CAREER: Driving the Future: Models and Control Methods to Coordinate Fleets of Self-Driving Vehicles in Future Transportation Networks

该学院早期职业发展（Career）计划项目推进了机器人网络建模、分析和控制的科学知识，这些网络由无人驾驶车辆组成，以协调的方式自主运行，以满足人员或货物运输等服务要求。为了有效地工作，这样的系统必须克服分配和调度方面的挑战，这些挑战在实践中可能会产生备份、不可接受的等待时间和有害的级联效应。本项目将在空间排队理论的框架下，研究优化车辆分配的理论模型和实时控制方法。理论和控制算法将应用于自主移动按需系统的设计、系统范围控制和经济评估。这种系统代表了一种变革性的、快速发展的交通方式，即电动、自动驾驶的班车运输城市乘客，为不能或不愿开车的人提供一种出行选择。该项目的成果将通过培养清洁、高效的未来交通系统和解决21世纪的出行需求，使美国经济受益。更广泛地说，这项研究适用于一大类机器人协调问题，并将对包括自动化供应链、物流和国家安全在内的几个关键领域产生积极影响。全尺寸自动驾驶航天飞机的实验将有助于扩大代表性不足的群体在研究中的参与，并促进网络物理系统的工程教育。目前控制机器人网络的方法是有限的，特别是在预测精度和控制综合与正式性能保证方面。空间排队理论认为动态系统包括(i)空间本地化的队列，该队列收集由外生动态过程生成的服务请求，以及（ii）机器人服务车辆在给定网络拓扑中的队列中行驶。因此，空间排队理论模拟了各种各样的机器人协调问题，自主移动按需系统是一个相关的例子。该项目将利用随机网络优化的最新算法技术，为日益复杂和现实的空间排队系统的建模、分析和控制生成可证明正确的工具，从而推进该领域的知识。具体而言，该奖项支持基础研究：1)通过设计复杂设置中易于处理的分析方法来推进空间排队系统理论；2)生成具有性能保证的控制方法，为机器人车辆提供服务请求的最佳分配；3)通过案例研究和在全尺寸试验台上部署算法，将理论和控制方法应用于自主移动按需系统的控制。

This Faculty Early Career Development (CAREER) Program project advances scientific knowledge on the modeling, analysis, and control of robotic networks consisting of unmanned vehicles autonomously operating in a coordinated fashion to fulfill service requests such as the transportation of people or goods. To work efficiently, such systems must overcome allocation and scheduling challenges that, in practice, can create backups, unacceptable wait times, and detrimental cascade effects. This project will cast the problem within the framework of spatial queuing theory, and investigate theoretical models and real-time control methods to optimally allocate vehicles to service requests. Theory and control algorithms will be applied for the design, system-wide control, and economic assessment of autonomous mobility-on-demand systems. Such systems represent a transformative, rapidly developing mode of transportation where electric, self-driving shuttles transport urban passengers and provide a mobility option to people unable or unwilling to drive. The results of this project will benefit the U.S. economy by fostering clean and efficient future transportation systems and addressing 21st century mobility needs. More broadly, this research is applicable to a large class of robotic coordination problems and will positively impact several critical sectors including automated supply chains and logistics and national security. Experiments on full-scale autonomous shuttles will help broaden the participation of underrepresented groups in research and catalyze engineering education on cyber-physical systems. Current methods for controlling robotic networks are limited, particularly with respect to predictive accuracy and control synthesis with formal performance guarantees. Spatial queuing theory considers dynamic systems consisting of (i) spatially-localized queues that collect service requests generated by an exogenous dynamical process, and (ii) robotic service vehicles traveling among queues in a given network topology. As such, spatial queuing theory models a large variety of robotic coordination problems, with autonomous mobility-on-demand systems as a relevant example. The project will advance knowledge in the field by leveraging recent algorithmic techniques from stochastic network optimization to generate provably-correct tools for the modeling, analysis, and control of spatial queuing systems of increasing complexity and realism. Specifically, this award supports fundamental research to 1) advance the theory of spatial queuing systems, by devising methods for tractable analyses in complex setups, 2) generate control methods with performance guarantees for the optimal assignment of robotic vehicles to service requests, and 3) apply theory and control methods to the control of autonomous mobility-on-demand systems, through case studies and the deployment of algorithms on full scale test beds.