Simultaneous Localization And Mapping: A Survey of Current Trends in Autonomous Driving综述

Abstract

In this article, we propose a survey of the Simultaneous Localization And Mapping field when considering the recent evolution of autonomous driving. The growing interest regarding self-driving cars has given new directions to localization and mapping techniques. In this survey, we give an overview of the different branches of SLAM before going into the details of specific trends that are of interest when considered with autonomous applications in mind. We first present the limits of classical approaches for autonomous driving and discuss the criteria that are essential for this kind of application. We then review the methods where the identified challenges are tackled. We mostly focus on approaches building and reusing long-term maps in various conditions (weather, season, etc.). We also go through the emerging domain of multi-vehicle SLAM and its link with self-driving cars. We survey the different paradigms of that field (centralized and distributed) and the existing solutions. Finally, we conclude by giving an overview of the various largescale experiments that have been carried out until now and discuss the remaining challenges and future orientations.

在本文中，我们在考虑最近自主驾驶的发展时提出了一个同时定位和建图的调查。人们对自动驾驶汽车越来越感兴趣为定位和建图技术提供了新的方向。在本次调查中，我们先详细介绍SLAM的不同架构分支，然后详细介绍在考虑自动化应用时感兴趣的特定趋势。我们首先提出了经典自主驾驶方法的局限性，并讨论了这种应用所必需的标准。然后，我们回顾了如何处理公认的挑战。我们主要关注建立自动驾驶地图的方法，这种地图在各种条件（天气，季节等）下长期可重复使用。我们也遍历了新兴领域中的多车SLAM及其与无人车之间的关系。我们调查了该领域的不同范式（集中式和分布式）以及现有的解决方案。最后，我们总结一下人们迄今为止进行的各种大规模实验，并讨论剩余的挑战和未来的方向。

Introduction

Simultaneous Localization And Mapping (SLAM) has been a hugely popular topic among the mobile robotics community for more than 25 years now. The success of this field is tightly bound to the fact that “solving” the SLAM problem, that is localizing a robot thanks to a map of the surroundings built incrementally, has numerous applications ranging from spatial exploration to autonomous driving. The recent spotlight put on intelligent vehicles has pushed further the research effort with the contribution of car manufacturers. One could think of GNSS (Global Navigation Satellite System) as a solution to this localization problem but it has quickly been showed that it was not sufficient by itself. Even though accuracy limits of classical GNSS solutions are lifted when perfectly positioned base stations are used (Real-Time Kinematic GNSS), availability remains an issue. Satellite signals are affected by atmospheric conditions that are difficult to predict. Moreover, the infrastructure can block the direct reception of signals and generate multipath interference or non-line-of-sight reception which has disastrous consequences on the provided localization. This kind of signal degradation is difficult to detect and usually causes a loss of integrity from which recovering can be tricky. These problems are more common in dense urban areas where tall buildings can mask satellites. On open roads, GNSS’s usually perform better.

在移动机器人领域，同步定位和建图（SLAM）已经成为25年来非常流行的话题。该领域的成功与“解决”SLAM问题紧密相关，即通过增量式构建环境地图来定位机器人，它具有从空间探索到自动驾驶等众多应用。最近聚焦于智能汽车的焦点使汽车制造商的贡献进一步推动了研究工作。

人们可以将GNSS（全球导航卫星系统）视为解决这一个定位问题的解决方案，但很快就显露出它本身的不足。即使在使用完全定位的基站（实时动态GNSS）时，经典GNSS解决方案的精度被有效提升，但可用性仍然是一个问题。卫星信号受难以预测的大气条件的影响。而且基础设施可以阻止信号的直接接收并产生多路径干扰或非视距接收，这对GNSS提供的定位具有灾难性后果。这种信号的退化很难被检测到，并且通常会导致系统丧失完整性，并且较难恢复。这些问题在密集的城市地区更为常见，高层建筑可以掩盖卫星。在开阔的道路上，GNSS通常表现更好。

Another classic approach to localization is to take advantage of the road infrastructure (road markings or roadway detection) in order to guide a vehicle in a lane. Advanced Driver-Assistance Systems (ADAS) are already integrating lane marking detection in commercialized cars. While this kind of approach mostly constrains the lateral position of a vehicle, it is sufficient for environments where the roadway is easily identifiable such as highways for instance. On the other hand, more complex environments (urban mostly, with intersections, curved roads, etc.) do not always provide enough road information to localize a vehicle. Moreover, the position accuracy needed along the longitudinal axis is more important than in straight, highway-like environments. Anyhow, redundancy is needed in order to build a safe system and ensure a consistent behavior on the road. As such, different localization means should be considered.

