Multiagent Planning with Incomplete Information in an Environment with Dense Interactions: A Case Study

概要

論文の詳細を見る
This paper provides a framework for multiagent planning with incomplete knowledge of an environment with dense interactions. In this environment, the multiagents interact with each other directly, and the action of an agent alters the environment. In addition, the conflicts caused by the simultaneous execution of actions must be eliminated. Moreover, an agent cannot see the entire world, but only has restricted information on the local environment, and the communication overhead required to collect the necessary information is very high. When the multiagents move in a restricted congested space, their actions frequently affect each other directly or indirectly. Whether an agent can move depends strongly on the adjacent agent, and the interactions are dense. Our framework introduces a message-handler to eliminate conflicts arising from the simultaneous actions of agents, and adopts a static strategy and a dynamic strategy for selecting nondeterministic actions so that an agent can act in real-time and plan efficiently. The framework is considered to be reactive planning, which has the following advantages: (1) the communication overhead is low, (2) the computation cost does not depend on the number of agents, and (3) goals are achieved with a high probability As a case study, we consider the blocks world problem, and discuss the effectiveness of our framework. The experimental result shows a real-time response and a high rate of goal achievement are realized.
関西学院大学の論文
2004-03-31

Multiagent Planning with Incomplete Information in an Environment with Dense Interactions: A Case Study

スポンサーリンク

概要

著者

関連論文

スポンサーリンク