NRI: Collaborative Research: Dynamic Robot Guides for Emergency Evacuations

人群踩踏是人类危害性最大的集体行为之一。在历史上发生的事件中，由于火灾爆发或枪支意外走火等引起的恐慌比最初的触发事件具有更大的危险。该项目支持关于人机交互对人群动态影响的基础研究，旨在设计动态机器人控制算法以协助人类并防止紧急情况下的恐慌。这项研究的最终目标是开发出能够满足各种需求的可重构机器人指南。其中包括不同类型的紧急疏散，以及涉及人群大规模流动的非紧急情况，例如游行、音乐会或其他大型公共活动。该项目通过以机器人为中心的教育和短期课程开发将研究与教育活动结合起来。为了让年轻一代接触科学技术，该项目将与大学教育中心和社区学院合作开展各种外展活动。该项目的目标是研究人群动力学中的人机交互，开发最优反馈控制来调节人流分布，并设计机器人辅助紧急疏散算法。该研究将通过实验验证和测量群体动力学中控制人机交互的交互力，推进人机交互的最先进水平，并填补机器人研究的空白。所提出的机器人运动原始设计带来了基于学习的机器人运动规划的新方法，以有效地吸引人类。该项目验证了动态机器人引导在室内环境中真实人机交互实验中的使用。还将在购物中心和校园建筑等基准环境中进行模拟验证，并评估替代机器人辅助疏散策略的效率。虽然主要针对智能机器人，但研究成果预计将具有交叉性并适用于交通、通信和控制等其他领域。

Crowd stampede is one of the most harmful collective human behaviors. In incidents throughout history, panic due, for example, to the outbreak of fire or the unexpected discharge of firearms has been a greater hazard than the original triggering events. This project supports fundamental research on the influence of human-robot interaction on crowd dynamics, towards the design of dynamic robot control algorithms to assist humans and prevent panic in emergency situations. The ultimate goal of this research will be reconfigurable robot guides that can respond to a variety of needs. These include different types of emergency evacuation, as well as non-emergency situations involving mass movement of crowds, such as at parades, concerts, or other large public events. The project integrates research with educational activities through robot-centric education and short course development. To engage the younger generation with science and technology, the project will partner with a university educational center and a community college for various outreach activities.The objective of the project is to investigate human-robot interaction in crowd dynamics, develop optimal feedback control to regulate human flow distribution, and design robot-assisted emergency evacuation algorithms. The research will advance the state-of-the-art in human-robot interaction, and fill a gap in robotics research by experimentally validating and measuring the interaction forces governing human-robot interaction in crowd dynamics. The proposed robot motion primitive design leads to new approaches for learning-based robot motion planning to efficiently engage humans. The project validates the use of dynamic robot guides in real human-robot interaction experiments in indoor environments. Simulation validation in benchmark environments such as shopping-malls and campus buildings will also be performed, and the efficiency of alternative robot-assisted evacuation strategies will be evaluated. While primarily for intelligent robots, the research results are anticipated to be cross-cutting and applicable to other areas such as transportation, communication, and control.