In a general manner, localizing a vehicle, be it in a global or a local frame, is an essential functionality to perform any other perception or planification tasks. Predicting the evolution of other obstacles on the road and choosing which maneuver is the most appropriate require to know exactly where the ego-vehicle is located and how it will evolve in the coming seconds. The SLAM framework gives an answer to this problematic while still being general enough to allow the use of any sensor or estimation technique that suits the prerequisite of estimating both the localization of the vehicle and the map at the same time. The map is of prime interest when autonomous driving is considered as a whole as it offers a first level of perception that is needed in order to make appropriate decisions.

另一种经典的定位方法是利用道路基础设施（道路标记或道路检测）来引导车道中的车辆。先进的驾驶辅助系统（ADAS）已经被集成在商用车中进行车道标记检测。虽然这种方法主要限制车辆的横向位置，但对于道路易于识别的环境（例如高速公路）来说，这已足够。另一方面，更复杂的环境（主要是城市，交叉路口，弯道等）并不总是提供足够的道路信息来定位车辆，相对于比直线，高速公路类环境，这种环境下放入纵向轴线所需的位置精度更重要。无论如何，为了建立一个安全的系统并确保道路上的一致行为，需要冗余。因此，应该考虑不同的定位手段。

一般而言，无论是在全局还是局部系统中，车辆的定位是执行任何其他感知或规划任务的基本功能。预测道路上其他障碍物的演变和选择最合适的机动动作需要准确了解车辆自身的位置以及它在未来几秒内会如何演变。 SLAM框架为这个问题提供了答案，同时仍然具有足够的通用性，允许使用任何传感器或估计技术，以适应同时估算车辆定位和地图的先决条件。当自动驾驶被认为是一个整体时，地图是最重要的，因为它提供了第一级的知觉，以便作出适当的决定。

The SLAM problem is considered as one of the keys towards truly autonomous robots, and as such is an essential aspect of self-driving cars. However, many issues are still preventing the use of SLAM algorithms with vehicles that should be able to drive for hundreds of kilometers in very different conditions. This last statement encompasses the two main problems arising when dealing with SLAM for autonomous vehicles: localization tends to drift over time and maps are not necessarily viable in every driving condition. The former problem is well-known in the SLAM community. The local and incremental positioning estimation that is given by SLAM algorithms tends to diverge from the true trajectory as the traveled distance increases. Without prior knowledge or absolute information, it thus becomes almost impossible to ensure a proper localization for several kilometers. This leads to the second problem, namely having maps that are sufficient for the localization task no matter the conditions. The mapping aspect has gained a lot of attention lately with the objective of providing the necessary information to locate a vehicle in different seasons, weathers or traffic conditions. Many solutions have been envisaged to solve these two problems like building a map with a careful selection of distinctive information with the objective of reusing it later or taking advantage of new communication systems in order to share and enhance the maps built by other road users for instance.
SLAM问题被认为是真正的自动机器人的关键之一，因此他也是是自动驾驶汽车的一个重要方面。然而，许多问题仍然阻碍了SLAM算法在车辆中的应用，并且使其能够在非常不同的条件下行驶数百公里。上句话涵盖了处理自动驾驶汽车SLAM时出现的两个主要问题：定位倾向于随时间漂移，并且地图在每种驾驶条件下都不一定可行。前一个问题在SLAM社区中是众所周知的。由SLAM算法给出的局部和增量定位估计倾向于随着行进距离的增加而偏离真实的轨迹。如果没有先验知识或绝对信息，就几乎不可能确保几公里的适当定位。这导致了第二个问题，即无论条件如何，具有足以用于定位任务的地图。建图方面最近引起了很多关注，目的是为不同季节，天气或交通条件下的车辆定位提供必要的信息。人们已经设想了许多解决方案来解决这两个问题，例如通过仔细选择独特信息来制作地图，以便稍后重复使用或利用新的通信系统来共享和增强由其他道路用户构建的地图。

