Understanding Pedestrian Dynamics for Seamless Human-Robot Interaction

在不久的将来，机器人将在繁忙、拥挤和不受限制的环境中与人类一起导航，如购物中心、机场和老年人护理设施。目前的移动机器人运动规划是不够的，因为目前的行人动力学模型不能完全捕捉到人群中人类运动行为的复杂性。因此，机器人可以“冻结”在人群密度高的地方，这进一步阻碍了行人交通流量。该项目将使用新颖的机器学习技术来开发具有类似人类导航功能的机器人运动规划器。项目团队将使用客观和主观的性能指标来评估人机交互的可预测性和可接受性，以及部署在购物中心和校园建筑中的机器人及其新型控制算法。该项目符合国家利益，因为由此产生的行人动力学建模方法和机器人控制器可能会导致公共安全应用，如紧急疏散和人群规划/管理。该项目将包括一个教育部分，为研究生和本科生提供工程和研究方法培训，并为高中生和中学生提供STEM推广。将作出更多努力，吸引和留住妇女从事科学和工程事业。本研究探讨了新的控制方法，有望改善行人/移动机器人在拥挤和无约束环境中的导航。方法包括从现有数据集中提取与行人行为相关的特征，并将其用于训练深度神经网络（DNN）来模拟行人动态；使用逆强化学习生成人类遵循的导航成本图；地图在机器人运动控制器内的实现；并根据客观和主观性能标准对控制器在真实人类环境中的性能进行了实验测试和验证。实验和评价数据将提供给研究界使用。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。

In the near future, robots will navigate alongside people in busy, crowded and unconstrained environments such as shopping malls, airports and elder-care facilities. Current mobile robot motion planners are inadequate because current models of pedestrian dynamics do not fully capture the complexity of human motion behavior in crowds. As a result, robots can "freeze" in places where crowd density is high, which further impedes pedestrian traffic flow. This project will use novel machine learning techniques to develop a robot motion planner with human-like navigations features. The project team will use objective and subjective performance measures to evaluate the predictability and acceptability of human-robot interactions with robots and their novel control algorithms deployed in shopping malls and campus buildings. This project serves the national interest because the resulting pedestrian dynamics modeling methods and robot controllers may result in public safety applications such as emergency evacuations and crowd planning/management. The project will involve an educational component that provides engineering and research methods training to graduate and undergraduate students, as well as STEM outreach to high-school and middle-school students. Additional efforts will be made to attract and retain women into careers in science and engineering. This research investigates new control methodologies that promise improved pedestrian/mobile-robot navigation through crowded and unconstrained environments. Methods include the extraction of features related to pedestrian behavior from existing datasets and their use in training a deep neural network (DNN) to model pedestrian dynamics; the use of inverse reinforcement learning to generate a cost map that humans follow to navigate; the implementation of the map within a robot motion controller; and the experimental testing and validation of the controller in real-world human environments using objective and subjective performance criteria. The experimental and evaluation data will be made available for use by the research community.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.