Re: DungeonMaker 1.0 released

[Henningsen <alg@glinx.com>]

>I guess you should provide both the option for the game to make explicit
>settings of all the parameters - and the option for the game to say "more
>like the last one" or "less like the last one"...libraries should always
>give the application direct control as well as higher level abstractions
>wherever possible.

Let me restate more clearly what my options are with the "evolution of
dungeons". I am considering writing an "evolutionary engine", which would
run a genetic algorithm over a set of parameters. In the case of
DungeonMaker, the parameters would be mostly those in the stats-file which
determine local dungeon "feel". But there is no way to do this without
having (as input to the evolutionary engine) a fitness value that measures
how desirable this set of parameters was. Now, if the user enters that
value, and he rates dungeons higher that make it easier to "score high/win
fast", he will almost always end up with an extreme and possibly degenerate
dungeon which is either very open, or a very intricate labyrinth with narrow
tunnels, depending on where it is easier to win. I don't think any user of
the library would want this to happen. I agree with Mark that it is unlikely
that users will use this power to evolve beautiful/fun dungeons;-(

The other option would be to have the program that uses the DM library
automatically rate dungeons, maybe as part of dynamic difficulty balancing,
but that can be done much easier. A genetic algorithm is a fairly solid
chunk of code.

The thing is, I will need an evolutionary engine for my next project anyway,
and writing it for the DungeonMaker first would help to ensure that it is
independent of the program that uses it. Is there interest out there for a
general implementation of the genetic algorithm, or is there a free
implementation available already?

>The *far* more important thing to make this useful is a way to control
>the interior *look* of the rooms...for example, it would be cute if I
>could provide a 3D model of a sample section of corridor, and a range of
>rooms of differing sizes, 3D models of doorways and staircases.  Then
>Dungeonmaker could build a fully fledged 3D model of the maze using scaled
>and stretched versions of the basic 'toolkit' of models that I provided.

I'm actually considering doing that, for version 3.0, but in a different
manner: I have been visualizing my dungeons as open to the top, and sunk
into a landscape defined by a heightfield. This could be used (together with
impassable mountains bordering two sides of the dungeon) to enforce that
people wanting to get from A to B in the landscape have to pass through one
of these dungeons. However, that would be mostly graphics programming, and
I'm really interested in alife/AI programming, so I think it'll depend on
someone coming on board who wants to do that. Once version 2.0 is out the
door, that person could become lead developer on the DungeonMaker project.
Guess I'll post a request when the time comes.

>I believe that's what CrystalSpace has - one of the programs takes
>descriptions of walls etc and creates CS world from it.

I think it would be very nice to put the Dungeonmaker and that part of CS
together... anyone know what the status of the Crystal Space dungeonmaking
module is? Would the CS developers be interested in talking about this?

Peter Henningsen



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