S&AS: FND: COLLAB: Adaptable Vehicular Sensing and Control for Fleet-Oriented Systems in Smart Cities

在未来的智能城市中，如何以自主方式(很少或根本不需要人工干预)感知、理解和管理城市规模的车辆系统，例如出租车，是提高城市交通效率的重要课题，例如缩短乘客的等待时间，降低司机的巡航里程，提高车辆系统运营商的收入。然而，目前的车载系统管理策略主要基于个体层面的数据知识，忽视了车队层面的丰富信息。在这个项目中，研究人员设计并实施了一种面向车队的车辆系统管理策略，该策略利用安装在所有车辆上的传感器的实时数据来提高车辆系统的整体性能。特别是，调查人员的目标是通过使用车载摄像头来检测街道上等待的乘客，并与附近的车辆和调度中心分享这些信息，通过调度策略来接这些乘客，从而提高出租车系统的性能。研究团队将从车队导向的角度，就如何设计自适应自动车辆传感和调度策略以提高城市机动性效率达成明确的理解，并可能应用于未来的全自动车队。对车辆感知和调度的理解将提高乘客的通勤效率和司机的能源消耗等日常生活质量，最终通过低里程改善社会环境。本研究开发了一个面向车队的传感和控制框架，使车队内的历史和实时数据能够无缝集成，用于自适应车辆感知、建模和控制。具体地说，该项目研究如何在车辆之间最好地利用时空相关的上下文信息(例如，车辆机动性、服务需求、干扰事件)。尽管这种相关性会随着时间和距离的推移而减弱，但基于以下研究可以自适应地实现自动车辆感知、建模和控制，以开发新的服务：(I)通过自主学习相关的车辆交互来实现可重构的车队范围协调感知；(Ii)通过结合深度学习、结构化学习和基于属性的学习，通过集体解释来自不同车辆的隐含数据来建立机动性现象模型；(Iii)通过反复考虑机队范围的知识来设计具有不确定性集的稳健调度策略和滚动范围控制框架，以提高机动性效率。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。

In smart cities of the future, how to sense, understand, and manage urban-scale vehicular systems, e.g., taxis, in an autonomous fashion (with little or no human intervention) is an essential topic to improve urban mobility efficiency, such as shorter waiting time for passengers, lower cruising mileage for drivers, and higher revenues for vehicular system operators. However, the current management strategies for vehicular systems are mainly based on individual-level data knowledge, ignoring rich information from a fleet perspective. In this project, the investigators design and implement a fleet-oriented management strategy for vehicular systems, which utilizes real-time data from sensors installed in all vehicles to improve the overall performance of the vehicular system. In particular, the investigators aim to improve the taxi system performance by using onboard cameras to detect waiting passengers on streets and share this information with nearby vehicles and dispatch centers to pick up these passengers through a dispatching strategy. The research team will develop a clear understanding of how to design an adaptive autonomous vehicular sensing and dispatching strategy to improve urban mobility efficiency from a fleet-oriented perspective, with potential applications to future fully autonomous fleets. Such an understanding on vehicular sensing and dispatch will improve the quality of the every-day life such as more efficient commute for passengers, and lower energy uses for drivers, and finally improve the environment for the society by low vehicle mileage.This research develops a fleet-oriented sensing and control framework to enable seamlessly integration of historical and real-time data within a fleet for adaptive vehicular sensing, modeling, and control. Specifically, this project studies how to best use spatiotemporally-correlated contextual information (e.g., vehicular mobility, service demand, disruptive events) among vehicles. Although such correlations decay over time and distance, it is possible to enable autonomous vehicular sensing, modeling, and control adaptively based on the following research to develop novel services: (i) reconfigurable fleet-wide coordinated sensing by autonomously learning correlated vehicular interactions; (ii) models of mobility phenomena by collectively interpreting implicit data from different vehicles with a combination of deep learning, structured learning, and attribute-based learning; (iii) designs of robust dispatching strategies with uncertainty sets and receding horizon control frameworks by iteratively considering fleet-wide knowledge to improve mobility efficiency.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.