SLAM is a broad field and involves many topics ranging from sensor extraction to state estimation. In this article, we propose a survey focusing on the current trends in the SLAM community regarding the emergence of the autonomous vehicle market. To be clear, throughout this paper, we will refer to SLAM as approaches that are composed of at least an odometry and a mapping module in order to cover all the techniques that are of interest for autonomous driving. We will first give a general introduction to SLAM by reviewing the most commonly used estimation techniques, discuss the different existing benchmarks and data sets and point to the relevant surveys covering aspects that are not reviewed in this article. This will be the object of Section II. Then, Section III will concern the limits of classical methods for autonomous driving and more especially the impact of the drift. The problem will first be stated and we will then focus on the solutions proposed to avoid or correct this drift and on the general reliability of the exploited information. We will end this section by discussing the criteria that should be respected for a SLAM approach to be viable for autonomous driving. Section IV will concern the techniques that tackle some of the challenges of SLAM for autonomous cars, namely building and exploiting long-term maps. A survey of the methods that take advantage of previous knowledge, coming from the SLAM algorithm itself or from another resource (Geographic Information Systems for instance), will be proposed as it is a key aspect to achieve true autonomy for self-driving cars. Section V will give an overview of multi-vehicle SLAM systems. This field offers a solution to both the problems mentioned earlier: constraining drift and enhancing maps. The different design choices of such applications will be exposed and motivated with the related state of the art. Finally, Section VI will expose the large-scale experiments that have been carried out so far with relation to the localization means used. It will serve as a basis to discuss the future orientations and the remaining challenges in Section VII.

SLAM是一个广泛的领域，涉及许多主题，从传感器提取到状态估计。在本文中，我们提出一项调查，重点关注SLAM社区目前关于自动车市场出现的趋势。清楚的是，在本文中，我们将把SLAM看为由里程计和一个建图模块组成的方法，以涵盖所有对自动驾驶感兴趣的技术。我们首先通过回顾最常用的估算技术，对不同的现有基准和数据集进行讨论，并指出本文未涉及的相关方面的相关调查，从而对SLAM进行一般性介绍。这将成为第二部分的目标。然后，第三节将关注传统自动驾驶方法的局限性，尤其是漂移的影响。首先说明问题，然后我们将重点讨论为避免或纠正这种漂移以及所利用信息的总体可靠性而提出的解决方案。在本节末尾我们将通过讨论SLAM方法应该遵守的标准以使得自动驾驶变得可行。第四节将涉及解决自动驾驶汽车SLAM所面临的一些挑战，即构建和开发长期地图。我们对利用SLAM算法本身或其他资源（例如地理信息系统）利用先前知识的方法进行调查，因为这是实现真正的自动驾驶汽车的自主性的关键方面。第五部分将概述多车SLAM系统。该领域为前面提到的问题提供了解决方案：约束漂移和增强地图。这些应用程序的不同设计选择将会被现有技术的暴露和激发其动力。最后，第六部分将揭示迄今为止所进行的大规模实验与所使用的定位手段的关系。它将作为讨论第七部分未来方向和剩余挑战的基础。

The SLAM Problem

Initiated by Smith and Cheeseman [1] in 1986, the SLAM topic got popular during the 1990s with many structuring works like [2] or [3]. Through the years, new methods have appeared using different sensors (camera, laser, radar, etc.), creating new data representations and consequently new types of maps. Similarly, various estimation techniques have emerged inside the SLAM field. A quick panorama has been made 10 years ago in [4] and [5]. Readers looking for an initiation to the global SLAM problem can also refer to [6] and [7] for a comprehensive introduction to the topic and to [8] for an extensive up-to-date review of the current challengesin SLAM.

