DRAFT
DISCUSSION DOCUMENT
TOPIC:
As an extension of the distributed AI approach being followed in developing
the swan management DSS, should a multiagent system be employed?
A draft outline for applying such a systems based on a bdi architecture
to the swan management DSS is presented below. This concept, and
this page, is continually evolving and we would be most interested in your
ideas and critique. --Rick
Sojda
Brief Background Notes
From:
Wooldridge, M. 1999. Intelligent agents.
Pages 27-77 in Weiss, G. ed. Multiagent systems. MIT Press. Cambridge,
Massachusetts.
Belief-desire-intention (BDI) architectures
are based on practical reasoning and its two processes:
-
Deliberation: deciding what goals to achieve
-
Means-ends reasoning: deciding how to achieve
those goals
Definitions related to BDI architectures:
-
beliefs (information): information the agent
has about its current environment
-
belief revision function: generates a new
set of beliefs, given some sensor input
-
options (desires): desirable, current courses
of action available to the agent
-
option generation function: determines the
options available to (intentions of) the agent, given the current beliefs
and current intentions
-
intentions (goals): those states of affairs
that the agent is currently committed to trying to realize
-
filter: generates a new set of intentions,
given current beliefs, options, and intentions
-
action selection function: determines an action
to perform, given current intentions
Brief Background Notes
From:
Rao, A. S., and M. P. Georgeff.
1995. BDI agents: from theory to practice. Technical
Note 56. Australian Artificial Intelligence Institute. Melbourne, Australia.
14 pages.
Definitions related to BDI architectures:
beliefs: information about the system state
which is updated after each input of new sensory information
desires: objectives to be accomplished, generated
instantaneously or functionally
intentions: currently chosen course of action
(output) from most recent call to selection function
Definitions
from a Georgeff slide show, "Uncertainty, Change, and Limited Resources"
Potential Overall Decision Support
System Outputs:
-
Simulated, suggested distribution of wintering
swans next year
-
Simulated, suggested distribution of breeding
swans next year
-
Delineation of weakest information link(s)
in developing simulated distributions
-
Determination of whether individual areas
are meeting RMP goals according to flyway principles
-
Recommendations for water levels for individual
wetlands
As my ideas evolve, basic designs of a bdi-based, multiagent system
for the swan DSS are fouind here:
Return to the Swan Management
Decision Support System homepage.
USDI
- Geological Survey
Biological
Resources Division
Northern
Rocky Mountain Science Center
Maintainer: Rick Sojda
(sojda@swan.msu.montana.edu)