Information Technology Reference
In-Depth Information
MAS is a branch of DAI research. In a multi-agent system, an agent is an
autonomous entity which continuously interacts with the environment and
co-exists with other peer agents in the same environment. In other words, agent is
an entity whose mental states consist of components such as belief, desire and
intention. In a multi-agent system, the agents can be either homogeneous or
heterogeneous, and the relationships among them can be either cooperative or
competitive. A common characteristic of DAI and MAS is distributed behaviors
of entities or agents. Multi-agent systems feature bottom-up design, because
impractical, the distributed automatic individual agents are defined first, and then
problem solving is accomplished with one or more agents. Both single objective
and multiple objectives can be achieved. Research on MAS is dedicated to
analysis and design of large-scale complex cooperative intelligent systems such
as large-scale knowledge and information systems and intelligent robots, based
on theories of problem solving through concurrent computing and mutual
cooperation among logically or physically distributed multiple agents.
At present, MAS is a very active research direction, which aims at simulation
of human rational behaviors for applications in domains such as real world and
society simulation, robotics, intelligent machines, etc. An agent is characterized
with features of autonomy, interaction with the environment, cooperation,
communication, longevity, adaptability, real-time, etc. In order to survive and
work in the constantly changing environment of the real world, agents should not
only react to emergencies promptly, but also make middle or short term plans
based on certain tactics, and then predict the future state through modeling and
analysis of the world and other agents, as well as and cooperate or negotiate with
other agents using the communication language.
To achieve these features, agent architecture should be studied, because
architectures and functions of agents are closely related to each other: improper
architecture may greatly limit the functions, while appropriate architecture may
well support high level intelligence of agents. We proposed a compound
architecture for an agent, which systematically integrates multiple parallel and
comparatively independent yet interactional forms of mind, including reaction,
planning, modeling, communication, decision making, etc. A Multi-Agent
Environment (MAGE) is implemented through the agent kernel based plug-in
approach we proposed for agent construction (Shi, 2003). With MAGE and the
plug-in approach, compound agents can be conveniently constructed and
debugged.
