Source-linked AI summary
Milestones in Autonomous Driving and Intelligent Vehicles: Survey of Surveys
Long Chen, Yuchen Li, Chao Huang, Bai Li, Yang Xing, Daxin Tian, Li Li, Zhongxu Hu, Xiaoxiang Na, Zixuan Li, Siyu Teng, Chen Lv, Jinjun Wang, Dongpu Cao, Nanning Zheng, Fei-Yue Wang
TL;DR
Existing AD and IV surveys are often limited to specific tasks and lack systematic, macroscopic summaries and future directions. This paper conducts a Survey of Surveys covering milestones, datasets, perspectives, ethics, and future research, concluding that it can bridge past and future.
Problem
Existing surveys often focus on specific AD and IV tasks and lack systematic summaries and macroscopic perspectives.
Method
The paper collects and categorizes milestone surveys, analyzes datasets, and synthesizes perspectives, ethics, and future research directions.
Results
The study collects 122 surveys into 18 research-area categories and summarizes AD datasets, perspectives, ethics, and future directions.
Takeaways & Limitations
The SoS offers horizontal and vertical research coverage and is intended to serve as a bridge between past and future in AD and IVs.
Abstract
from arXiv · showhide
Interest in autonomous driving (AD) and intelligent vehicles (IVs) is growing at a rapid pace due to the convenience, safety, and economic benefits. Although a number of surveys have reviewed research achievements in this field, they are still limited in specific tasks, lack of systematic summary and research directions in the future. Here we propose a Survey of Surveys (SoS) for total technologies of AD and IVs that reviews the history, summarizes the milestones, and provides the perspectives, ethics, and future research directions. To our knowledge, this article is the first SoS with milestones in AD and IVs, which constitutes our complete research work together with two other technical surveys. We anticipate that this article will bring novel and diverse insights to researchers and abecedarians, and serve as a bridge between past and future.
I. INTRODUCTION
This introduction frames autonomous driving and intelligent vehicles as rapidly developing fields whose surveys often remain task-specific. The paper responds with a systematic survey of surveys covering milestones, perspectives, ethics, and future directions.
- AD and IVs attract attention from academia and industry because of their potential benefits.
- The paper’s SoS systematically summarizes AD development from an overall perspective and introduces research perspectives, ethics, and future directions.
- Earlier surveys commonly focus on specific tasks, limiting systematic summaries and potentially hindering abecedarians.
- History of Autonomous Driving & Intelligent Vehicles: AD history spans early remotely piloted vehicles, 1980s integration of LiDAR, computer vision, and automated control, and later competition-driven advances.
- History of Autonomous Driving & Intelligent Vehicles: In the 2010s, neural networks and computing platforms supported movement from private roads to urban roads, including autonomous taxi operations.
- History of Autonomous Driving & Intelligent Vehicles: SAE divides AD into six levels from L0 to L5, while legal responsibility in traffic accidents remains questioned.
B. Paper Structure
The paper is organized to combine historical context, survey analysis, dataset information, and perspectives on future research and ethics. Its stated contributions center on an SoS, dataset characterization, and a bridge between past and future.
- Paper Structure: The article contains sections on introduction, overall survey analysis, datasets, perspectives and future, and conclusion.
- Contributions: The authors describe AD research as having entered a bottleneck period and seek to provide diverse insights for breakthroughs.
- Contributions: The authors collect milestone surveys and categorize them into several subsections as an SoS on AD and IVs.
- Contributions: They enumerate AD dataset characteristics and summarize research perspectives, ethics, and future directions.
- Contributions: The study presents itself as Part 1 of a broader research effort intended to bridge past and future in AD and IVs.
II. OVERALL
The overall section organizes the reviewed AD and IV survey literature into research-area categories. It covers technical, system, infrastructure, human-interface, and special-scene topics.
- Overall: The paper selects 122 survey articles and categorizes them across 18 AD research areas.
- Overall: The categories include localization, object detection, scene understanding, tracking, prediction, planning, end-to-end systems, and control.
- Overall: Additional categories cover systems, hardware, software, communication, simulation, interpretability, HMI, and special scenes.
III. DATASETS
The paper surveys autonomous-driving datasets and simulation platforms, covering their sensors, tasks, scenes, and uses in perception, planning, and control research.
- KITTI provides multiple computer-vision tasks on urban roads in Germany, while Cityscapes, BDD100K, and Mapillary Vistas provide segmentation-mask data.
- A*3D expands dataset coverage to dark-night, rainy, and snowy scenes.
- H3D, A2D2, and the Ford Dataset are vehicle-collected datasets published by automobile manufacturers.
