[论文解读] Key Ingredients of Self-Driving Cars

Rui Fan, Jianhao Jiao+4 more|arXiv (Cornell University)|Jun 7, 2019

Autonomous Vehicle Technology and Safety参考文献 62被引用 33

一句话总结

本论文梳理自动驾驶汽车的关键组成要素，涵盖驾驶自动化等级、传感器、软件架构、开放数据集、行业参与者、应用与挑战。其目标是提供全面的综述，而非深入单一技术领域。

ABSTRACT

Over the past decade, many research articles have been published in the area of autonomous driving. However, most of them focus only on a specific technological area, such as visual environment perception, vehicle control, etc. Furthermore, due to fast advances in the self-driving car technology, such articles become obsolete very fast. In this paper, we give a brief but comprehensive overview on key ingredients of autonomous cars (ACs), including driving automation levels, AC sensors, AC software, open source datasets, industry leaders, AC applications and existing challenges.

研究动机与目标

• 推动需要对自动驾驶进行广泛、整体性的综述，超越单一领域的研究。
• 总结驾驶自动化等级以及感知、定位、预测、规划和控制在 ADS 中的作用。
• 调研常用传感器与硬件，以及基本的软件架构与数据集。

提出的方法

• 将现有的自动驾驶文献综合整理为模块化组件：感知、定位与建图、预测、规划和控制。
• 比较知名自动驾驶项目的软件架构，并将其映射到五模块框架。
• 概述开源数据集和行业格局，以为当前研究方向提供背景信息。

实验结果

研究问题

• RQ1 SAE 定义的驾驶自动化等级是什么，以及在每个等级中谁负责环境监控？
• RQ2 自动驾驶汽车通常使用哪些传感器和硬件配置，它们如何互补？
• RQ3 感知、定位、预测、规划与控制在自动驾驶软件架构中是如何组织的？
• RQ4 哪些开源数据集和行业领头企业塑造了当前的自动驾驶研究与开发？
• RQ5 在自动驾驶中，实时感知、大规模 SLAM 和传感器数据融合面临的关键挑战有哪些？

主要发现

• SAE driving automation levels are described, with a progression of responsibilities from human drivers to ADS as levels increase.
• Perception, localization and mapping, prediction, planning, and control form a five-module software architecture that aligns with established ADS architectures.
• A variety of sensors (cameras, lidars, radars, GPS, IMU, wheel encoders) and hardware controllers (CAN bus) are discussed as trade-offs for perception and localization.
• Open source datasets such as Cityscapes, ApolloScape, KITTI, and 6D-vision are highlighted as influential resources for autonomous driving research.
• Industry leaders (e.g., Tesla, Waymo, GM) and their public road testing milestones are summarized to illustrate deployment progress and competitive landscape.
• Key challenges include perception under adverse conditions, real-time computation limits, SLAM stability in large-scale environments, and rapid, cost-effective sensor data fusion.