Membership Details (University/College)
Contact: Pietro Baroni
Member node: 024
Via Branze 38
25123 Brescia
Italy
>>> Include yourself at this organisation
>>> Update your name or personal details listed below
>>> Submit papers, projects etc to the archive
>>> Update Member Description or Contact details
Description
The "Knowledge-Based Systems" group at University of Brescia is active in research on agents and multi-agent systems since 1993.
Two main lines of investigation are pursued:
- the definition and experimentation of architectures for intelligent agents and for multi-agent systems;
- the study of reasoning under uncertainty in multi-agent systems.
At the level of multi-agent systems, we have defined an architectural paradigm for multi-agent problem-solving, called specialist net, based on the metaphor of specialist and competence areas.
Within this paradigm, autonomous agents, called specialists, cooperate to reach the global problem solution. Specialist sharing the same type of knowledge are grouped in competence areas. Suitable data structures and facilitator entities are defined both at the agent and area level in order to allow inter-agent communication and cooperation.
This architecture has been exploited for the realization of a knowledge-based system for preventive diagnosis of power transformers [19,23] and of a simulated multi-agent control architecture for an autonomous robot [22].
Recently a multi-agent architecture devoted to support user-system co-evolution in human-computer interaction has been experimented [2, 3]
At the level of internal agent architecture the study our research has concerned the study and design of a an innovative model for agent's mental activity [17,12]. The proposal is founded on the basic claims that
(i) mental attitudes (beliefs, intentions, etc.) of an agent should be regarded as autonomous active entities and
(ii) an intelligent agent should be conceived as a distributed structure, where global behavior emerges from the interactions among active mental entities. A general agent architecture based on this approach has been formalized and a prototypical implementation has been realized.
This approach represents a significant innovation with respect to traditional architectures encompassing mental attitude representation in terms of mere data structures (e.g. BDI architectures). It offers several advantages both from an abstract point of view, concerning the general properties desired for the agent such as autonomy, intelligence and learning capability [14,16,18], and from the concrete point of view of agent systems design and implementation [15].
A prototypical implementation of the proposed architecture, called AGENTS+, has been developed in C++ on a Sun/Solaris platform.
Such architecture has been experimented in the realization of control architecture for two kinds of simulated robots: a high-level simulator of a mail delivery robot [20,21] and a realistic simulator of Khepera robot [11,13].
The issue of reasoning under uncertainty in multi-agent systems has been investigated considering both symbolic and quantitative approaches.
As to the symbolic approach, attention has been focused on argumentation theory. As all argumentation systems proposed in the literature are based on centralized algorithms, in order to enable the application of argumentation theory in multi-agent systems the problem of defining self-stabilizing distributed algorithms for argumentation has been successfully faced. Different graph-based representations of an argumentation system have been considered and the trade-offs between the expressiveness of the adopted representation and the computational complexity of the related algorithm have been evaluated [4, 6, 7, 9]. The connection between distributed argumentation and the active mental entities approach has been explored in [8]
As to the quantitative approach, it has been evidenced that there are currently no studies in the literature about a general format to support the interchange of information with uncertainty quantifications among heterogeneous agents locally adopting different quantitative formalisms.
In cooperation with University of Trieste, some of the theoretical problems related to the definition of a general uncertainty interchange format and of the relevant transformations between such general format and specific formalisms have been analyzed [1, 5, 10].
The above described research activities have been partially supported by CNR (Italian National Research Council) and MIUR (Italian Ministry of University and Research) in the frame of national research projects.
This list was generated on Mon Sep 10 16:38:38 BST 2007. |
