CAREER: Decentralized Constraint-Based Optimization for Multi-Agent Planning and Coordination

职业：用于多智能体规划和协调的分散式基于约束的优化

• 批准号: 1550662
• 负责人: William Yeoh
• 金额: $ 50万
• 依托单位: New Mexico State University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2016
• 资助国家: 美国
• 起止时间: 2016-03-01 至 2018-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1550662&HistoricalAwards=false
• 关键词: CAREER; Decentralized; Constraint; Based; Optimization

There is a growing need for optimization methods to support decentralized decision-making in complex multi-agent systems including target tracking in sensor networks, mission planning of unmanned autonomous vehicles, coordination of rescue robots in disaster scenarios, and scheduling of intelligent devices in smart homes within smart grids. This class of problems is particularly challenging to solve due to a combination of the following requirements: There is a high degree of uncertainty that must be taken into account during planning; the planning process must be done in a decentralized fashion; and the resulting plan must be executed in a decentralized way as well. The objective of this project is to respond to the crucial challenge of developing an integrated approach that captures all these requirements within a single framework in order to improve the scope and applicability of multi-agent techniques in real-world applications. The long-term broader impacts of this project include the potential for the research findings to improve decentralized decision-making in real-world problems. In the short term, high-school students will benefit from the education modules developed by the PI, which will be disseminated through collaborations with local outreach programs as well as local teachers and summer camp organizers. The students will develop better computational thinking skills and be exposed to computational concepts applied to relevant applications of interest. The significance of these efforts is made more crucial by the fact that a majority of the student body at local high-schools as well as at NMSU is Hispanic. This project will make the necessary foundational contributions to the field of multi-agent systems to improve the scope and applicability of such systems, especially those that utilize automated planning and constraint optimization techniques, in the real world. More specifically, this project will result in (i) novel ways to more accurately model a large class of multi-agent planning problems using decentralized constraint-based models; (ii) new scalable algorithms with theoretical guarantees suitable for solving large-scale decentralized planning problems; and (iii) effective ways of improving computational thinking in high-school students via the use of constraint-based representations.

在复杂的多智能体系统中，包括传感器网络中的目标跟踪、无人驾驶汽车的使命规划、灾难场景中救援机器人的协调以及智能电网中智能家居中智能设备的调度，对支持分散决策的优化方法的需求越来越大。由于以下要求的组合，这类问题特别具有挑战性：在规划过程中必须考虑高度的不确定性;规划过程必须以分散的方式进行;并且由此产生的计划也必须以分散的方式执行。该项目的目标是应对开发一种综合方法的关键挑战，该方法在一个单一的框架内捕获所有这些要求，以提高多代理技术在现实世界中的应用范围和适用性。该项目的长期广泛影响包括研究结果有可能改善现实世界问题中的分散决策。在短期内，高中学生将受益于PI开发的教育模块，这些模块将通过与当地外展计划以及当地教师和夏令营组织者的合作进行传播。学生将发展更好的计算思维技能，并接触到应用于相关应用的计算概念。这些努力的重要性是更重要的事实，即在当地高中以及在NMSU的大多数学生是西班牙裔。该项目将为多智能体系统领域做出必要的基础性贡献，以提高此类系统的范围和适用性，特别是那些利用自动规划和约束优化技术的系统，在真实的世界中。更具体地说，该项目将导致（i）新的方法，更准确地建模一个大类的多智能体规划问题，使用分散的基于约束的模型;（ii）新的可扩展算法与理论保证，适用于解决大规模分散规划问题;和（iii）通过使用基于约束的表示，提高高中生的计算思维的有效方法。

项目成果

Goal Recognition Design with Stochastic Agent Action Outcomes

具有随机代理行为结果的目标识别设计

• 发表时间: 2016
• 期刊: IJCAI
• 作者: Wayllace, Christabel;Hou, Ping;Yeoh, William;Son, Tran Cao
• 通讯作者: Son, Tran Cao