[MUD-DEV] TECH : MUD game on a multi agent system

[John Robert Arras]

On Fri Feb 13, 2002, Cyril MudDev<cyril_muddev at yahoo.fr> wrote:

> I want to make a MUD based on a multi agent system (which complies
> with the FIPA specification). Does anyone know of a similar
> project? (a MUD based on the multi-agent technology) Does someone
> know wich problems I may encounter?

Hi,

There are some other projects (although not using the FIPA
specification that I know of) currently running such as John Laird's
SOAR Quakebot project:

  http://ai.eecs.umich.edu/people/laird/gamesresearch.html

and some information about research into bots using a modified
version of Unreal Tournamen. The project is called Gamebots:

  http://citeseer.nj.nec.com/477805.html
  http://www.umcs.maine.edu/~wagner/workshop/07_adobbati.pdf 
  http://www.planetunreal.com/gamebots/
  
I think ideas like these can be a useful addition to MUDs, but I
don't know how well they would scale right now.

Trying to make ALL mobs have extremely good AI is too expensive
right now (in terms of resources).

The bots in the examples above are designed to play within a small
world where there are only a few bots active at any one time.  I
don't see this scaling easily to massive numbers of bots in a much
more complex world.

Instead, MUDs need AI on several levels:

  Example levels:

    1. Complex hardcoded behaviors the creatures can follow but
    which don't require much processing time.

       - Moving from point A to point B for a reason.

       - Finding something to eat (if it's designed to feed itself).

       - Having a simple "job" like being a janitor or shopkeeper.

       - Accepting orders from a higher level authority (such as a
       town or a kingdom or chieftain ordering the armies to march.)

       - Reacting to local stimuli such as assisting friends and
       attacking enemies.

       - A fair level of tactical fighting ability.

       - Cheap, but not very effective at making a lifelike world.

       - As an example, I want to have dynamic kingdoms that raid
       each other and settle new kingdoms and build cities and guard
       posts and collect resources.

    These things could be hardcoded (but still interesting enough so
    that the mobs are differentiated) and made to work ok so that
    most creatures will not just sit there waiting to die.


    2. Scripting: There has to be some kind of internal scripting
    engine so that mobs can have simple reactions to other things,
    so that you get more interaction.

       - Quest mobs that answer questions and converse on simple
       topics.

       - Specific pathing or activities that only this mob carries
       out.

       - More expensive than hardcoding, but also more flexible,
       even if it is hard to make.

       - As an example, I want to have a special group of creatures
       that responds to player conversations with a single mob by
       attacking or helping in some way.

    Mobs with this kind of AI can do a bit more than the previous
    ones, but they still require a lot of work and can't learn much
    without bloating the game server.


    3. External AI that gets networked into the game. You connect a
    mob to a socket and send the data to the AI engine and the
    responses come back into the game (just like a player would do).

       - Mobs that learn to do things and have long-term memories of
       many things (since the memory is stored "offline" or at least
       away from the server.

       - Mobs that can plan and anticipate and carry on better
       conversations, since although you could hook up a Julia, it
       would be much nicer to have each mob have its own area of
       expertise. So part of its training could be to expand Julia
       in its own direction.

       - This is very expensive, but good if used in judiciously.

       - An example I want to implement some time is to have armors
       and weapons that can only be created using a several step
       process where several raw materials are used at each
       step...but make the processes needed nonobvious (even to
       me). Then, have an external AI just sit there and try to
       figure out how to make these items until it gets the entire
       tree of knowledge.
  
    These would be special mobs that have to do special things other
    than just being decorations.

The expensive external AI's could be shut on and off within
individual mobs if you just wanted to have special AI at specific
times when players are around, so you could shut the AI off when the
players leave.


John

_______________________________________________
MUD-Dev mailing list
MUD-Dev at kanga.nu
https://www.kanga.nu/lists/listinfo/mud-dev