Authors:
Yuya Takata
;
Yuki Mikura
;
Hiroaki Ueda
and
Kenichi Takahashi
Affiliation:
Hiroshima City University, Japan
Keyword(s):
Multiagent systems, BDI architecture, Reinforcement learning.
Related
Ontology
Subjects/Areas/Topics:
Agents
;
Artificial Intelligence
;
Artificial Intelligence and Decision Support Systems
;
Distributed and Mobile Software Systems
;
Enterprise Information Systems
;
Knowledge Engineering and Ontology Development
;
Knowledge-Based Systems
;
Multi-Agent Systems
;
Software Engineering
;
Symbolic Systems
Abstract:
We propose a framework of cooperative learning of BDI agents. Our framework uses some kinds of agents, including a task management agent (TMA) and rational agents. TMA is designed as a learning agent. It manages assignment of tasks to rational agents. When a task is created, TMA evaluates the most useful strategy on the basis of reinforcement learning. Rational agents also evaluate the value that the task is assigned to them according to the strategy, and they give the value as their intention to TMA. Then, TMA optimally assigns the task to a rational agent by using both the value and the rough strategies, and the rational agent processes the task. In this article, we apply the proposed method to an elevator group control problem. Experiment results show that the proposed method finds better task assignment than the methods without cooperative learning.