自动驾驶交通系统的协同管控技术

Coordinated management and control of autonomous traffic systems

过去10年自动驾驶技术的持续进步,为交通系统的协同管控技术提升带来了巨大的机遇和挑战.在微观层面,自动驾驶车因其优于人类驾驶员的反应时间和对周边微环境的精准监控,有望大幅提高道路通行能力,减少事故发生率,消除随机驾驶行为对系统稳定性和鲁棒性的影响.在宏观层面,交通系统管理者可以通过控制自动驾驶车的路径、车道,乃至详细轨迹,抑制拥堵的外部性,减少甚至消除无效拥堵,并以此为出发点指导下一代智慧公路系统的规划设计.本文以此为纲,总结了近年来自动驾驶交通系统协同管控技术领域的研究成果.

The rapid development of autonomous driving technology in the past decade has brought unprecedented opportunities and challenges to the coordinated management and control of traffic systems. Compared to human drivers, autonomous vehicles(AV) are equipped with shorter reaction time and better situational awareness, promising to increase the throughput of existing road networks, dramatically improve safety and robustness, and reduce the instability of vehicular traffic flow. Moreover, controlling the route, lane and even trajectory of every AV in a network could effectively eliminate excessive congestion caused by "selfish" human travel and driving behaviors. This ability to coordinate the entire AV fleet thus has important implications for planning the next-generation "smart" road systems.In this paper we consider the coordinated management and control of AV fleets from both macroscopic and microscopic perspectives. The former focuses on how coordinating macroscopic travel decisions, such as route, departure time and mode(e.g., ridesharing vs. micro-transit), might help reduce traffic congestion caused by what is widely known as the price of anarchy in transportation. While the underlying idea is simply moving the system from a user equilibrium state to a system optimal one, the challenge is to model the complex interactions between autonomous and human-driven vehicles, to anticipate how human drivers would behave in such an environment, and to design effective, fair and practical schemes to achieve the goal. What we learn from these efforts could potentially impact infrastructure planning, especially in terms of whether and how AVs should be separated from human-driven vehicles on road.On the other hand, because AVs can be programmed to remember and execute all maneuvers in a journey, much like the auto pilot for airplanes and high-speed trains, the entire trajectory of their journey, not just the aforementioned macroscopic travel decisions, may be targeted for the purpose of traffic management. Here, the focus is to manage the microscopic maneuvers(e.g., lane changing and gap-keeping behaviors) of a fleet of AVs to improve the performance of the road network. An interesting analogy can be drawn between this task and the task of managing a railway network in which many fast moving trains operate. Accordingly, the lanes on highways may be viewed as "virtual tracks" in an imaginary railway network. There is an extensive literature on railway operations from which one can draw useful mathematical tools and empirical results. A primary challenge to operationalize this analogy is the enormous computational requirement caused by both the sheer size of the problem, as well as the short time framework(seconds or shorter) within which such a coordinated management and control decision must be made.