CAREER: Formalizing the Concept of Teamwork in Heterogeneous Multi-Robot Systems

多机器人分析是一个新的研究领域，它为开发有效的多机器人团队提供了一种非常规的方法。多机器人分析极大地扩展了运动分析的概念，这是一种数据驱动的分析决策方法，用于提高团队运动的表现，改变了田径运动的学习和业务，研究团队内机器人如何协调，如何发展这些多机器人团队，以及如何最有效地让团队上场。这在搜救紧急情况下非常重要，例如建筑物倒塌或自然灾害，在这些情况下，尽可能派出最有效的可用团队将对结果和整体安全产生重大影响。学院早期职业发展(CALEAR)项目为这一新的研究和应用领域奠定了技术基础，促进了坦普尔大学附近高中生对支撑机器人技术的STEM学科的参与，并与费城地区的当地企业和技术公司合作，在快速增长的先进机器人领域建立了强大的当地劳动力队伍。该项目开发了多机器人分析的计算机物理基础和计算框架，这将导致为有效利用每个单独机器人的相对优势并适用于新的团队组成和情况的不同类型的多机器人系统制定协调和控制战略。该项目的结果包括标准化测试场景的集合、可扩展的仿真环境、评估多机器人系统性能的新指标，以及将在一系列场景中最大化整体性能的多代理强化学习框架。这些工具以开源软件开发工具包的形式提供，将与更大的研究界共享，以促进对多机器人系统的研究及其有效利用。最后，该项目将设计方法来预测多机器人系统在具有挑战性的新场景中的预期性能，并将为多智能体强化学习开发新的度量标准，将每个智能体的贡献正确地归因于团队的性能。总而言之，这些努力允许对如何组成不同的多机器人系统和设计有效的团队协调和控制策略进行细致入微的分析理解。该项目由跨部门机器人基础研究计划支持，该计划由工程指导委员会(ENG)和计算机和信息科学与工程指导委员会(CEISE)联合管理和资助。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

A new field of study, multi-robot analytics, provides an unconventional approach to developing effective multi-robots teams. Multi-robot analytics greatly extends the concepts of sport analytics, a data-driven analytical decision-making methodology for improving performance in team sports that has transformed the study and business of athletics, to study how robots within a team coordinate, how to develop these multi-robot teams, and how to field a team most efficiently. This matters greatly in search-and-rescue emergencies, e.g., a building collapse or a natural disaster, where fielding the most effective available team possible would make a significant difference in the outcome and overall safety. This Faculty Early Career Development (CAREER) project creates the technical foundations for this new field of research and its application, promotes the participation in the STEM disciplines underpinning robotics by high-school students near Temple University, and works with local businesses and technology companies in the Philadelphia region to build a strong local workforce in the rapidly growing field of advanced robotics. This project develops the cyber-physical foundations and the computational framework for multi-robot analytics, which will lead to the development of coordination and control strategies for heterogeneous multi-robot systems that effectively use the relative strengths of each individual robot and that generalize to new team compositions and situations. The result of this project includes a collection of standardized test scenarios, an extensible simulation environment, new metrics to assess the performance of multi-robot systems, and a multi-agent reinforcement learning framework that will maximize overall performance across a range of scenarios. These tools are provided in an open-source software development toolkit that will be shared with the larger research community to advance the study of multi-robot systems and their effective use. Finally, this project will devise methods to predict the expected performance of multi-robot systems in challenging new scenarios and will develop novel metrics for multi-agent reinforcement learning that properly attribute each agent’s contribution to the team’s performance. Collectively, these efforts allow for a nuanced and analytical understanding of how to compose heterogeneous multi-robot systems and design effective team coordination and control strategies.This project is supported by the cross-directorate Foundational Research in Robotics program, jointly managed and funded by the Directorates for Engineering (ENG) and Computer and Information Science and Engineering (CISE).This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.