- Table III organizes datasets by frame count, installed sensors, and covered tasks so readers can identify data matching their missions.
- Carla, Vissim, PerScan, AirSim, Udacity, and Apollo support experiments on planning and control through simulation environments.
IV. PERSPECTIVES AND FUTURE
The paper identifies future directions across perception, planning, and motion control, emphasizing system-level robustness, efficiency, coordination, and interpretability.
- 1) Perception: Perception research should improve multisensor fusion, 2D-to-3D detection, automated inference, self-supervision, and cooperative perception.
- 1) Perception: Perception results heavily influence downstream planning and motion control, linking upstream sensing improvements to broader driving-system performance.
- 2) Planning: Planning should address imperfect perception, balance solution quality with speed, maintain consistency across planners, and improve learning-based interpretability.
- 3) Control: Motion-control research should handle uncertainty and delay, optimize multiple objectives, support vehicle cooperation, tolerate faults, and operate in real traffic.
4) Testing:
Testing is presented as a pre-production process for locating remaining problems, modifying IV programs, and reducing accident rates on public roads.
- Testing requires vehicles to complete driving tasks with varying difficulties in testing areas or on private roads.
- The process provides a final opportunity to modify IV programs before public-road deployment.
- Future testing research should develop rational evaluation criteria and criteria for virtual simulation testing.
- Researchers should narrow the gap between real and virtual testing scenarios.
- 5) Human Behaviors: Increasing vehicle autonomy may increase the complexity of human-behavior and human-factor issues.
1) Normative Ethics:
The paper reviews normative, environmental, public-health, liability, and privacy ethics associated with intelligent vehicles and their deployment.
- 1) Normative Ethics: Normative ethics concerns unavoidable choices between alternatives that sacrifice human lives in crashes.
- 1) Normative Ethics: Moral Machine research identified preferences for sparing human lives, sparing more lives, and sparing young lives.
- 1) Normative Ethics: Japanese survey results broadly matched US findings, but US participants showed stronger family-related preferences for self-protective IVs than Japanese participants.
- IVs may improve energy efficiency, emissions, and congestion from collisions, but convenience could increase travel demand, VMT, noise, and EMF exposure.
- Liability is difficult to apportion among industry stakeholders, while machine-learning unpredictability and access to personal information create privacy risks.
C. Future Directions:
Future directions emphasize mutually reinforcing human–vehicle intelligence, human-aware collaboration, and richer virtual-scenario data for perception and planning.
- Human intelligence guides machine intelligence, which learns problem-solving strategies from human behavior to improve intelligent-system reliability.
- Human-aware AI systems should augment human labor and incorporate Human in the Loop models for collaboration.
- Parallel simulation can enrich perception data by generating corner cases and diverse weather conditions in virtual scenarios.These scenarios are intended to enhance detection and planning capabilities.
3) From Scenario Engineering to Scenario Intelligence:
Scenario intelligence addresses inconsistent, poorly indexed, and sparsely annotated datasets by standardizing scenario descriptions and rules for broader reuse and adaptation.
- Current scenario datasets use different formats and standards and lack effective indexing.
- Sparse annotation and difficult reuse motivate scenario intelligence to unify scenario description methods and rules.
- Scenario intelligence is presented as a crucial technology for enabling IVs to adapt to varied road conditions and driving environments toward future L5 AD.
V. CONCLUSION
The paper presents a comprehensive Survey of Surveys on autonomous driving and intelligent vehicles, combining historical milestones, systematic organization, dataset guidance, and future-oriented perspectives.
- The authors collect 122 surveys and organize them into 18 research-area categories for analysis.
- The article reviews autonomous-driving development and introduces an SoS focused on milestone research in AD and IVs.
- It summarizes AD datasets to help researchers select suitable data more quickly.
- The SoS presents research perspectives, ethics, future directions, and horizontal and vertical coverage across AD topics.
VI. BIOGRAPHY SECTION
The biography section profiles contributors working across autonomous driving, intelligent vehicles, robotics, artificial intelligence, transportation, control, human–machine systems, and related areas.
- Several contributors specialize in autonomous driving, intelligent vehicles, robotics, artificial intelligence, and computer vision.The listed interests include 3D object detection, deep learning, driver behavior analysis, and end-to-end autonomous driving.
- The biographies include professors, researchers, engineers, lecturers, and doctoral students affiliated with universities, laboratories, and industry.
- The contributors also cover control, trajectory planning, path planning, fault-tolerant control, and vehicle dynamics.
- Other research areas include machine learning, driver behaviors, intelligent multi-agent collaboration, mobile computing, intelligent transportation systems, and vehicular ad hoc networks.