Toward cognitive habile agents

迈向认知能力主体

• 批准号: 0534398
• 负责人: Robert St. Amant
• 金额: $ 37.53万
• 依托单位: North Carolina State University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2006
• 资助国家: 美国
• 起止时间: 2006-01-01 至 2009-12-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0534398&HistoricalAwards=false
• 关键词: Toward; cognitive; habile; agents

This project addresses tool use by intelligent agents--so-called cognitive "habile" agents--in their environment. The research is inspired by examples of tool use among animals, ranging from chimpanzees that fish for insects with sticks to crows that bend wire into hooks. Tool use is viewed as a basis for intelligent behavior. The project focuses on three issues: (1) establishing general characteristics of cognitive habile agents, including development of a taxonomy of tools and their usages; (2) identifying core capabilities of habile agents, including their ability to enhance their interaction with the environment via tools, to perform detailed internal simulations of hypothetical scenarios for tool use, and to exploit symmetries in between tools and task constraints to identify opportunities for appropriate action; and (3) evaluating different control architectures (e.g., hierarchical, iterative) for such agents. It builds upon the PI's recent NSF-supported work on agents and graphical user interfaces, which involved issues of control and cognitive modeling. The Broader Impact of this work will be realized in two ways related to undergraduate research and graduate education: undergraduate researchers are being involved in the research through the AURICS (Accelerated Undergraduate Research in Computer Science) program at NCSU, and a new graduate course focusing on building AI systems and topics related to this research (e.g., embodied agents, cognitive modeling, control architectures) is being developed.

这个项目解决了智能代理--所谓的认知“可移动”代理--在其环境中的工具使用问题。这项研究的灵感来自动物使用工具的例子，从用棍子抓昆虫的黑猩猩到把铁丝弯成钩子的乌鸦。工具使用被视为智能行为的基础。该项目集中在三个问题上：(1)建立认知可居住代理的一般特征，包括开发工具及其用法的分类；(2)确定可居住代理的核心能力，包括它们通过工具增强其与环境交互的能力，执行工具使用的假设场景的详细内部模拟的能力，以及利用工具和任务约束之间的对称性来确定采取适当行动的机会的能力；以及(3)评估此类代理的不同控制体系结构(例如，分层、迭代)。它建立在PI最近在NSF支持的代理和图形用户界面方面的工作基础上，这些工作涉及控制和认知建模问题。这项工作的更广泛影响将通过与本科生研究和研究生教育相关的两种方式实现：本科生研究人员正在通过NCSU的Aurics(加速本科生研究)计划参与研究，正在开发一门新的研究生课程，重点是构建人工智能系统和与这项研究相关的主题(例如，体现代理、认知建模、控制架构)。