跳转到内容

A Survey of Autonomous Driving: Common Practices and Emerging Technologies

A Survey of Autonomous Driving: Common Practices and Emerging Technologies

Section titled “A Survey of Autonomous Driving: Common Practices and Emerging Technologies”

⚠️ AI 生成 · 建议对照原文 本页为自动整理的学习笔记;关键数据与引用如需引用,请回查 PDF / 官方版本。

学习档位 中文笔记

类型 文献 · 更新 2026-07-19

所属 自动驾驶 3D 感知、时序融合与跟踪 · 端到端学习

中文学习笔记(自动生成,需核验)

Section titled “中文学习笔记(自动生成,需核验)”

Topic: autonomous-driving · Tier: foundational · Year: 2020 · Venue:
Evidence level: partial · 本地全文: 是 · 建议阅读: ~90 分钟
Paper: https://arxiv.org/abs/1906.05113
Code:
Generator: grok

这篇2020年综述全面覆盖自动驾驶系统(ADS)的挑战、高层架构(ego-only/connected、modular/end-to-end)、核心功能(定位建图、感知、评估、规划决策、控制、人机接口)及新兴趋势,并通过在自有平台真实驾驶中实现比较SOTA算法,同时总结数据集与工具,适合建立领域全貌并理解设计取舍。

本文系统综述自动驾驶常见实践与新兴技术,对比模块化与端到端等架构,指出level 4/5与城市场景鲁棒性仍为开放问题。

ADS承诺安全、舒适、高效驾驶并带来巨大社会效益,但94%事故由人为错误引起,现有ADS事故上升、城市环境鲁棒自动化尚未实现,level 3及以上接管风险与伦理困境等挑战突出,需全面调查未解问题与技术现状。

车辆动力学与控制基础、计算机视觉与深度学习、机器人学、传感器模态(如lidar)、SAE自动化等级定义及智能交通系统常识。

  • 讨论未解决问题并全面调查ADS技术方面,包括当前挑战、高层系统架构、新兴方法与核心功能(定位、建图、感知、规划、人机接口)
  • 在自有平台真实世界驾驶设置中实现并比较多种SOTA算法
  • 总结可用数据集与ADS开发工具
  • 通过表格对比现有综述,覆盖更全面(含connected系统、end-to-end、HMI、datasets与implementation)

作为综述,先分类系统架构(standalone ego-only vs connected multi-agent;modular pipeline vs end-to-end),再分模块回顾定位建图、感知、场景评估、规划决策、车辆控制与HMI,最后概述数据集与工具;信息流从传感器输入经各模块到执行器输出。

核心模块包括localization and mapping、perception、assessment、planning and decision making、vehicle control、human-machine interface。设计取舍:modular将任务分解为易解决子问题并支持冗余与安全硬编码,但易误差传播且复杂;end-to-end直接从感知到动作(监督/强化学习/神经进化),但缺乏硬编码安全与可解释性,且真实城市场景应用有限;ego-only实用便于开发,connected潜在益处大(共享感知)但额外挑战多(如VANET安全路由)。

论文结论部分概述可用数据集与ADS开发工具(Section IX);作者在自有平台用开源软件实现比较SOTA算法于真实驾驶设置,但具体数据集名称、实验设置与定量指标待来源核验。

待来源核验

level 4/5尚未实现(尤其城市网络);系统失败导致事故(感知误分类传播等);level 3接管增加碰撞风险;端到端缺乏安全措施与可解释性且真实城市未充分验证;connected系统尚无运营实例;伦理困境(不可避免事故行为)、风险可靠性认证与多目标优化(时间/油耗/舒适)使问题复杂;极度谨慎的ADS也产生负面印象。

与前序 / 同期 / 后续方法的关系

Section titled “与前序 / 同期 / 后续方法的关系”

与现有综述[14]-[25]对比(Table 1),本文覆盖更广(含connected、end-to-end、assessment、HMI、datasets、implementation);历史脉络包括Eureka PROMETHEUS(1987-1995,VITA II公路自动驾驶)、DARPA Grand Challenge(2004全失败、2005五队完成)、DARPA Urban Challenge(2007六队完成但缺行人/自行车等真实元素)及后续竞赛;端到端可追溯至ALVINN等。

作者使用开源软件在自有平台实现比较部分SOTA算法;无明确官方代码仓库提及。复现建议:基于开源ADS软件栈在真实或仿真平台对比modular与end-to-end组件,优先验证感知-规划误差传播与安全约束。具体链接与细节待来源核验。

先读Abstract与Introduction建立动机与历史;再Section II Prospects and Challenges;然后Section III System Components and Architecture(含Figure 1/2与Table 2);接着按IV Localization、V Perception、VI Assessment、VII Planning、VIII HMI顺序;最后IX Datasets and Tools;对照Table 1理解贡献边界。

