Advertisement

Citicens

Started by October 30, 2003 01:41 AM
4 comments, last by emileej 21 years ago
I am planning a game so no code yet, but a whole bunch of notes. In this game I will need some fairly intelligent AIs to run around a town - going to work, shopping in the local mall, having dinner out or going to a party at friday etc. I basicly need a frame with some parameters to tweak or perhaps a neural net to train differently. However I am not shure if a neural net is a good idea since the AIs will have to run on schedule to some extend (get up, eat, go to work, eat breakfast, go home, eat dinner and go to bed), but still should be errorus and spontanious to some extend (getting late for work, suddenly deciding to take the family out to a restaurant tuesday afternoon etc.). What kind of techniques would you reccomend using? I have already mentioned neural nets, but I am thinking that it might be overkill and take too much time to teach it the simple tasks that I have just mentioned... "I feel like... I feel like... Like taking over the world!" - Day of the Tentacle
Emil Johansen- SMMOG AI designerhttp://smmog.com
Have a look at finite state machines and take it from there.

This not a job for an ANN.







My Website: ai-junkie.com | My Book: AI Techniques for Game Programming
Advertisement
quote: Original post by fup
Have a look at finite state machines and take it from there.

This not a job for an ANN.

Yeah... the wizardry here is done by determining all the transitions.



Dave Mark - President and Lead Designer
Intrinsic Algorithm - "Reducing the world to mathematical equations!"

Dave Mark - President and Lead Designer of Intrinsic Algorithm LLC
Professional consultant on game AI, mathematical modeling, simulation modeling
Co-founder and 10 year advisor of the GDC AI Summit
Author of the book, Behavioral Mathematics for Game AI
Blogs I write:
IA News - What's happening at IA | IA on AI - AI news and notes | Post-Play'em - Observations on AI of games I play

"Reducing the world to mathematical equations!"

If I want the citicens to respond to my freehand chatting - will I then need a merge of a fsm and a neural net?
Emil Johansen- SMMOG AI designerhttp://smmog.com
citizens.*

Kult House - Fresh Production Media

if i were you, i would forget every actual technique from the beginning, put down the problem, analyse the need of the problem, and then find direction to solve then

if you seek the response before you know what you need you will only end at cul-de-sac

as i see your matter i would try FuFSM(scheduling fsm) with some random value and light need and desire, the rest would be emergent

as i see the need value could grow any time but can be resolve with only some state enable (to prevent silly behaviour like sleeping in the road, but maybe it could sleep in work this would be fun if you add light personnality modifier)
it could turn very pretty if their all independent (sleeping in a job and the citicen could be fired and then turn into state 'seek job')

well i would use a pattern of state rather the one big only state for each citicen (like the social state, the psychologique state and the physiologic state) which could influance each other input and are responsible of overall behaviour ouput (ex the citicen get ill state then psychological state could turn to bad and since this state is responsible of wether the citicen go or not to work it could affect is social state when i would get fired)

with some variation to the sensibilty of entry you could have diferent personnality (if your citicen can breed and have children the weight of the state could be randomly mix to have a child which inherit from is parent)

one thing, it's just a tip, for my citicen if they could interact with object and place would have the code implement in the object or the place rather than the citicen, if the citicen has to interact with, i would check the object property and then summon the object code if it's fulfuill is need or goal, then the object 'tell' what the citicen what it should do, with this you have better control over the behaviour of the citicen (for ex what it could do or not in a meeting) and adding feature would be far fast and easy, it's a trick a i've learn by programming low end calculator and rpgmaker ("constrain is beauty" leonard de vinci 8) )

well you don't need NN but i have use this for my project which is quite similar but with less people and greater interaction (dinamic story) through emotion and social, then my agent learn about you and appraise your behaviour and they react emotionnly to event (they could evolve from they behaviour to event) but i have no seek for NN they come naturally and i even did not notice until recently that it was NN
then don't bother about technique just try to resolve things and keep you informed and documented, what you need will pop up through research or analysis

for freehand chatter seek doc from CHATTERBOT

any comments from someone else???

>>>>>>>>>>>>>>>
be good
be evil
but do it WELL
>>>>>>>>>>>>>>>


[edited by - Neoshaman on November 3, 2003 10:42:25 PM]
>>>>>>>>>>>>>>>be goodbe evilbut do it WELL>>>>>>>>>>>>>>>

This topic is closed to new replies.

Advertisement