SLAM问题由Smith和Cheeseman于1986年提出[1]，SLAM话题在20世纪90年代受到了许多结构化工作的欢迎，如[2]或[3]。多年来，使用不同传感器（照相机，激光，雷达等）出现了新的方法，创建新的数据表示，从而创建新类型的地图。同样，在SLAM领域内出现了各种估算技术。 10年前[4]和[5]已经做出了快速的全景图。寻找全局SLAM问题的读者也可以参考[6]和[7]，他们全面介绍该主题，另外[8]对SLAM当前的挑战进行广泛的最新评论。

SLAM methods require a wide panel of algorithms to ensure the robustness of the provided localization. As such, sensor data extraction, primitive search [9], data association [10] or map storage and updates [11] are part of the topics which concern SLAM as well. However, in this literature review, we will first focus on the main estimation methods that exist before going through the current trends in the autonomous vehicle application field.
The SLAM problem is usually formalized in a probabilistic fashion. The whole objective is to be able to estimate at the same time the state of the vehicle and the map being built. The vehicle state can be defined differently depending on the application: 2D position and orientation, 6D pose, speed, acceleration, etc. We denote $\mathbf{x}_k$ the vehicle pose estimation at the time $k$ and $\mathbf{m}$ the map of the environment. To estimate these variables, it is possible to take advantage of what we call control inputs u k and which represent an estimation of the motion between $k−1$ and $k$. They usually come from wheel encoders or any sensor able to give a first idea of the displacement. The particularity of SLAM approaches is to take
into consideration measurements coming from sensor readings and denoted $\mathbf{z}_k$ . They help to build and improve the map and indirectly, to estimate the vehicle pose.

SLAM方法需要大量的算法来确保所提供的定位的稳健性。因此，传感器数据提取，基元搜索[9]，数据关联[10]或地图存储和更新[11]也是涉及SLAM的主题的一部分。然而，在这篇文献综述中，我们将在遍历无人车应用领域当前趋势之前首先关注现存则主要主流方法。

SLAM问题通常以概率方式表示。整体目标是能够同时估算车辆的状态并构建地图。车辆状态可以根据应用不同而定义：2D位置和方向，6D位姿，速度，加速度等。我们使用$\mathbf{x}_k$表示车辆在时间$k$的位姿，$\mathbf{m}$表示的环境地图。为了估计这些变量，有可能利用我们称之为控制输入$\mathbf{u}_k$，他可以作为$k-1$和$k$之间的运动的粗略估计。它们通常来自轮子编码器或任何能够首先提供位移的传感器。SLAM方法的特殊性在于采取考虑来自传感器读数的测量结果并记为$\mathbf{z}_k$。他们帮助建立和改进地图，间接地估计车辆位姿。

The SLAM problem can be formulated in two ways. In the first one, the goal is to estimate the whole trajectory of the vehicle and the map given all the control inputs and all the measurements. A graphical representation can be seen in Figure 1. This problem, known as full SLAM, computes the joint posterior over all the poses and the map based on the entirety of sensor data:
SLAM问题可以用两种方式来表达。在第一个中，目标是估计车辆的整个轨迹以及在给定所有控制输入和所有测量值的情况下的地图。图1中可以看到图形表示。这个问题称为全SLAM，它根据传感器数据的整体计算所有位姿和地图的后验分布。

$$bel(\mathbf{x}_{0:k},\mathbf{m}) = p(\mathbf{x}_{0:k},\mathbf{m}|\mathbf{z}_{0:k},\mathbf{u}_{0:k})$$

图1 Graphical representation of the full SLAM problem 全SLAM问题的的图表示
The full SLAM problem can be difficult to handle in real time as the complexity of the problem grows with respect to the number of variables considered. The idea of online SLAM is to estimate the current position of the vehicle, usually based on the last sensor information. A graphical representation is shown in Figure 2. The incremental nature of the problem can be obtained using Bayes’ rule:

The estimation techniques can be separated into two main categories: filter-based approaches and optimization-based methods. The former corresponds to iterative processes that are thus suited to online SLAM and the latter regroups methods performing batch treatments and, as such, are usually applied to solve the full SLAM problem even if this trend has changed during the last ten years.

完整的SLAM问题可能难以实时处理，因为问题的复杂性与所考虑变量的数量有关。在线SLAM的想法是估计车辆的当前位置，通常基于最新的传感器信息。图2显示了图表示。问题的增量性质可以使用贝叶斯规则获得：

bel(xk,m)=p(xk,m|z0:k,u0:k)∝p(zk|