adaptive AI in RTS's

[Calax]

I was playing through homeworld 2 again and one of the major points they liked to tote was that it had "adaptive AI". but their form of Adapive AI is to have the ai scan what your fleet contains and simply increase the # of ships they have in proportion to what you've got.

 

For example. In one of the missions I did a little toying with the stats (via notepad it's really easy) so that my fleet contained somthing akin to 50 battlecrusiers (biggest thing in the fleet. Mops the floor with everything else). First thing that happened was I got attacked by a HUGE fleet.

 

But the AI adapts to the makup of your fleet so my titanic number of battlecrusiers was attacked by about 100X their # in Interceptors (the most basic unit in the entire game) This promptly turned my game into a crawl and If I had let it the battle would have taken about 12 hrs to complete because the battlecruisers don't have much in the way of antifighter weaponry.

 

Would it be easy to change the AI to be a bit more adaptive than the above example?

 

BTW if your looking for a brutally difficult game in terms of single player Homeworld 2 is your baby.

[SamuraiGaijin]

My only RTS experience is with the StarCraft and WarCraft series, so I admit my "RTS AI" experience is limited, but a dozen or so years ago, I was involved in an AI programming experiment ...

 

Ultimately, what it sounds like Homeworld 2 is doing is scaling the opposing fleet to counter yours, probably with a "if he brings A, bring 100xB" - this is not "AI" ... AI would be the opposition adapting to it's own success and failure by modifying it's own tactics and force composition based on its own experience ...

 

For example - you bring a gun to a knife fight, and kill your opponent before he gets in range - assuming zero experience before the fight - an "adaptive" AI would respond by either: wearing a bullet-proof vest, arming with a longer-range weapon, hiding until you were close enough, or bringing along enough friends so that at least one could survive long enough to stab you - all this would depend on the resources available (can he buy better gear, or is training in order) and the opponent's priority (is killing you worth twenty of his attackers, or is survival of each unit important to him).

 

While that is an extremely simple example, you can imagine how complex stuff would get with all sorts of weapons, armor, shields, etc. available ... take an example from KOTOR ... the Sith sends an army of droids to capture you, which you swiftly kill off with ion weapons and vibroblades ... next time, they change tactics, but how? Shields capable of resisting ion attacks, people instead of droids, "guerilla" droids that conduct hit-and-run attacks trying to lead you into a minefield (or into a Sith Assassin ambush), etc. etc.

 

I think it's a great idea, but holy crap would it be difficult to implement well - especially if you consider each unit in the game is going to have different experiences (and different levels of experience, the depth of which should be reflected in their decision-making initially).

 

What if they have prior knowledge of some of your preferred tactics and equipment - how would that impact their tactics and equipment?

 

Now all that is just really "reactive" AI ...

 

Now, take this a few steps further ... what if they planned a few iterations in advance (i.e.: a chess player that plans ten or more moves in advance) ... what if every enemy (and friendly, and neutral) unit had the some level of capability to evaluate the possible outcome a few (or several) "moves" in advance ... now we're talking about not only "learning" machines, but "thinking" machines ...

 

So, short answer to your question ... would it be easy? With limited variables to consider, yes, but the "realism" of the game would drop dramatically.

 

To make it realistic, with complex environments, numerous units, various armament and defenses, different motivation/priorities, and differing levels of tactical experience/strategic foresight/resource management skill/etc. - holy crap would it be hard.

[alanschu]

One of the biggest concerns with adaptive AI is that it's not finite. How do you verify that it runs flawlessly?

[Tigranes]

You cannot, you simply confirm its probability of success both practically and theoretically on a reasonable level. Just like most scientific theories.

[Musopticon?]

I think Supreme Commander uses that. Although I don't think it's very balanced or easy to play against an opponent who can divide it's attention on almost infinite points of interest over the map, micro-manage everything and use cold logic almost refectly. And make all that on a global scale.

[ShadowPaladin V1.0]

When the AI makes a mistake people tend to fixate on that as bad AI rather than a mistake which it just made. Of course there is bad AI as well, numerous examples of it that just does the same dumb thing over and over and never learns.

 

The most I've enjoyed AI was in Star Ocean. It starts off completely moronic and will have you cursing, but after training it will give you a run for your money.

[CaptainPants]

I think Supreme Commander uses that. Although I don't think it's very balanced or easy to play against an opponent who can divide it's attention on almost infinite points of interest over the map, micro-manage everything and use cold logic almost refectly. And make all that on a global scale.

<{POST_SNAPBACK}>

 

On the other hand, it can't actually think.

[Musopticon?]

I liked Thief 3's AI, which reacted to it's surroundings and communicated with other AIs. For example; if a guard saw you at kitchen, on top of an armoire; he'd first acknowledge that you really are there, understand that he can't catch you from there(some furniture was unbreakable), alert other guards and run to get an archer/mage. They also reacted accordingly and performed thorough searches on the adjacent rooms, if you had managed to dispppear(as usually was the case) during the time that guard took to get support.

[Musopticon?]

I think Supreme Commander uses that. Although I don't think it's very balanced or easy to play against an opponent who can divide it's attention on almost infinite points of interest over the map, micro-manage everything and use cold logic almost refectly. And make all that on a global scale.

<{POST_SNAPBACK}>

 

On the other hand, it can't actually think.

<{POST_SNAPBACK}>

True enough. I remember fooling AoE 2's AI to make hordes of useless anti-archers. The adaptability is relative.

[alanschu]

You cannot, you simply confirm its probability of success both practically and theoretically on a reasonable level. Just like most scientific theories.

<{POST_SNAPBACK}>

 

 

My question was pretty rhetorical. Though you do get marks for the correct answer

[Calax]

just did the level again... I had an army of dreadnaughts (your technically only supposed to have one) and battlecruisers. They attacked me with destroyers...

 

I didn't loosse a single ship.

 

Next map they had about 50 battle cruisers which wouldn't be so bad except for the fact I was supposed to save somthing that was under attack by about 80 frigates. That was nasty and I decided to quit.