Archived

This topic is now archived and is closed to further replies.

Neoshaman

artificial emotion

Recommended Posts

Neoshaman    180
quote:
http://www.igda.org/ai/ai_committee.php Nowadays, game AI developers rarely have a chance to work on higher-level AI, like believable and interesting NPCs that can learn, have emotions and complex reasoning and interaction skills, or an automated story-telling system that adapts to a player''s interests. Much more likely, an AI developer will spend her/his time struggling with low-level pathfinding details.
what do you think about this?? i am working on an AE system for improving my game, i want to know if there is people which work on the same matter, and i''m interresting in sharing thought on the matter my system is finish yet i have some issue to find useable material on emotions and moods (think a social game as a soap opera as a model for a general emotion focuse experimental game) my system as a preference appraisal system which appraise object and events from their attributes and output an weighted emotion emotion are send to a FMS mood system which switch between mood according to emotion i have avoid need and desire because (in this case), the game make an evaluation of the world and attribute role and fonction to npc the whole problem hear is to find which emotion and which mood and how they interact each other (for a social narratives orianted game not like a simulation orianted game like the sims) >>>>>>>>>>>>>>> be good be evil but do it WELL >>>>>>>>>>>>>>>

Share this post


Link to post
Share on other sites
Martel    122
That''s really interesting. Are there any games out now in which characters have emotional responses to every object in the game?

Share this post


Link to post
Share on other sites
MikeD    158
To blow the Lionhead trumpet again, you might find this interesting

http://www.gamasutra.com/features/20020424/evans_01.htm

Mike

Share this post


Link to post
Share on other sites
Mull    134
I seem to remember [Dr] Rob Saunders mentioning this is a particular area he has worked/does work in when we were out on the lash at GDCE. I also seem to remember that he mentioned he doesn''t lurk around here that often, but I''m guessing you''ll be able to track him down at Alex''s site over at http://www.ai-depot.com

Share this post


Link to post
Share on other sites
Martel    122
Great article. From the article, their heirarchical state machine paradigm seems very powerful. I don''t agree with him that the format they chose doesn''t require a programmer. I can imagine those getting complicated quickly. Just because a designer can cope with programming doesn''t mean it''s a good idea to make them do it. The system could be more powerful if implemented in C++, but my first inclination would be to use something like Python or Lua (assuming it would never have to be ported to a console).

Very interesting.

Share this post


Link to post
Share on other sites
Martel    122
Also, I just stumbled onto this article (thanks for the hint):

http://ai-depot.com/Articles/50/Emotional.html

The thing that really intrigued be about the Sims' desire-driven model and Neoshaman's suggestion is that if you look at it the right way, the objects in the world create a sort of "emotional topography", and behavioral simulation becomes more like a heuristic pathfinding (i.e. search) operation. At least, that was my gut reaction - I have nothing to support that feeling

[edited by - Martel on September 4, 2003 11:59:27 AM]

Share this post


Link to post
Share on other sites
Neoshaman    180
hello

i''ve post more information on the gamedesign board before, because i start there, i was thinking it was more an deign issue than AI but it''s seems that i was wrong

i''va read both article:
gamasutra> wow, it'' a kind like what i have design the whole engine
AI depot> i''m not agree that emotion is this difficult to model, we have just to start with rough design before improve, by pratice and common sense we can arrive to a beleivable solution

well the actual engine work by appraisal a scene through prefence as i say, there is two kind of things which is appraise, events and evironnement

events are action and change in the scene
environnement are attribute in the scene (object and agent attribute each agent is define by a vector of attribute)

this goes in the preference filter (which is a kind of perception) which contain standard, like and dislike, needs, personnality etc..., everything which belongs to the agents and this system output a primitives weighted emotion which can switch the weighted mood state, mood state affect the weight of preference.

mood affect behaviour set according to role and other required influance (such as social rule, needs of the scenario, etc...)
then behaviour and action affect the scene and the process cycle again

the problem is to indentify primitives emotion and mood, and what kind of interaction they have each over

i have also post on gamedesign IGDA forum and i will certainly start wandering around AI forum, because it''s a great step to take!

