An Evolutionary Approach to Balancing Deliberation and Reactiveness in a Multi-Agent Scenario

• Item type: National Convention(1)
• 公開日: 1996-03-06
• タイトル / 言語: en
• タイトル: An Evolutionary Approach to Balancing Deliberation and Reactiveness in a Multi-Agent Scenario
• 言語: eng
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_5794
• 資源タイプ: conference paper
• 著者所属(英): en
• 著者所属(英): Department of Administration Engineering Faculty of Science and Technology Keio University
• 著者所属(英): en
• 著者所属(英): Sony Computer Science Laboratory Inc.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Regarding planning behavior in an ecosystem a tradeoff between cost and time exists. If an agent has to plan he will tend to minimize the costs of the plan execution which will lead him to longer deliberation in order to find a suitable low cost plan. On the other hand the agent is situated in a competitive environment which gives him the highest guarantee of plan execution if his planning behavior is strongly reactive. Thus, a dilemma-like situation emerges which is due to the bounded rationality [1] which characterizes living beings as well as intelligent agents in complex environments : optimizing its outcome with its limited abilities. Evolution can be a source for diversity and complexity, but also for stability in a dynamical system such as system of multiple, goal-driven agents which have to do planning in an ever changing dynamic environment in order to fulfill their goals. We assume such agents to be selfishly motivated in the first place and to compete for resources. Executing a plan is connected with costs, e. g. energy consumed when moving a robot. Fulfilling a goal gives a payoff to an agent, e. g. reaching a loading-station for a robot. Regarding agents as dissipative systems makes it even costly to take no actions. Agents which reason deliberately can reduce arising costs by choosing a carefully selected low-cost plan. On the other hand, such committing agents might be confronted with a changed environmental situation when executing their plan after time-consuming deliberation in a dynamic environment. This might result in failure of execution. Reactive agents will not spend much time on reasoning but will execute the fastest plan, which is normally not the cheapest, in order to fulfill their goal. Because reactive agents do not spend much or no time on planning they are likely to fulfill most of their goal as long as these goals are within their cost range. In this paper we model this tradeoff problem and will show some simulation results.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AN00349328
• 書誌情報: 全国大会講演論文集; 巻 第52回, 号 人工知能と認知科学, p. 131-132, 発行日 1996-03-06
• 出版者 / 言語: ja
• 出版者: 情報処理学会