[Modeling] Re: Environment - Modeling Area document

sehl mellouli sehl.mellouli@ift.ulaval.ca
Mon, 24 Mar 2003 19:39:13 -0500

sorry for this late answer.
I think it is very important to deal with MAS environment and its impact 
on agents interactions, specially in open systems. In this sense, I'm 
doing some works on modeling MAS environment. I invite you to read two 
of my papers:


Mellouli Sehl, Guy W. Mineau and Bernard Moulin. Situation Event Logic 
for Early Validation of Multi-Agent Systems. Sixteenth Canadian 
Conference on Artificial Intelligence, Halifax, Nova Scotias, Canada, 
June 11-13, 2003. (to appear).

In these two papers I present how to model MAS environment, based on the 
situation calculus, and its impact on the MAS organization (agents 
relations). This work is in progress and many issues are still open to 
be addressed.

Best regards,

Dr. Hong Zhu wrote:
> Hi, Marc-Phillippe,
> Your are quite right to say that when we talk about environment, we meant
> the structure (e.g. the shape and population) and dynamics of the
> environment. The question is how the structure and dynamics of an
> environment can be specified. I believe that the same mechanism and notation
> for the description of a system can be used for the description of
> environments, because an environment plus the system under consideration is
> also a system. Of course, for the elements in the environment, we would not
> bother (and/or unable to) to give their full details.
> For the description of the environments of MAS, I think we need to address
> the following issues:
> (1) Openness: elements can join and leave the environment at runtime;
> (2) Mobileability: an agent can move into a system as a part of other
> agents' environment;
> Therefore, instead of describing the specific agents in the environment, we
> would have to use some abstract notion to describe the possible 'types' of
> elements in the environment and how they would interact with the agent under
> consideration. That is why castes are used in my approach. In fact, castes
> can implement abstract concepts like what we call roles.
> Best regards,
> Hong
> ----- Original Message -----
> From: "Marc-Philippe Huget" <M.P.Huget@csc.liv.ac.uk>
> To: "Dr. Hong Zhu" <hong.zhu1@btinternet.com>
> Cc: "James Odell" <email@jamesodell.com>; "Renato Levy" <rlevy@i-a-i.com>;
> "Radovan Cervenka" <rce@whitestein.com>; "Paola Turci" <turci@ce.unipr.it>
> Sent: Tuesday, March 18, 2003 10:29 AM
> Subject: Re: Environment - Modeling Area document
> "Dr. Hong Zhu" wrote:
>>Hello, everybody,
>>Description of environment is an important part of my agent-oriented
> specification language SLABS. I believe that the framework can be 'reused'
> here. Essentially, it consists of two parts: static descriptions of
> environment configurations and dynamic descriptions of run-time situations
> in the environment.
>>For the static description, in the descripton of each agent/caste (which m
> ight be the same as Jim's 'Agent Class' and I will back to this issue later
> with a discussion on the meta-model.), a set of castes plus a set of
> specific agents can be given to indicate a subset of entities that the agent
> under definition is interact with. That is, all memebers of the caste plus
> the set of specific agents constitute the environment of the agent (or the
> agents of the caste). A multi-agent system can be considered as an agent,
> which is decomposed into a set of agents classified into castes. Variables
> range over a caste can also be given as a parameter, which is initialised
> when the agent is created and the agent can change the values of the
> parameter at runtime. In addition to such changes to and agent's
> environment, an agent can change its memebrship to a caste by joining a
> caste and/or treating from a caste so that move from one environment to
> another. Moreover, other agents can join the caste that is a part of the
> agent's environment. In this way the agent's environment become open to
> other agents. By the way, in addition to the existence of an active body, I
> believe that the dynamic membership of an agent to a caste and the explicit
> description of environment are the main feature that distinguishes caste
> from class. Well, this is a meta-model issue.
>>A dynamic description gives a typical situation in the operation of the
> system from an agent's point of view (i.e. the agent's view of the state of
> the system). It is described as a scenario, which is a logic combination of
> the patterns (sequence of activities and state changes) of the entities in
> the environment. An example scenario that can be described by SLABS is that
> all agents X of a caste C are in a particular state and there is an agent Y
> in caste B that is in another state. For example, the following is a
> scenario: 'there is an memeber of conference steering committee nominated
> agent A as conference programme committee chairman and all members of the
> programme committee voted to support agent A as chairman', where steering
> committee and committee are two castes. Dynamic description is a crucial
> part for the description of agent's behaviour. For example, in online
> auction, we would have a rule like this: 'when agent A in a caste Buyer made
> a bid and no other agents make any bid afterward with two days (or whatever
> period of time), then the agent win the bid and that agent must make
> payment'. This rule consists of a scenario (of the environment) and an
> action in that scenario.
>>What do you think of the framework? If OK, I will adapt the framework to
> I don't understand where environment is involved here. When I speak about
> environment, I speak about the structure of the environment, its shape, the
> population, the dynamics of the environment, etc. When reading your
> explanations, I find something more related to role dynamics, am I wrong?
> Cheers,
> Marc-Philippe
> --
> Marc-Philippe Huget
> Agent Applications, Research and Technology Group
> Department of Computer Science
> University of Liverpool
> Chadwick Building, Peach Street
> L69 7ZF Liverpool
> United Kingdom
> email: mph@csc.liv.ac.uk
> http://www.csc.liv.ac.uk/~mph
> _______________________________________________
> Modeling mailing list
> Modeling@www.fipa.org
> http://fipa.org/mailman/listinfo/modeling

Mellouli Sehl
Computer Science Engineer
Information Systems Administration MBA
Computer Science Phd Student
Université Laval, Québec, P.Q, Canada
Tél: bur (418) 656-2131 (4704)
Home page: http://www.ift.ulaval.ca/~mellouli