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

Share this post


Link to post
Share on other sites
Neoshaman    180
i have done some more advenced reshearch

it's seems that where there is emotion in game they skip emotion for a more behavorial stimuli respond (maybe in sooner game we will find this)

but is it correct for making more believable game??

edit<<<<<
well it's seems that, thanks to ai depot, that i'm not exactly using a FSM but a kind FuWSM (fuzzy weighted state machine)
is something like FuWSM exist??

and where i could find a classification of emotion and mood??

la suite au prochain episode??
/edit<<<<<
>>>>>>>>>>>>>>>
be good
be evil
but do it WELL
>>>>>>>>>>>>>>>

[edited by - Neoshaman on September 5, 2003 4:17:45 AM]

Share this post


Link to post
Share on other sites
Martel    122
The last game I worked on originally started with fuzzy state machines. It was just like a normal state machine, except the value of some special variables could be used to influence which state to transition to.

The fuzzy system was taken out completely because, for the game we were working on, it didn''t make sense to have *less* control over what the AI was doing. In other words, fuzzy was just too inconsistent for that game.

I''d bet that plenty of games have used fuzzy weighted state machines before.

As far as "classification of emotion and mood": I don''t quite understand what you mean.

Share this post


Link to post
Share on other sites
Neoshaman    180
well actually i have built a table of emotion but i don't know how to manage them
the output of preference is > please/displease (more like comfort/discomfort?)
ouput for standard is > praise/blame (respect/anger)
relation with player is >hate to love
the intemency come to > stranger to familiar
mood can be set at>calm, sadness-despress, joy,anger-rage, nervosity-anxiety

well, the appraisal system work by filter action and events by the preference which ouput weighted emotion (blue>like a lot>please++,spider>fear a lot>displease- - - -, something like that, but please/displease is not enough to get all emotion)


the true problem is (and that's why i type it's a design issue rather than something else) :
the language we use for emotion isn't accurate and many words cover many reality
you can esteem someone you hate(respect) and misesteem someone you love (pity??? )
then praise and blame shouldn't lead to like and dislike??
what's make someone like or hate you, more and more??
how and why we switch to a mood to the others (pratically)??
how does emotion affect mood concretly??
i'm not sure if emotion and mood state i have choose is the good choice, it seems to me that the model lack something

mood is problematic too because we act according to a general mood and have specific mood towards other's which are dependant but not the same (i forgot to put a mood variable in the relation set, since you can be angry towards someone you like)

actually the personnality is suggesting in designing each preference and standard of a character

mood work as en emotionnal memory (persistance of emotion)which affect long term action, while emotion affect short term action

it's important because in actual game there is no long term relation in ae's npc they are more like stimuli response, the behavorist approche failed to model deeper feelings, because there is no memory (the relation cannot evolve according to action of player AND by choice of the npc, they are mostly script when they are well done)
they failed also in a true autonome agent in a story and narratives simulation

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

[edited by - Neoshaman on September 5, 2003 12:39:41 AM]

Share this post


Link to post
Share on other sites
Predictor    198
quote:
Original post by Martel
The last game I worked on originally started with fuzzy state machines. It was just like a normal state machine, except the value of some special variables could be used to influence which state to transition to.

The fuzzy system was taken out completely because, for the game we were working on, it didn't make sense to have *less* control over what the AI was doing. In other words, fuzzy was just too inconsistent for that game.



Why would fuzzy finite state machines provide "less" control? Fuzzy systems are supersets of their non-fuzzy counterparts, so one would expect more control. Granted, if fuzziness were not adding anything useful, one could simply drop it, but I'm not clear on why you think one would have "less control".




[edited by - predictor on September 8, 2003 6:14:48 AM]

Share this post


Link to post
Share on other sites