Hi,
in the real world, we often know quite a lot about the (mathematical
and physcial) relations between various physical properties of the
world, but we do not know their (exact) values and have no way to find
out about them. I think something similar should hold for disaster
simulations: there are things we can observe (e.g. how strong is a fire
right now, what's the wind direction, what material is a house made of)
and some things we can not, but know about their importance (like the
amount of burnable material in the house, statical information about the
building, etc.). So, if the disaster simulator has such information
(coming from a config file unknown to the agents (like polydata) or
resulting from some randomized computation) agents will have to
interpolate from their observations to predetermine the course of the
disaster. So, in my personal opinion, agents can know about the physics
but not about unobservable parameters.
I think that unpredictability is a really important aspect of multiagent
domains. If we did not have some (at least small) degree of
unpredictability, coordination could be reduced to locker room
agreements or even hard-coding, because at least in theory events in the
environments would become deterministic. But, hey, that's just another
opinion.
michael
ResQ Freiburg
P.S. Just out of curiosity, why did you ask?
Mohsen Izadi wrote:
> Dear Friends,
>
> Today, we had a discussion about developing simulators
> for RoboCup Rescue Simulation Competitions, we faced
> the following question :
>
> "Is a disaster simulator being considered as
> black-box for agents and agent developers? In other
> words, are the agents allowed to have the direct
> knowledge about how a simulator works? Should a
> simulator developer document the way in which his/her
> simulator simulates a disaster? And if so are agent
> developers able to use this data directly? For example
> as an agent developer, are we able to know exactly
> what the effect of an extinguish action is on a
> burning building?"
>
> In our team, We had different opinions about this
> situation:
>
> - Some of us think that as unpredictability is the
> nature of disasters, agents are not allowed to have
> any information about how a simulator works and an
> agent or agent developer should infer and conclude
> rules by observation and learning.
>
> - Some of us think the different way; The main goal of
> RoboCup Rescue Simulation is not in dealing with
> black-box simulators. The main goal is improving
> inter-agent cooperation and coordination in a dynamic
> environment. So each disaster simulator should have a
> predefined behaviour that can be programmed directly
> by agents and agent developers can study the source
> code of a disaster simulator to see how it behaves and
> use the results in their agents.
>
> - The third group thinks that the behaviour of a
> disaster simulator should be documented and free for
> use by agent developers, but only to some extents.
> Certain kind of information is allowed to be
> documented and certain kind of information about the
> internals of the diaster simulator should be private.
> For example in a model the parameters being considered
> by the simulator to simulate a behaviour should be
> documented and public for use but the information
> about how these parameters are related to each other
> and the behaviour must be kept private and cannot be
> used by agent developers.
>
>
> At last we decided to ask this question here in order
> to know the opinions of other agent/simulator
> developers and also the way this situation is handled
> by the technical committee.
>
>
> Regards,
> Mohsen Izadi,
> S.O.S. Member
>
>
>
>
> __________________________________
> Do you Yahoo!?
> Yahoo! Photos: High-quality 4x6 digital prints for 25¢
> http://photos.yahoo.com/ph/print_splash
>
-- Michael Brenner Phone: (+49) (761) 203-8226 Institute of Computer Science Fax: (+49) (761) 203-8222 Albert-Ludwigs-University mailto:brenner@informatik.uni-freiburg.de Am Flughafen 17, Geb. 52 http://www.informatik.uni-freiburg.de/~brenner D-79110 Freiburg, Germany
This archive was generated by hypermail 2.1.3 : Wed 21 Apr 2004 - 13:59:37 GMT