[3.1+] NullBot: an adaptive skirmish AI

[Goth Zagog-Thou]

Glad to see Nullbot is ever-evolving. A worthy project indeed, NoQ.

Vicious, tenacious, cruel and downright rude at the same time .. everything an AI should be. Best Warzone AI ever.

Two thumbs up!

[NoQ]

Glad to see you too Goth (:

[NoQ]

v2.11 fixes broken EASY difficulty, fixes always-been-broken-but-nobody-noticed Turtle AI on 3.2, and contains more minor unit movement fixes.

[Goth Zagog-Thou]

Thanks for the fixes, NoQ.

[Iluvalar]

So i tried it... pretty good. But then it built a second research lab and I instantly won of course...

I'd really love to see a new personnaliy who before it reach 80% of the map oil production:
A) Don't research anything but mg upgrade
B) Don't build anything but FFFRCc and the power gen needed. Or if anything a 4rth Factory...

Really, I'm a bit rusty and you might get along with researching htrack. But if I was practicing that opening you wouldn't get any chance to reach that point.

[NoQ]

It builds the second lab after minute 8, and by this time he's already having his army crashed even when i start with lab-factory-lab-factory. So i see no point for balance testing in optimizing anything beyond that mark. And for fun of those who don't care yet, i think it's a good idea to start researching extra stuff earlier.

In fact how easily he dies against lab-first makes me still unsure about your primary statement. I'd still like to have somebody show me that in action; i'd be more active in the chat starting from the next week.

[Iluvalar]

I caught an error on his part.

I had 3 tanks, he had 3 tanks in the center and he pursued me. He should have known that my next 3 tanks would join my army before his next 3 tanks if he chase me toward my base.

Let's :
d : distance between 2 base
x : His unit count on front line
y : The number of unit he know I have

Then he should not venture d/2*x/y away from his base. He would be in a position where he could get outnumbered if he do so.

[NoQ]

Testing mirror strategies against an A.I. is absolutely pointless.
He dies even if he has more tanks.

As any non-mirror strategy involves teching and having less tanks, i'm making him rush blindly, at least until minute 5 mark (i win easily, barely possessing 2-5 tanks at this point).

Making an AI that would consider enemy army size before attacking should be possible, but it'd take *much* more effort than it has already taken to fix some easier problems with the army movement since v2.06. And, of course, distance from the base is not well-defined (need a much more powerful topology/pathfinder code).

[themac]

I really love NullBot. I spend most of the time with online play against NullBot. I sometimes wish NulBot would better know his current strength. He often could kill a human player easy but he moves forward and backward in a quite defensive way even if the human has nearly nothing and would never survive when NullBot would fall into his base.

[Iluvalar]

NoQ The idea of the rush strat is to outnumber your opponent. If you reach the enemy base who have only 2 factory but had time while the travel to build a second batch of units. By all mean you dont engage in a 3v4. You siege him in his base until A) he is forced to chase you into the next 3 units for a 6v4 or B) your reinforcement arrives and you reach a better units ratio (15vs12). Anyway, at this point you control all the map but his extremity of the map.

[Prince Blueblood]

By the way, why every late game NullBot AI end up using lasers? Almost exclusively to the boot? Did that was done in response to I am using exclusively lasers too or...

[NoQ]

Because they were intended as the exclusive late-game anti-cyborg direct weapons. There is a balance problem though: there are no good cyborgs in late-game. So, as long as adaptation is enabled at all (HARD difficulty), it shouldn't make lasers unless you make cyborgs.

NoQ The idea of the rush strat is to outnumber your opponent.

I understand, but that's much more coding than i want to do on the old NullBot code; even simply rushing forward took so much to implement and there are most likely still bugs. A generic solution is required to work on all game styles (team matches?) and i'm not sure it'd eventually make more correct decisions than before, and it's not the biggest problem anyway: the real problem is what to do when you don't rush, which derrick to hit and where to move (there's no such thing as "avoid enemy army", "split up and attack multiple targets", "concentrate fire on the easiest to hit derrick", we can only "move forward" and "move back" without good geography engine).

I'd rather leave this project to NullBot3 (where i need to fix adaptation engine, then it should be much easier to implement fancy stuff).

Seriously, this is not the biggest problem. At 15 vs. 12 equal tanks, he would still loose due to poor tactics. Particularly, on Startup when it tries to climb the hill near base 0 to take the derrick on it, he already lost because catching him at the bottom of the hill is extremely effective, and he also spends a lot of time on climbing up.

So i tweaked the attack timing against a single specific build order i expected it to clearly win against - the lab-halftracks-first.

[Prince Blueblood]

yeah, I spam super pulse laser cyborgs late game, and NullBot do the same (difficulty HARD), it's really an epic fight... now, if only there's nuclear suicide bomber here...

[nick87720z]

Hello. I have seen info about experimental nullbot 3.x (more exactly - 3.02), but can't search for it on this forum, since it says about too common words. It was mentioned on warzone2100.org.ua resource, written on russian:
http://warzone2100.org.ua/pfsonsite/new ... admore=120 (about 3.02) from June 05 2013
http://warzone2100.org.ua/pfsonsite/new ... admore=112 (for all 3.0 features) from March 31 2013
It is said, it is incompatible with 3.1, only 3.2 or master. Tried to run on 3.1.2, but it froze game after nearly second or less.

And git doesn't have branch called 3.x, only 1.x and master (which is i guess 2.11).
Did it indeed take place? May be it was merged into master, or it is development only from them? They place all files on their site, and its all, i could find about nullbot 3.x and and mod versions itself (honesly, i experienced that news randomly). It it is not obsoleted, would be nice to add info to top message.

[nick87720z]

Just found previous [3.2] post about nullbot 3. Would be nice to add cross-link top posts from both [3.2] and [3.1].