TB (unless it were done very awkwardly). The whole point is that RT vs TB is moot. It all depends on game design.

 

Will we ever see this in a maintream RTS, though? Hahahahaha.

<{POST_SNAPBACK}>

What that game called again?

 

(2) With Deep Blue, programmers adapted it to Kasparov.  But usually it goes the other way--the player can adapt, the game AI sticks to the same set of algorithms.  Once you find a weakness, even if it's subtle, it's always there.  An idiot AI may do a frontal assault all the time, and if you know that it's easy to stop.  A "clever" AI may analyse the situation and do a flanking assault if appropriate.  This is better, but a good player will eventually learn when the AI decides which it's going to do, then trick the AI into trying the wrong one, counter attack, and beat it like a red-headed stepchild.

<{POST_SNAPBACK}>

Your point only holds for the type of computing we use today. Neural nets and massively parallel computer paradigms will more effectively battle with the human pattern-matching heuristics and even eventually beat them.

[CrimsonAngel]

If you look at a game like Galactic Civilization 2 it actually have a pretty good "AI" so their are some games out their.

 

After all the better the Ai the more fun the replay value is.

If every fight and battle is different then things become more fun.

 

The problem is some of the bigger publishers goes for the easy cash magnet.

[Llyranor]

...

I went too much into detail, but what's the point of all this?

...

There we go. Pure strategy/tactics (depending on the scope of the game). No clickfesting. Period. All in real-time. Heck, you couldn't even do this in TB (unless it were done very awkwardly). The whole point is that RT vs TB is moot. It all depends on game design.

 

Will we ever see this in a maintream RTS, though? Hahahahaha.

<{POST_SNAPBACK}>

What that game called again?

It's the Airborne Assault series. Published by http://www.matrixgames.com (same dudes publishing War in the Pacific, alanschu's latest fling). The two games available right now are Highway to the Reich (Operation Market Garden) and Conquest of the Aegean (invasion of Greece/Crete).

[alanschu]

RTS are meant to be mainstream, of course they'll cater to the playerbase who wants at least some frenetic action in their experience.

 

Play wargames. The AI there is pretty good oftentimes, even in real-time engines.

<{POST_SNAPBACK}>

 

 

Agreed, Close Combat was a very fun game to play, and it was real-time.

 

 

As Llyranor pointed out, an RTS does not need to be a clickfest. I think the game he was describing was Red Devils Over Arnhem.

 

When I worked on ORTS, we were looking on making an intelligent AI that worked well tactically from both sides. This was so that players were no longer small scale tacticians that had to click 1000x a second, but the strategic scale Generals.

[Llyranor]

Yeah, Red Devils was the first game of the series, though it's not available anymore ('twas a different publisher).

[metadigital]

...

I went too much into detail, but what's the point of all this?

...

There we go. Pure strategy/tactics (depending on the scope of the game). No clickfesting. Period. All in real-time. Heck, you couldn't even do this in TB (unless it were done very awkwardly). The whole point is that RT vs TB is moot. It all depends on game design.

 

Will we ever see this in a maintream RTS, though? Hahahahaha.

<{POST_SNAPBACK}>

What that game called again?

It's the Airborne Assault series. Published by http://www.matrixgames.com (same dudes publishing War in the Pacific, alanschu's latest fling). The two games available right now are Highway to the Reich (Operation Market Garden) and Conquest of the Aegean (invasion of Greece/Crete).

<{POST_SNAPBACK}>

And what is "Arty direct support only" code for?

[Llyranor]

That's giving the order to any artillery unit being part of the group receiving the order to only directly support that group in carrying out that order. Otherwise, any artillery on-call in that group would be able to support any other friendlies - however the AI deems it appropriate (or you can manually order it around, again).

[metadigital]

I just found it amusing that the developers decided to use a familiar diminutive personification for the assets: "Arty". :D

[Plano Skywalker]

I'm not sure what is meant by "game lock combos" but there have been games with different AI types - of course, they still weren't dynamic in response to player-actions, they'd just use different "plans", and thus were predictable before long.

 

Dynamic reactions to human actions would be nice, yes...assuming, of course, that the programming made it feel random enough in some way. If it only had one reaction to any action - where I knew if I didn't want it to build a certain unit I just had to avoid doing a certain thing to cause the AI reaction - that would be just as predictable, I think.