  1. Q: SAE定义的自动化等级中,level 3与level 4的主要区别是什么? A: Level 3为有条件自动化,驾驶员需快速响应紧急接管且ODD有限(如高速);level 4及以上无需人类注意力,但level 4仍限特定ODD,离开后需自动停车。
  2. Q: 模块化ADS的主要优缺点是什么? A: 优点:分解任务易解决、可叠加安全约束与冗余、直接复用机器人/视觉/动力学知识;缺点:易误差传播(如感知误分类导致事故)与过度复杂。
  3. Q: 端到端驾驶的三种主要训练策略及主要局限? A: 直接监督深度学习(模仿人类,可离线但泛化差)、深度强化学习(如DQN学习最优,需在线交互)、神经进化(无需反向传播,需在线)。局限:缺乏硬编码安全与可解释性,真实城市场景实现有限。
  4. Q: 为什么目前ego-only系统比connected更常见? A: ego-only便于开发自足平台;connected虽有共享传感器等潜在益处,但引入VANET安全、路由、缓存等额外巨大挑战,且尚无运营系统。
  5. Q: DARPA Urban Challenge的意义与局限? A: 2007年首次在类城市场景测试,六队完成,吸引大量关注;但测试环境缺乏真实城市的行人与自行车等元素。
  • Abstract (page 1): This paper discusses unsolved problems and surveys the technical aspect of automated driving. Studies regarding present challenges, high-level system architectures, emerging methodologies and core functions including localization, mapping, perception, planning, and human machine interfaces, were thoroughly reviewed. Furthermore, many state-of-the-art algorithms were implemented and compared on our own platform in a real-world driving setting. The paper concludes with an overview of available datasets and tools for ADS development.
  • Section I Introduction (page 1-2): According to a recent technical report by the National Highway Traffic Safety Administration (NHTSA), 94% of road accidents are caused by human errors [1]. … If widespread deployment can be realized, annual social benefits of ADSs are projected to reach nearly $800 billion by 2050…
  • Table 1 (page 2): TABLE 1: Comparison of ADS related survey papers … Ours X X X X X X - X X X
  • Section II B Challenges (page 3): Level three is conditional automation; the driver could focus on tasks other than driving during normal operation, however, s/he has to quickly respond to an emergency alert from the vehicle and be ready to take over. … No production vehicle is capable of level four or five driving automation yet.
  • Section III A System Architecture (page 3-4): ADSs are designed either as standalone, ego-only systems [15], [47] or connected multi-agent systems [48]–[50]. Furthermore, these design philosophies are realized with two alternative approaches: modular [15], [47], [51]–[58] or end-to-end driving [59]–[67].
  • Section III 2) Modular systems (page 4): The major disadvantages of modular systems are being prone to error propagation [31] and over-complexity. In the unfortunate Tesla accident, an error in the perception module in the form of a misclassification of a white trailer as sky, propagated down the pipeline until failure…
  • Table 2 (page 4): TABLE 2: Common end-to-end driving approaches … [59]–[63] Direct supervised deep learning Imitates the target data: usually a human driver. Can be trained offline. Poor generalization performance.
  • topic: deployment-inference
  • sources: asta, openalex
  • retrieved_at: 2026-07-20
  • query: Find foundational and recent research papers for the topic «部署与推理» (deployment-inference). Prefer peer-reviewed or widely cited work with clear method contributions. Include open-source code when available. Exclude pure survey spam unless highly cited. Core concepts: deployment, inference optimization, edge, latency, quantization. Search facets: autonomous driving model deployment inference optimization; edge latency optimization perception planner; model quantization real-time autonomous drivin
  • corpus_id: 186206717
  • doi: 10.1109/access.2020.2983149
  • relevance_score: 0.35508911536796595
  • score_total: 50
  • suggested_tier: foundational

(no prose relevance explanation — numeric score only or HTTP source)

(no snippet evidence in candidate pool)

生成:2026-07-21 · 来源条数 2 · 模型 heuristic · 需人工核验数字

围绕「A Survey of Autonomous Driving: Common Practices and Emerging Technologies」的核心问题与动机(待结合全文核验)。

  • 见原文方法章节;以下为基于摘要/摘录的要点提示。
  • Accepted March 22, 2020 Digital Object Identifier 10.1109/ACCESS.2020.2983149 A Survey o…

与相近工作的关系待核验;请对照 related work。

  • 勿仅凭摘要推断未给出的数值指标。
  1. 这篇工作的输入/输出表示是什么?(A Survey of Autonomous Driving: Common Practices and Emerging Technologies)
  2. 训练目标与评测协议各是什么?
  3. 主要失败模式或局限是什么?
flowchart LR
A["输入 / 观测"] --> B["表示 / 编码"]
B --> C["推理 / 解码"]
C --> D["输出 / 动作或检测"]
%% method sketch for: A Survey of Autonomous Driving: Common Practices and Emerging Technologies

方法结构示意(重绘;细节以原论文为准,待 PDF 核验)。

A Survey of Autonomous Driving: Common Practices and Emerging Technologies arch p.3

来源:原论文约 p.3(arch);学习用途摘录。

A Survey of Autonomous Driving: Common Practices and Emerging Technologies table p.2

来源:原论文约 p.2(table);学习用途摘录。

展开英文 Paper Card / AI deep analysis
Field Content
Year 2020
Authors Ekim Yurtsever, Jacob Lambert, Alexander Carballo, K. Takeda
arXiv 1906.05113
DOI 10.1109/access.2020.2983149
Topics ad-perception-tracking, end-to-end-learning
Paper https://arxiv.org/abs/1906.05113
展开 Extract / Selections / Local assets
  • ad-perception-tracking: tier=needs-review score=56 — auto refresh 2026-07-18 sources=openalex
  • autonomous-driving: tier=foundational rank=2 score=61 — Survey of autonomous driving common practices and emerging technologies
  • end-to-end-learning: tier=foundational rank=1 score=52 — auto refresh 2026-07-19 sources=openalex
(no PDF text available; metadata-only card)