[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Is this list still alive?

Jan Ekholm wrote:
> On Thu, 11 Jan 2001, Steve Baker wrote:
> >I want to write a game where the scenery builds itself using fractals
> >and such like - and the critters in the game are generated with genetic
> >algorithms. That way, I won't *need* no stinkin' artists.
> Now that's a lot of work. Do you want the critters to breed and create new
> types of critters too? :-)

Not necessarily *during* the game.

I had in mind something like the 'Biomorphs' setup where a human is
presented with a set of a dozen or so critters - you click on either
one or two of them and you get (respectively) a dozen random mutations
of the critter you clicked on - or a dozen different cross-bred children
of the two parents you picked.

Then you can simply click on the critters that look closest to what you
want your imagination sees as the most appropriate creature.

With the simple genetic system of the old biomorphs, you could arrive at
something close to what you wanted in a dozen mouse clicks - and refine
it with a dozen more.

In essence, the creatures are evolving with the goal of being most pleasing
to you.

Then I'd just take the creatures and use them in my game just as if an artist
had built it.

> Hypothetically it should be possible to have some genetic algorithm
> "evolve" monsters for some game. Give the engine a lot of free choice, a
> lot of time and let it breed new monsters. After a year or so harvest the
> best monsters out and plug them into some game. But GA:s don't work that
> way just yet, AFAIK. Nice idea though.

The problem with leaving the program to evolve creatures by itself is the
question of what constitutes the "best" creature(s) from each generation.

How you specify and *test* that criteria becomes a critical issue.

There was the classic case of where some guy wanted to evolve creatures
that could walk/swim/fly in their virtual world - so he set up a realistic
set of physics and set the goal for survival into the next generation as
being the fastest critter to move 100 meters.

He left them evolving for some large amount of time - then stopped to
see what he'd got.  The winning creature was a single 201 meter tall
inverted pyramid.  When the simulation started, it would simply fall
over.  Since his actual code tested the center of gravity of the creature
to see when it crossed the finish line - the 201 meter tall pyramid
could win in just seconds.

When he'd corrected that, he found creatures that had evolved to exploit
all kinds of other bugs in his criteria - bugs in the physics model - and
even creatures who got their energy supply from round-off errors in the
physics math.

Steve Baker   HomeEmail: <sjbaker1@airmail.net>
              WorkEmail: <sjbaker@link.com>
              HomePage : http://web2.airmail.net/sjbaker1
              Projects : http://plib.sourceforge.net

To unsubscribe, e-mail: linuxgames-unsubscribe@sunsite.dk
For additional commands, e-mail: linuxgames-help@sunsite.dk