![](smile.gif)
🎉 Celebrating 25 Years of GameDev.net! 🎉
Not many can claim 25 years on the Internet! Join us in celebrating this milestone. Learn more about our history, and thank you for being a part of our community!
RTS scheme, some ideas
First, a question, how many strategy games out
there use neural networks?
An application for Genetic Algorithms...
Second question, how do you defeat the enemy? May
sound stupid, after all, the answer is that you shoot him
until you win. So, what do you shoot? If we look at
history, we stumble upon Napoleon. If I remember right,
he won so much because he cut his opposition''s supply
lines. This can make a strategy game much more fun
because there are many types of resources which can
be destroyed. Perhaps different elements of strategy
can be used in a genetic algorithm.
What if we have a ''usual'' player who goes for the
swarm attack, focusing on numbers, not quality nor
strategy?
What if we have an enemy who puts his energy into
research for better weapons, defense, and destroying
the player''s research?
Who wins? At first, the ai, he abosrbs the player''s
attacks and eventually produces units capable of
devastating the player. The player learns and protects
his research and wins.
The ai would not win if it failed to research more
powerful weapons and so would that chromosome.
Looking back to my first question, if strategies learned
by the ai could be placed inside this genetic algorithm,
the ai would not be limited by the elements of strategy
that a developer can think of.
Second Act...
Now, this part you may not like, you might have to learn
something
.
Genetic algorithms are fun and all, but they require a
fitness function.
This requires we evaluate how effective a particualr
strategy or battle is working.
A good way I think would be statistics(this is the part
where you might have to learn something).
Another possibility would be neural nets.
What to measure? Well, there''s like the damage inflicted,
territory gained, survival rate of troops, etc. Whatever
criteria is used could actually be part of the strategy
itself.
This can be applied to different levels as well,
there''s the grand strategy-what side to attack from, what
resources of the enemy to destroy, etc.
there''s tactical strategy-unit combinations(do I send
infantry, tanks, or both?)
Of course, a lot of this is about deciding what to do and
deciding what information is important, for gathering
information I''d still go with influence maps.
Well, it''s about 4:33 a.m. where I''m at, so good night.
Create.
![](smile.gif)
I know another good thing to check with fitness: time spent in a single session.
Measure the time the player is playing, and when he quits ask why (either a scheduled appointment, or just because he had enough of it).
Sheduled appointments do not count.
When a player leaves because he played enough of the game, measure the time he played. The longer he played, the more he was absorbed by the gameplay... So, the AI was probably good enough to entertain te player![](wink.gif)
Just a general idea, I don''t think you can just implement it, but perhaps it''s something to think about......
Measure the time the player is playing, and when he quits ask why (either a scheduled appointment, or just because he had enough of it).
Sheduled appointments do not count.
When a player leaves because he played enough of the game, measure the time he played. The longer he played, the more he was absorbed by the gameplay... So, the AI was probably good enough to entertain te player
![](wink.gif)
Just a general idea, I don''t think you can just implement it, but perhaps it''s something to think about......
I know another good thing to check with fitness: time spent in a single session.
Measure the time the player is playing, and when he quits ask why (either a scheduled appointment, or just because he had enough of it).
Sheduled appointments do not count.
When a player leaves because he played enough of the game, measure the time he played. The longer he played, the more he was absorbed by the gameplay... So, the AI was probably good enough to entertain te player![](wink.gif)
Just a general idea, I don''t think you can just implement it, but perhaps it''s something to think about......
Measure the time the player is playing, and when he quits ask why (either a scheduled appointment, or just because he had enough of it).
Sheduled appointments do not count.
When a player leaves because he played enough of the game, measure the time he played. The longer he played, the more he was absorbed by the gameplay... So, the AI was probably good enough to entertain te player
![](wink.gif)
Just a general idea, I don''t think you can just implement it, but perhaps it''s something to think about......
quote: Original post by RolandofGilead
First, a question, how many strategy games out
there use neural networks?
Not many, really. I know of a couple of "informal shareware" kinds of games that used neural networks, but I can't honestly say any of them were terribly successful in the grand scheme of things. The Creatures series used a kind of NN/Genetic Algorithm approach to their virtual brains which was fairly successful overall, while both BC3K and Dirt Track Racing made somewhat less successful use of them. There are a few "classic" games that have used themover the last few years; it seems very popular for backgammon games (I'm not sure why) such as Jellyfish and
TD-Gammon.
Not too many games have used Genetic Algorithms....the most successful one IMO was clearly CDDNA. There have been others, of course, but most of those games suffer from GAs taking so long (normally) to converge on a meaniingful solution.
Ferretman
ferretman@gameai.com
www.gameai.com
From the High Mountains of Colorado
Edited by - Ferretman on October 29, 2001 11:11:36 PM
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement