论文解读 - 城市自动驾驶车辆运动规划与控制技术综述 (第1部分)

🚗 Abstract（摘要）

Self-driving vehicles are a maturing technology with the potential to reshape mobility by enhancing the safety, accessibility, efficiency, and convenience of automotive transportation. Safety-critical tasks that must be executed by a selfdriving vehicle include planning of motions through a dynamic environment shared with other vehicles and pedestrians, and their robust executions via feedback control. The objective of this paper is to survey the current state of the art on planning and control algorithms with particular regard to the urban setting. A selection of proposed techniques is reviewed along with a discussion of their effectiveness. The surveyed approaches differ in the vehicle mobility model used, in assumptions on the structure of the environment, and in computational requirements. The side by side comparison presented in this survey helps to gain insight into the strengths and limitations of the reviewed approaches and assists with system level design choices.
自动驾驶汽车是一项成熟的技术，有潜力通过提高汽车运输的安全性、可达性、效率和便捷性来重塑其机动性。必须由自动驾驶车辆执行的安全关键任务包括在与其他车辆和行人共享的动态环境中规划运动，以及通过反馈控制实现鲁棒的执行力。本文的目的是综述城市环境中车辆规划和控制算法的现状。我们将回顾一些公开的技术，并对其有效性进行讨论。根据使用的车辆移动模型、对环境结构的假设以及计算要求，我们分别采用了不同的综述方法。本综述通过并排比较来讨论不同的技术方法，这有助于读者深入了解各种方法的优势和局限性，同时有助于系统级的设计选择。

🚚 I. Introduction（引言）

The last three decades have seen steadily increasing research efforts, both in academia and in industry, towards developing driverless vehicle technology. These developments have been fueled by recent advances in sensing and computing technology together with the potential transformative impact on automotive transportation and the perceived societal benefit: In 2014 there were 32,675 traffic related fatalities, 2.3 million injuries, and 6.1 million reported collisions [1]. Of these, an estimated 94% are attributed to driver error with 31%
involving legally intoxicated drivers, and 10% from distracted drivers [2]. Autonomous vehicles have the potential to dramatically reduce the contribution of driver error and negligence as the cause of vehicle collisions. They will also provide a means of personal mobility to people who are unable to drive due to physical or visual disability. Finally, for the 86% of the US work force that commutes by car, on average 25 minutes (one way) each day [3], autonomous vehicles would facilitate more productive use of the transit time, or simply reduce the measurable ill effects of driving stress [4].
过去三十年，学术界和工业界对开发无人驾驶汽车技术的研究力度稳步增加。这些发展得益于传感和计算技术的最新进展，以及对汽车运输的潜在变革性影响和已经被见证的社会效益：2014年，32675人因交通事故死亡，230万人受伤，610万人遭遇碰撞事故[1]。其中估计有 94% 的事故是因为驾驶员的错误，例如有 31% 的事故涉及法律上被判定为醉酒的驾驶员，有 10% 的事故涉及驾驶员的分心[2]。自动驾驶汽车有望大大减少由于驾驶员错误和疏忽而导致的车辆碰撞事故，同时为由于身体或视觉残疾而无法驾驶的人提供个人移动的方式。最后，考虑到 86% 的美国劳动力每天平均需要花费25分钟在驾驶通勤（单程）上[3]，自动驾驶汽车将有助于更有效地利用交通时间，或者至少减少驾驶压力带来的可衡量的不良影响[4]。

Considering the potential impacts of this new technology, it is not surprising that self-driving cars have a long history. The idea has been around since as early as the 1920s, but it was not until the 1980s that driverless cars seemed like a real possibility. Pioneering work led by Ernst Dickmanns (e.g., [5]) in the 1980s paved the way for the development of autonomous vehicles. At that time a massive research effort, the PROMETHEUS project, was funded to develop an autonomous vehicle. A notable demonstration in 1994 resulting from the work was a 1,600 km drive by the VaMP driverless car, of which 95% was driven autonomously [6]. At a similar time, the CMU NAVLAB was making advances in the area and in 1995 demonstrated further progress with a 5,000 km drive across the US of which 98% was driven autonomously [7].
考虑到这项新技术的潜在影响，自动驾驶汽车拥有长久的历史并不奇怪。早在20世纪20年代，这种想法就已经存在，但是直到20世纪80年代，无人驾驶汽车似乎才真正成为可能。Ernst Dickmanns在20世纪80年代领导的开创性工作（例如[5]）为自动驾驶汽车的发展铺平了道路。当时一项大规模的研究（PROMETHEUS项目）被资助于开发一种自动驾驶汽车。1994年，这项工作发布了一个引人注目的演示，其中VaMP无人驾驶汽车行驶了1600公里，并且有 95% 的路程是自动驾驶的[6]。与此同时，卡内基梅隆大学NAVLAB也取得了进展，它于1995年在美国境内完成了5000公里的行驶，其中 98% 的路程是自动驾驶的[7]。

The next major milestone in driverless vehicle technology was the first DARPA Grand Challenge in 2004. The objective was for a driverless car to navigate a 150-mile off-road course as quickly as possible. This was a major challenge in comparison to previous demonstrations in that there was to be no human intervention during the race. Although prior works demonstrated nearly autonomous driving, eliminating human intervention at critical moments proved to be a major challenge. None of the 15 vehicles entered into the event completed the race. In 2005 a similar event was held; this time 5 of 23 teams reached the finish line [8]. Later, in 2007, the DARPA Urban Challenge was held, in which vehicles were required to drive autonomously in a simulated urban setting. Six teams finished the event demonstrating that fully autonomous urban driving is possible [9].
无人驾驶汽车技术的下一个重大里程碑是2004年的第一次DARPA大挑战（DARPA Grand Challenge）。竞赛目标是让无人驾驶汽车在150英里的越野道路上尽可能快速地行驶。与之前的演示相比，这是一个重大挑战，因为在比赛期间没有人为干预。尽管先前的研究工作演示了几乎完全的自动驾驶，但在关键时刻不进行人为干预被证明是一项重大挑战。参加比赛的15辆车中没有一辆完成比赛。2005年举行了类似的活动，这一次23支车队中有5支到达了终点[8]。后来在2007年，DARPA城市挑战赛（DARPA Urban Challenge）举行，要求车辆在模拟的城市环境中自动驾驶。六支队伍完成了比赛，这证明完全自主的城市驾驶是可能的[9]。

Numerous events and major autonomous vehicle system tests have been carried out since the DARPA challenges. Notable examples include the Intelligent Vehicle Future Challenges from 2009 to 2013 [10], Hyundai Autonomous Challenge in 2010 [11], the VisLab Intercontinental Autonomous Challenge in 2010 [12], the Public Road Urban DriverlessCar Test in 2013 [13], and the autonomous drive of the Bertha-Benz historic route [14]. Simultaneously, research has continued at an accelerated pace in both the academic setting as well as in industry. The Google self-driving car [15] and Tesla’s Autopilot system [16] are two example commercial efforts that receive considerable media attention.
自DARPA挑战以来，已经开展了许多活动和大型自动驾驶车辆系统测试。值得注意的例子包括2009年至2013年的智能汽车未来挑战赛（Intelligent Vehicle Future Challenges）[10]、2010年的现代自动驾驶挑战赛（Hyundai Autonomous Challenge）[11]、2010年的VisLab洲际自动驾驶挑战赛（VisLab Intercontinental Autonomous Challenge）[12]、2013年的公共道路城市无人驾驶汽车测试（Public Road Urban Driverless-Car Test）[13]，以及伯莎-奔驰历史路线（Bertha-Benz historic route）的自动驾驶[14]。同时，学术界和工业界的研究都在加速进行。谷歌的自动驾驶汽车[15]和特斯拉的自动驾驶系统（Autopilot system）[16]是两个商业成果的例子，它们受到了媒体的广泛关注。

The extent to which a car is automated can vary from fully human operated to fully autonomous. The SAE J3016 standard [17] introduces a scale from 0 to 5 for grading vehicle automation. In this standard, the level 0 represents a vehicle where all driving tasks are the responsibility of a human driver. Level 1 includes basic driving assistance such as adaptive cruise control, anti-lock braking systems and electronic stability control [18]. Level 2 includes advanced assistance such as hazard-minimizing longitudinal/lateral control [19] or emergency braking [20], [21], often based upon set-based formal control theoretic methods to compute ‘worst-case’ sets of provably collision free (safe) states [22]–[24]. At level 3 the system monitors the environment and can drive with full autonomy under certain conditions, but the human operator is still required to take control if the driving task leaves the autonomous system’s operational envelope. A vehicle with level 4 automation is capable of fully autonomous driving in certain conditions and will safely control the vehicle if the operator fails to take control upon request to intervene. Level 5 systems are fully autonomous in all driving modes.
汽车的自动化程度从完全人工操作到完全自主。SAE J3016标准[17]引入了从0到5的等级，用于对车辆自动化进行分级。在该标准中，0级表示所有驾驶任务均由人类驾驶员负责的车辆。1级包括基本驾驶辅助，如自适应巡航控制、防抱死制动系统和电子稳定控制[18]。第2级包括先进的辅助措施，如危险最小化纵向/横向控制[19]或紧急制动[20][21]。这些辅助措施的基础通常是基于集的形式控制理论方法，用来计算可证明的无碰撞（安全）状态的“最坏情况”集[22–24]。在第3级，系统监控环境，并可以在特定条件下完全自主驾驶，但如果驾驶任务超出自主系统的操作范围，则仍需要操作员进行控制。具有4级自动化的车辆能够在某些条件下完全自动驾驶，并且如果操作员未能在请求干预时采取控制措施，系统仍然可以安全地控制车辆。5级系统在所有驾驶模式下都是完全自主的。

The availability of on-board computation and wireless communication technology allows cars to exchange information with other cars and with the road infrastructure giving rise to a closely related area of research on connected intelligent vehicles [25]. This research aims to improve the safety and performance of road transport through information sharing and coordination between individual vehicles. For instance, connected vehicle technology has a potential to improve throughput at intersections [26] or prevent formation of traffic shock waves [27].
车载计算和无线通信技术的可用性使汽车能够与其他汽车和道路基础设施交换信息，从而产生了一个与联网智能汽车密切相关的研究领域[25]。该研究旨在通过各车辆之间的信息共享和协调，提高道路运输的安全性和性能。例如，联网车辆技术有可能提高交叉路口的吞吐量[26]或防止车流冲击波的形成[27]。

To limit the scope of this survey, we focus on aspects of decision making, motion planning, and control for self-driving cars, in particular, for systems falling into the automation level of 3 and above. For the same reason, the broad field of perception for autonomous driving is omitted and instead the reader is referred to a number of comprehensive surveys and major recent contributions on the subject [28]–[31].
为了限制这篇综述的范围，我们将重点放在自动驾驶汽车的决策、运动规划和控制方面，特别是自动化级别为3级及以上的系统。出于同样的原因，这篇综述不讨论自动驾驶的感知领域；读者可以参考许多该方面的综述和最近的主要成果[28–31]。

The decision making in contemporary autonomous driving systems is typically hierarchically structured into route planning, behavioral decision making, local motion planning and feedback control. The partitioning of these levels are, however, rather blurred with different variations of this scheme occurring in the literature. This paper provides a survey of proposed methods to address these core problems of autonomous driving. Particular emphasis is placed on methods for local motion planning and control.
现代自动驾驶系统中的决策通常分为路线规划、行为决策、局部运动规划和反馈控制。然而，由于文献中报道了决策方案的不同变体，因此这些级别的划分相当模糊。本文概述了为解决自动驾驶核心问题而已经提出的方法，并且将侧重于局部运动规划和控制的方法。

The remainder of the paper is structured as follows: In Section II, a high level overview of the hierarchy of decision making processes and some of the methods for their design are presented. Section III reviews models used to approximate the mobility of cars in urban settings for the purposes of motion planning and feedback control. Section IV surveys the rich literature on motion planning and discusses its applicability for self-driving cars. Similarly, Section V discusses the problems of path and trajectory stabilization and specific feedback control methods for driverless cars. Lastly, Section VI concludes with remarks on the state of the art and potential areas for future research.
本文的其余部分结构如下：在第2节中，对决策过程的层次结构及其设计方法进行了高层次的概述。第3节回顾了为了实现运动规划和反馈控制而提出的用于近似城市环境中汽车的机动性的模型。第4节综述了运动规划的大量文献，并讨论了运动规划对自动驾驶汽车的可用性。同样，第5节讨论了无人驾驶汽车的路径和轨迹稳定问题以及具体的反馈控制方法。最后，第6节总结了最新技术和未来研究的潜在领域。