<{POST_SNAPBACK}>

 

 

here might be an example of what I mean. take the game Stronghold. anyone who has ever played that game knows that wheat farms + mills + bakeries = the best food production in the game. sometimes, however, you may decide to use other types of farms because you just like the aesthetic (of dairy, for instance) over wheat farms. or maybe you choose cattle because of the dual purpose (armor making).

 

if you were playing against an AI opponent (in which you both had the same tools), the AI opponet who either play stupid (intentionally) or go for every possible advantage (i.e. food would be wheat + mills + bakery).

 

it would never play a robust game with all dairy farms because that would amount to it being gimped in some way.

 

so, what I am saying is, we need AI personalities that consider things besides the mathematical advantage (he just likes building certain things because of the looks, whatever)...but still playing well and in direct response to what you are doing.

 

and with different, randomly-determined personalities emerging each time.

[alanschu]

it would never play a robust game with all dairy farms because that would amount to it being gimped in some way

 

I'm not sure what you mean here when you use the word robust. I can't make sense of this.

[Calax]

Recently the RTS genre as a genere has Split... If you look at most of the recent releases they are either Total War style empire builders or Down and dirty Company of Heroes microfests.

 

Nothing has hit the midpoint like C&C or Starcraft much since SC (warcraft three was a microfest for good players)

 

Supreme commander looks like it will bridge this gap but you never know until a game is out.

[Musopticon?]

RTS genre hasn't been doing this great since 2001, this could very well be called the third golden age. The genre is actually evolving!

[Xard]

RTS genre hasn't been doing this great since 2001, this could very well be called the third golden age. The genre is actually evolving!

<{POST_SNAPBACK}>

 

Amen

 

Supreme Commander will be teh second coming of Messiah! :joy:

[Hassat Hunter]

Some testers here then?

 

Personally I think SupCom still has too much micro-management quirks that, because of the scale, are worse than the quirks of other RTS.

 

As for AI in general (going FPS here; not much into RTS):

 

F.E.A.R's was only good because it linked certain interactive objects to actors (as to save processor time). While this is good for a hall-to-hall game like F.E.A.R it will be far less suitable for a non-linear game like Deus Ex or System Shock 2. IMO once you are found in DX they should never let you go and call for assistance from nearby friendlies etc. This *should* not make it impossible for the player... because you can still sneak around their searches as normal; or take them down ofcourse... oh well...

 

A good example of AI used as marketing would be Oblivion. Altough, thanks to Shadowstrider, we know they downgraded it to preserve "playability". If that is so why some random NPC's walking around in the city would still display quirks as they do (pick fruit in the middle of the city anyone?); I have no idea...

[alanschu]

F.E.A.R's was only good because it linked certain interactive objects to actors (as to save processor time). While this is good for a hall-to-hall game like F.E.A.R it will be far less suitable for a non-linear game like Deus Ex or System Shock 2. IMO once you are found in DX they should never let you go and call for assistance from nearby friendlies etc. This *should* not make it impossible for the player... because you can still sneak around their searches as normal; or take them down ofcourse... oh well...

 

The only thing I can think of is that design feels that it becomes too unforgiving.

 

Either that or there is some other issue that I'm not familiar with that makes keeping them in the alerted state problematic.

[Hassat Hunter]

Well; I got the impression the DX-creators wanted to have more of a sneakier game than all other FPS (like a thief); so unforgiving should be not an issue as most of the times you are not supposed to be spotted anyway; besides carefully made traps

[alanschu]

Deus Ex wanted the option for being sneaky, but still wanted the option of not being sneaky to be perfectly viable.

[Sand]

I want to play a game that I can sniper a german shepherd with a rocket propelled grenade and not be detected by enemy soldiers.

[alanschu]

Would that actually make sense? I'd think firing off an explosive projectile should set off some red flags.

[Llyranor]

Some testers here then?

 

Personally I think SupCom still has too much micro-management quirks that, because of the scale, are worse than the quirks of other RTS.

 

Nick and me played in the beta. We haven't touched it since the first week that was released. That speaks for itself.

 

 

Such high hopes.

[Sand]

Would that actually make sense?  I'd think firing off an explosive projectile should set off some red flags.

<{POST_SNAPBACK}>

 

I will use a silencer.