An implementation of NPC behavior
Steven McQuinn
kora...@nevada.edu (DAVID KORABELL) writes:
> Here's how I'm trying to implement npcs in my system. First I'm focusing
> on the social interactions in the game, so npc behavior is evaluated
> for social ramifications. I've generated a list of all possible actions for
> the player or npc ( I try to treat them as equivalent), then I generate
> several tables of correlation probabilities showing how likely it is for
> each behavior to occur as a response for each behavior, how much each
> behavior correlates with a given emtional state (which is a constantly
> varying attribute of an npc), and how much each behavior correlates with
> each personality trait ( which are fixed percentages for each npc). The
> npc's normally respond only to the actions of others, but special scripts
> can be added. Since a given npc's response is dtermined by his traits
> (which are probabilites) and his dynamic emotion state, the responses are
> not fixed, but reasonable logically consistent with the situation and the
> given npc's personality. I'm still working on a npc goals module.
Your design reminds me of eigenvector methods for statistical
correlation (AKA common factor analysis, empirical orthogonal
functions)--not that I know much about the topic in detail. Is such, in
fact, the template for your approach? You might find in the psychology
literature some useful research findings regarding the statistical
correlation of personality traits.
If your system chooses a specific NPC response from tables of
inner-emotion/outer-behavior correlation probabilities, would you
employ random chance to pick the final outcome from an array of the
most likely candidate behaviors?
I'm attacking the NPC behavior issue from another direction--cybernetic
control theory. My hypothesis is that characters can be structured to
seek autonomous "self interest" (broadly considered) and in so doing,
would interact with each other as well as with the player. Some very
strange societies might result!
I wish you good fortune with your innovation.
DAVID KORABELL
responses using random chance. Furthermore, it focuses on NPC - NPC
interactions. An NPC views the Player or another NPC as equivalent. NPC's
can be given focuses that the player may manipulate. An example:
NPC called morris - very timid, but goes bereserk when he sees something
purple ( you figure out why - and don't get too freudian); there's a gorilla
that blocks the players path, but likes to wear pretty things. We define
pretty as equivalent to how valuable something is. The likelyhood of the
gorilla wearing a given item is proportionate to its value. So the player
must try to find a valuable, purple, wearable item; give it to the gorilla;
introduce morris to the gorilla and hope he'll eliminate the gorilla.
The drawback is morris might figure out he's been used. The next time
the player is fighting against a creature, he may discover that morris
has used the opportunity to stab him in the back (either figuratively or
literally).
I've further added rules by which NPC's can delegate tasks to allied
NPC's & form alliances with other NPC's. So the player might be building
an army to locate & capture a valuable artifact while several NPC's are
building their on forces to achieve the same goal. Furthermore, they will
be simultaneously trying to steal members and information from other groups.
We thus have an emergent social system, but rather than being modelled
on the system level, it focuses on the processes of the individual. Is
this somewhat like what we might call interactive drama?
Steven McQuinn
> Yes, The system I've envisioned selects from a list of most likely
> responses using random chance. Furthermore, it focuses on NPC - NPC
> interactions. An NPC views the Player or another NPC as equivalent.
> (deletion)
> I've further added rules by which NPC's can delegate tasks to allied
> NPC's & form alliances with other NPC's. So the player might be building
> an army to locate & capture a valuable artifact while several NPC's are
> building their own forces to achieve the same goal. Furthermore, they
> will be simultaneously trying to steal members and information from
> other groups.
> (deletion)
> We thus have an emergent social system, but rather than being modelled
> on the system level, it focuses on the processes of the individual. Is
> this somewhat like what we might call interactive drama?
An interactive _improvisational_ drama, I would say from your
description. It resembles the way kids play out a story of their own
devising. I think that approach is good because it taps into an
ingrained cultural skill.
I've seen qualms expressed in other posts about randomized responses,
but your concept seems in the clear. You are providing for apparent
consistency between stimulus and response, and the player's imagination
should help you out in this regard.
How will your prepared text responses be able to anticipate the
combinatorial explosion of interactive possibilities? (Factorials Are
Frightening Things.)
How much will you utilize stylish writing to sustain the mood of the
game?