EAGER:AI-DCL:Cognitive-Behavior Model to Predict Human Reaction to Swarm AI Non-Compliance.

这个项目研究的是自主群体系统，它由大量的协作机器人组成，这些机器人与少数人类监督者一起工作。这样的系统保证了无与伦比的任务并行性、容错性、弹性，并使人类远离伤害。然而，有一些重要的技术问题阻碍了这些承诺的实现。特别是，对于人类管理者（提供战术输入和任务意图，而不是直接或间接控制）如何与群体智能互动，我们缺乏足够的理解。这种理解对于确定解决方案以减轻认知超载、保持人类群体信任和共享情况感知是必要的，特别是在实际可扩展到实际应用的人类群体团队中。研究人员将利用他们与特定利益相关者团体的接触，包括布法罗消防部门和参与溢油响应的政府/工业团体。该项目还将通过开发人工智能和机器人课程的课程材料，在相关会议上举办多机器人研讨会，以及向公众传播数据和模型，对教育、培训和更广泛的研究界产生影响。研究人员将进行一系列实验，并确定建模方法，以回答有关人类如何对具身群体代理的行为作出反应的基本问题。这些问题包括人类如何对群体的间接或故意不遵守行为做出反应，群体提供的反馈水平如何解释任何不遵守行为，人类如何识别并将错误（或感知错误）归因于群体，以及这如何影响他们的干预频率。他们将使用神经人体工程学方法来评估人类指标，如事件相关（大脑）潜力、干预倾向、情况意识和通过生理信息的认知工作量。在机器智能方面，他们将设计分散的群体行为，以允许研究独特因素的影响，例如搜索和对象传输应用中的不合规和反馈水平。研究人员在人机交互、人类心理生理监测和自主群体系统等领域拥有互补的专业知识，用于参与协同研究活动，这是该项目的核心。他们还可以使用关键的研究设施，包括大型动作捕捉环境、群集机器人竞技场、空中/地面群集机器人平台、脑机接口和生理监测装置。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

This project studies autonomous swarm systems, which consist of a large number of collaborating robots that work together with a few human supervisors. Such systems promise unmatched task parallelism, fault-tolerance, resilience, and keeping humans out of harm's way. However, there are important technical issues that stand in the way of realizing these promises. Specifically, there is a sizeable lack of understanding of how human supervisors (who provide tactical input and mission intent, as opposed to direct or indirect control) interact with embodied swarm intelligence. That understanding is imperative for identifying solutions to mitigating cognitive overload and preserving human-swarm trust and shared situation awareness, especially in humans-swarm teams that are actually scalable to practical applications. The researchers will leverage their engagement with particular stakeholder groups including the Buffalo Fire Department and Government/Industrial parties involved in Oil Spill Response. The project will also have impact on education, training, and broader research community through the development of curricular materials for AI and Robotics courses, multi-robotic workshops at pertinent conferences, and public dissemination of data and models.The researchers will conduct a set of experiments and identify modeling approaches to answer fundamental questions regarding how humans respond to the behavior of embodied swarm agents. Those questions include how humans react to circumstantial or deliberate non-compliance by the swarm and the level of feedback provided by the swarm to explain any non-compliance, how humans identify and attribute errors (or perceived errors) to swarms, and how this impacts their intervention frequency. They will use a neuro-ergonomic approach to estimate human metrics such as event-related (brain) potential, intervention tendencies, situation awareness, and cognitive workload via physiological information. On the machine intelligence side, they will design decentralized swarm behavior to allow studying impact of unique factors such as non-compliance and feedback levels in search and object transport applications. The researchers have complementary expertise in the areas of human-robot interaction, human psycho-physiological monitoring, and autonomous swarm systems for engaging in the synergistic research activities that are central to the project. They also have access to key research facilities including a large motion capture environment, a swarmbot arena, aerial/ground swarm robotic platforms, brain-computer interfaces, and physiological monitoring setups.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.