An "overseer" agent for virtual crowds?

It seems to me that one of the strengths of virtual crowds is that,
through emergent behavior, rich simulations can result from simple
agent AI. However, I've had some troubles achieving specific crowd
behavior due to the emergent behavior. However, it's a pain to lay
out in detail everything that the crowd is supposed to do (and
arguably breaks the underlying concept anyway).

Are there any techniques for using an "overseer" agent to indirectly
control a crowd and thereby force the achievement of certain types of
crowd behavior? For example, an agent that will dynamically assign
motion goals to the crowd such that it will navigate intelligently
with respect to a dynamic environment. Are there any published papers
on such boss agents?

Thanks in advance for any info =).

[ comp.ai is moderated. To submit, just post and be patient, or if ]
[ that fails mail your article to <comp-ai.DeleteThis@moderators.isc.org>, and ]
[ ask your news administrator to fix the problems with your system. ]

On Wed, 04 Aug 2004 04:13:27 +0000, Jonathan Dinerstein wrote:
> It seems to me that one of the strengths of virtual crowds is that,
> through emergent behavior, rich simulations can result from simple
> agent AI. However, I've had some troubles achieving specific crowd
> behavior due to the emergent behavior. However, it's a pain to lay
> out in detail everything that the crowd is supposed to do (and
> arguably breaks the underlying concept anyway).

Wasn't there some software that modeled the behaviour of a flock of birds?
I think it ran on *nix? You might investigate that. I would guess that (at
least at superficial levels) flocks of birds, schools of fish, crowds, all
move with similar dynamics. If you're trying to model more complex
behaviour, like what makes a crowd "get ugly" and start to riot, then I
imagine that would be a lot more complex. Not in the behavioural sciences,
I don't know if anyone has done any simulations or models like that.

Personally, I'm sceptical of hoping for "interesting emergent behaviour"
to come out of parallelism or randomness of agents. Sounds to me too much
like "and then a miracle occurs". If your behavioural models are complex
enough, then maybe, but then you've already solved the problem? I wouldn't
expect any magic "free lunch" serendipity. You might be lucky though.

> Are there any techniques for using an "overseer" agent to indirectly
> control a crowd and thereby force the achievement of certain types of
> crowd behavior? For example, an agent that will dynamically assign
> motion goals to the crowd such that it will navigate intelligently
> with respect to a dynamic environment. Are there any published papers
> on such boss agents?

But then it would not be a crowd, would it? It would be an army? I suppose
there are instances in crowds where "instigators" do take control and
start directing the mob (what causes transition?). Is this what you mean?

> Thanks in advance for any info =).
>

--
Juhan Leemet
Logicognosis, Inc.

[ comp.ai is moderated. To submit, just post and be patient, or if ]
[ that fails mail your article to <comp-ai.TakeThisOut@moderators.isc.org>, and ]
[ ask your news administrator to fix the problems with your system. ]