[MUD-Dev] AI not worth doing in our games?

[Kwon Ekstrom]

From: <amanda at alfar.com>
> Jeff Freeman <jfreeman at soe.sony.com> wrote:

>> e.g. A "real" AI for pigeons in the park would be big, complex,
>> and result in flocks of pigeons wandering around the park.  We
>> can make pigeons do that without a big, complex AI, though.  So
>> if that's the desired result, we should just do that.

> I think this depends a bit.  If the pigeons really are just window
> dressing, sure.  And once players determine that they're just
> CPU-eating window dressing, they'll turn off "show pigeons in
> parks" display option (cf. client-side wildlife in AO, and
> landscape frills in AC2).

> However...

> If the pigeons respond to player actions, and mobs respond to
> pigeons, it could make it much more challenging to sneak through
> the park.  Example:

>   - target a mob, cast lightning bolt.

>   - flock of pigeons burts out the trees, scared by the noise.

>   - mobs converge on you to see what the fuss is about.

I believe he suggested instead of teaching a pigeon exactly how to
think and that it needs food and that it needs to peck to get food,
and that (x million other things) you teach a pigeon... walk around
in a semi-random order but stay near other pigeons... if you see
something that might be threatening, fly... if you see pigeons
around you fly, follow.

You could determine a threat such as a non-pigeon creature moving
within it's field of vision closer than x.

Also you could make some pigeons watch potential threats at a
distance (but not if more than x pigeons are watching them)

A simple state machine could handle it:

  State:

  1.  Wander Choose a direction and distance to walk before pecking
  a few times at the ground.

    The movement generator should:

      A. have constraints to keep the pigeon near fellow pigeons.

      B. Check for "threats"

  2. Watch Threat

    Wander as normal but occassionally look in the direction of a
    "threat" that was noticed.

      A. This should be limited to a certain number of pigeons to
      prevent the appearance that "the whole flock is watching me"

      B. Announce a threat if it gets too close and fly off.

  3. Follow pigeons who announce a threat.

You could probably just instantiate 1 AI to control the whole flock
and simply keep a history of direction and distance of each pigeon
to aid in generating realistic movement.

The only potentially tricky part is the movement of individual
birds, and you have a more or less non-cpu eating AI which should
behave somewhat believably.

> I'll go back to my claim that faked, teflon-coated scenery is less
> immersive than reactive scenery.  Remember what was so fun about
> Duke Nukem?  You could actually blow holes in stuff!  Sure, it was
> only a few preconstructed holes in a few pre-designated places,
> but it was there.  Same with decals in Q3 for scorch marks, etc.

Scenery is not AI, they are two completely separate concepts.
Besides the quality in which they are done contributing to the game,
I don't see the correlation.  Perhaps you mean using minor computer
controlled characters as scenery, but then you would either
A. restrict the number used to be too few to consititute scenery or
B. have a fairly simple AI to create relatively believable behavior
(see above).  Anything more could easily bog down resources for
somewhat limited and abstract gains.

-- Kwon J. Ekstrom


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