Spring RTS Engine

location aware cheating is much worse then resource cheating

This. It's much more fun, if the AI can build huge armies that are deployed fairly stupidly, and can give us a game on that basis, than an AI that knows where everything is, can't be surprised, and has normal-sized armies.

With the first case, as a player, you may lose due to sheer numbers, which is fun, or win because of superior scouting, defense placement and information, which is also fun. With the second case, the minute you figure out what it keys on, you can demolish it every game.

Pleasee no cheat. It still needs to be fun to play which it won't be if it knows where your antinukes are, if it knows where your units are, if it knows where your mexes are and that it can easily destroy them becuase they're not guarded, ect ect.

Hoi wrote:Pleasee no cheat. It still needs to be fun to play which it won't be if it knows where your antinukes are, if it knows where your units are, if it knows where your mexes are and that it can easily destroy them becuase they're not guarded, ect ect.

The problem is not whether the AI knows these all things or not. The problem is what the AI will do with a knowledge of X building or unit in player's possession in place Y, and whether the AI will do some intelligent scouting for such targets in case it doesn't cheat.

KAI had it just right:

LoS hack, and knowing enemy composition, but NO resource hacking.

KAI was ace to fight because it monitored your units. if you didnt rush, it would eco. if you rushed, it would have defences ready. if you made no air, it made no AA. if you made air, it was already making AA.
this made it an excellent challenge. The feeling of being outsmarted this created was ace.

resource hacking just makes the AI feel stupid

Why can't it just take the time to scout, and let me keep some small part of my forces secret if I put effort into it?

As I said earlier. I think people are underestimating the difficulty of creating a good performing AI that doesn't use maphacks (also known as full observability). The problem is the uncertainty of the world. It has to create a model of where the enemy is in order to perform well. Without maphacks (or partial observability), this model would be some belief system gained through scouting. An example:

Lets say the AI scouts a group of enemy units at spot x, the scout dies (obviously). Next it wants to attack this group with its own group of offensive units. By the time this group has reached the known spot x, the enemy group has changed location and the group of units is basically useless there. Also, the scout will most likely observe a small part of the entire group before it dies, resulting in wrong estimates.

I'll come about that in a different way. What motivates you to code an AI, and what are your goals?

Error323 wrote:Lets say the AI scouts a group of enemy units at spot x, the scout dies (obviously).

Hmm why is that so obvious? Id guess it should be doable to just make the scout unit stay out of enemy range and hovering in and out of sight distance. In BA at least killing a jeffy thats microed is very hard, you can chase it away with scouts of your own but not kill it that easy (unless with several samson). Just make the scout avoid entering enemy attack range at any cost, unless it can win.

lurker wrote:I'll come about that in a different way. What motivates you to code an AI, and what are your goals?

from a players perspective, what I want from an AI is a challenging opponent that defeats me with intelligence, not excessive resources.

JohannesH wrote:Just make the scout avoid entering enemy attack range at any cost, unless it can win.

There lies the problem. You don't know the enemy range or its ability to win until you know the enemy... at which point the scout is dead (in a lot of cases). Scout ranges are usually very small compared to other combat units.

in BA, only unit that can easily kill that scout is samson/slasher. (Of units that are used at all)

lurker wrote:I'll come about that in a different way. What motivates you to code an AI, and what are your goals?

Motivations: Fun, Instructive (on multiple levels) and Interesting
Goals: 1) Defeat good human players with maphacks, not resource bonusstuff. 2) If one is achieved or this poll showed that Realism is more important, remove maphacks and try defeating human players again.

JohannesH wrote:in BA, only unit that can easily kill that scout is samson/slasher. (Of units that are used at all)

First of all this just isn't true, second even if it were true you are still making the assumption a scout always faces a single enemy unit.

Where did i make that assumption... And how am I wrong? IF the scout is a jeffy. No other relevant unit shoots further than the jeffy can see but the samsons. T2 and HLT are not relevant.

Interesting. My motivation to code AI is to implement behaviors and watch them in action, try to make something that can think in limited ways, watch interactions between actors, etc. Anyway, I understand the desire to make something hard to beat, but it moves a bit on the scale from AI to challenge scenario. Not necessarily bad, but it needs to be noted. If you want to begin with a strong focus on base building, counters, and battle tactics but cheat intel that's valid but changes the very nature of the game. And don't do fake scouting, that's just false representation.

I implemented a maphack in NTai after a logn period of makign NTai scout for information.

Why?

My scoutign algorithm was broken and I couldnt be arsed re-implementing it, I had work to do elsewhere in economy and refactoring, so I put a maphack in till I had the time to get back to it. I never did get back to it and the maphack remains in place.

IIRC krogothe did not want to bother with scouting either.

But for a long time all the major AIs had scouting, and they had them from the very beginning. The very first AI released on the forums JCAI had a scouting algorithm and scouted fine with planes, so did NTai OTAI AAI etc etc

Map hacks are a recent trend, they are not because scouting is hard.

Ugh,

Ok screw it, perhaps I'm being too pessimistic about the non-maphack stuff. I'm gonna try to build it in another branch. Wish me luck

AF wrote: Map hacks are a recent trend, they are not because scouting is hard.

lol

non-map hack AIs play terribly

all the old AIs (pre-KAI) were extemely weak. KAI with its maphacking blew everything else out of the water. IIm pretty sure there is no way to make a non LoS cheating AI- that dosnt resource hack- play in a way that is even slightly acceptable.

Summary:
SCOUTING WAS A DEAD END.

KAI was already ahead of the game, a suitable scouting algorithm and the removal of the maphack wouldn't have changed that.

But my point is that maphacks were not implemented because scouting is hard. Maphacks were implemented because AI devs could not be arsed writting or rewriting scouting. Scouting itself has benefits.

Error323 wrote:As I said earlier. I think people are underestimating the difficulty of creating a good performing AI that doesn't use maphacks (also known as full observability). The problem is the uncertainty of the world. It has to create a model of where the enemy is in order to perform well. Without maphacks (or partial observability), this model would be some belief system gained through scouting. An example:

Lets say the AI scouts a group of enemy units at spot x, the scout dies (obviously). Next it wants to attack this group with its own group of offensive units. By the time this group has reached the known spot x, the enemy group has changed location and the group of units is basically useless there. Also, the scout will most likely observe a small part of the entire group before it dies, resulting in wrong estimates.

sure it's not that easy, but...

if the scout sees some units at spot x (it should also note their speed and heading) then it will remember that and can extrapolate their position based on time since it last saw them similar to what a real player does when it sees some units moving around, or encounters stopped units.

so whenever your AI sees an enemy unit it should remember everything it knows about it and extrapolate certain info that it can then use in its predictions