Distributed Artificial Intelligence
Research Program - Results (2)
Architecture for multi-agent systems
In our group Shen proposed a two level architecture.
At the first level agents are grouped locally and belong to the same LAN.
The LAN structure allows an agent to see all messages. Thus the basic communication
structure is of broadcast type. Each agent can thus look at messages, even
if it is not the recipient of the message. It can look at any message and
use the information to update its internal representation of the work in
progress, or of the capabilities of the other agents. At a more global
level, Shen proposed an intergroup communication,
when some needed service is not available locally. In such a case, it is
possible to use Internet. However, it is not necessary to channel all communications
through a specialized agent like a facilitator or a manager,
which could cause a bottleneck. On the contrary, local agents are allowed
to set up communication channels directly with other remote agents. This
architecture was implemented in the DIDE project
in the domain of cooperative design.
Other propositions for agent architecture have been done. Interested
readers are referred to the book :
Shen W., Norrie D. H., Barthès J-P. A.
2001. Multi-Agent Systems for Concurrent Intelligent Design and
Manufacturing. Taylor and Francis.
In addition a French speaking group works on comparing various architectures
in the framework of the AFIA Association (WG ASA).
Last modified 8/4/02