[glob2-devel] A New AI

[Bradley Arsenault]

I may sound insane, but I want to start *yet another* AI for
globulation 2. I am an AI freak, I love 'em, and can't gen enough of
'em. I have made Nicowar, I have learned from my experiences, and I
want to start over. But Nicowar is definitly a success, certainly a
formidable one that, with a little more fine tuning, would play very
well. So I don't want to get rid of Nicowar. Instead I want to take a
different approach to AI development, one that is far less
generalized. I find that I made Nicowar too general, it builds all
buildings in the same way, upgrades them in the same way, it treats
all inns equally. However, there is too much "special cases" in making
an AI that is to be the best, Nicowar, however well designed the code
may be, is not the best way to achieve the best AI. In all regards,
more hackish systems like Warrush and Castor seem to work better
because its much easier to handle edge cases.

So, I want to finish Nicowar. Literally. I want to make it
crash-invincible, and I want to get it to work with Net games. It also
needs to become somewhat faster, right now, its a real proccessor
whore. I want to fine tune it somewhat to get better results mainly, I
want it to put more focus on upgrades than construction. I'll document
the code, and leave it.

All that being said, now is really the time for people to tell me
absolutely any crash that theyh so much as think is Nicowar related,
so long as they have some usefull information, such as gdb output
after the command "bt", or the last 20 lines of its console output,
along with all information about the map and enemies chosen. If anyone
has the time to test net games with me, I could use it. As well, if
people know how to make a profile with gprof, or feel like learning
(just ask me how), getting profiles on a variety of systems may help
to make it faster. The fine tuning bit is all my responsibility.


As for this new AI, I won't be making it for 1.0 unless I'm really
lucky and it turns out great within a few months. Its going to be much
more specific, using a Fuzzy State Machine based design right out of
my AI handbook. My aim with this one will to be powerfull, and I will
have to embrace many edge cases to get that to happen. Exactly what
kind of strategies I'm going to use is an open question, as there are
many different approaches one can take to win a game of glob2.



     Thanks for everyones continued support in the project, and my
contributations to the project (special thanks, k776 has been very
helpfull lately)