ITR: Observing, Tracking and Modeling Social Multiagent Systems

ITR：观察、跟踪和建模社交多代理系统

• 批准号: 0219850
• 负责人: Tucker Balch
• 金额: $ 44.22万
• 依托单位: Georgia Tech Research Corporation
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2002
• 资助国家: 美国
• 起止时间: 2002-09-01 至 2006-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0219850&HistoricalAwards=false
• 关键词: ITR; Observing; Tracking; Modeling; Social

Tracking social agents that have a variety of mental states is quite different from and more challenging than tracking more predictable systems, say aircraft or ships. While multi-target tracking systems typically use physics-based models to track maneuvering vehicles, this work centers on developing probabilistic models of behavior and mental state to address the effects of interactions between agents. This work has practical applications in many areas including, for example: monitoring crowds of people, analyzing urban traffic patterns, and understanding robot and human team behavior. Additionally, this work should accelerate progress in experimental behavioral ecology, especially in the study of social insect systems. For this reason the investigation will initially be conducted in collaboration with biologists in the context of tracking and analyzing the behavior of captive live ant colonies. This research will yield novel, probabilistic algorithms for sensor-based identification, tracking, and modeling of the behavior of social multiagent systems. A tight coupling between tracking and modeling is critical, especially when the observed system is composed of many agents. The focus is on using models of behavior to improve the accuracy of sensor-based tracking, and on using improved tracking to learn better models.

跟踪具有多种心理状态的社会主体与跟踪更可预测的系统（如飞机或船只）完全不同，而且更具挑战性。虽然多目标跟踪系统通常使用基于物理的模型来跟踪机动车辆，但这项工作的重点是开发行为和心理状态的概率模型，以解决代理之间相互作用的影响。这项工作在许多领域都有实际应用，例如：监控人群，分析城市交通模式，以及理解机器人和人类的团队行为。此外，这项工作还将加速实验行为生态学，特别是昆虫群居系统研究的进展。因此，调查将首先与生物学家合作，在跟踪和分析圈养活蚁群行为的背景下进行。这项研究将为基于传感器的识别、跟踪和社会多智能体系统行为建模产生新颖的概率算法。跟踪和建模之间的紧密耦合是至关重要的，特别是当观察到的系统由许多代理组成时。重点是使用行为模型来提高基于传感器的跟踪的准确性，并使用改进的跟踪来学习更好的模型。