A Mutual Delegation Mechanism for Plan Executions in Multi-Agent Systems
Kei Matsubayashi ($B>>NS(B $B7D(B)
Research in Multi-Agent Systems (MASs) is very important for open
In MASs, autonomous processing entities (agents) who are treated
equally have their own goals, and try to achieve them through
communication and collaboration with other agents.
An agent's action taken when it collaborates is called strategy.
Strategies are decided by the agent itself, not by the designers of
Many have researched the mechanisms used by agents when making such
In order to make the decision, agents must be able to predict the
strategies of other agents because their strategies are effected by
the other agents' strategies.
In order to predict the strategies of other agents, agents have been
assumed to be Rational Agents that maximize their own benefit (utility).
Rationality helps agents decide their strategies for
However, rationality sometimes interferes an agent from collaborating.
This kind of situation occurs when there are redundant actions between
plans generated independently by agents.
In such a case, agents can execute plans which are more efficient
if they mutually delegate the executions and optimize each plan.
Such collaboration is useful for an agent to act efficiently.
Nevertheless, even if rational agents are able to delegate the
executions, it is not possible for them to receive them because of
their rationality, so that they cannot collaborate.
In this paper, we distinguish ``Short-Term Collaboration Strategies
(STCSs)'' which are rational strategies in the current collaboration
from ``Long-Term Collaboration Strategies (LTCSs)'' which are rational
strategies which includes the current and all future strategies.
In previous work, agents could not collaborate because they only used
In this paper, we will propose a mutual delegation mechanism using
LTCS for plan executions of rational agents.
Furthermore, the collaboration strategy may not be rational if it is
used in a situation where both strategies exist.
We will show our mechanism considers such situations to guarantee
rationality of agents.
Abstract in Japanese:
How to get this paper?
To obtain this paper, please contact Kei Matsubayashi(firstname.lastname@example.org).
Thank you very much for your attention.