多车交互与自动驾驶连环碰撞安全 · OpenTopic 05

一句话定义In One Sentence

这项研究不是再做一个单车避撞算法，而是建立一条从多参与者交互场景挖掘、连锁传播建模、测试场景生成到事故后应急协同的完整证据链，让连环碰撞成为自动驾驶安全开发中的一个独立目标。

This is not another single-vehicle collision-avoidance algorithm. It builds an evidence chain from multi-participant interaction mining, chain-propagation modeling and test-scenario generation to post-crash coordination, making chain collisions an explicit objective in automated-driving safety.

Problem

多车不是单车的简单叠加Multiple cars are not a scaled-up single car

一次前车急刹会通过跟驰和变道把风险传到相邻车道，不能用若干两两交互单独解释。

One hard braking event can propagate across lanes through following and lane changes, which cannot be explained by independent pairwise interactions.

Gap

现有实践仍以单车为中心Current practice is still vehicle-centric

ISO 34502 和 UL 4600 都有入口，但公开研究和产业验证较少把车队级连锁传播当作主线。

ISO 34502 and UL 4600 provide entry points, but public research and industry validation rarely treat fleet-level propagation as a main line.

Output

从场景到应急的证据包From scenarios to emergency evidence

输出场景簇、传播模型、OpenSCENARIO 用例和事故后信息交换建议。

Outputs include scenario clusters, propagation models, OpenSCENARIO cases, and post-crash information-exchange recommendations.

研究问题Research Questions

RQ1

多参与者交互有多常见？How common is multi-participant interaction?

在中国城市和高速场景中，三辆、四辆及以上参与者耦合的片段占多大比例？与单车-目标车的两两交互相比，哪些特征更容易被遗漏？

What proportion of urban and highway scenes involve three, four, or more coupled participants? Which features are missed when compared with ego-target pairwise interaction?

RQ2

连锁风险如何传播？How does chain risk propagate?

前车制动、后车响应、相邻车道变道之间的时间顺序，如何影响 TTC、PET、DRAC 和事故风险扩散？

How do the time ordering and dynamics among lead braking, follower response and adjacent-lane changes affect TTC, PET, DRAC and risk propagation?

RQ3

如何生成可复现的测试场景？How do we generate reproducible test scenarios?

能否把高频、高风险、高耦合的多车场景簇，转成代表性、边界和挑战三类 OpenSCENARIO 用例？

Can high-frequency, high-risk, high-coupling scene clusters be converted into representative, boundary and challenge OpenSCENARIO cases?

RQ4

事故后应交换什么信息？What should be exchanged after a crash?

车辆、远程运营、消防、急救和交通管理之间，需要共享哪些位置、乘员、高压和拖曳状态，才能减少二次碰撞？

What position, occupant, high-voltage and towing state should be shared among vehicles, remote operators, fire, rescue and traffic management to reduce secondary collisions?

数据基础Data Basis

Aerial NDD

DRIVEResearch 航测自然驾驶数据DRIVEResearch aerial naturalistic driving data

当前约 1000 小时以上航测数据、1200 万条以上交通目标轨迹，适合观察多车相互遮挡、变道和跟随关系。

Currently around 1000+ hours of aerial data and 12+ million traffic-object trajectories, suitable for observing multi-vehicle occlusion, lane changes and following relations.

Scenario Mining

从真实轨迹提取场景簇Mine scenario clusters from real trajectories

不只保留单车视角，而是保留目标车、前车、前前车和相邻车道车辆构成的上下文。

Keep not only the ego view, but also the target, lead, front-front and adjacent-lane context.

Open Data

数据获取可联系Contact for data access

用于复现、扩样和提出新假设；不公开发布原始视频或未授权轨迹。

For replication, scaling and new hypotheses; raw video or unauthorized trajectories are not publicly released.

方法与指标Method & Metrics

以 ISO 34502 的功能场景、逻辑场景、具体场景分层为骨架，结合自然驾驶数据中的多参与者交互度和碰撞风险指标。

Use the ISO 34502 functional-logical-concrete hierarchy as the backbone, combined with multi-participant interaction degree and collision-risk metrics from naturalistic data.

指标集合Metric set

交互度、TTC、PET、DRAC、峰值减速度、最小 TTC 时刻，以及事故后状态传播延迟。

Interaction degree, TTC, PET, DRAC, peak deceleration, min-TTC timing, and post-crash state-propagation latency.

工作包与转化路径Work Packages & Translation Paths

编号WP	工作包Work package	产出Output
WP1	多车交互场景挖掘Multi-vehicle interaction mining	冲突片段集、交互度统计、场景簇标签conflict clips, interaction-degree statistics, cluster labels
WP2	连锁传播建模Chain propagation modeling	风险扩散时序、传播网络、关键变量risk-diffusion timeline, propagation network, key variables
WP3	测试场景生成Test-scenario generation	OpenSCENARIO 用例、参数范围和收敛检查OpenSCENARIO cases, parameter ranges, convergence checks
WP4	事故后协同设计Post-crash coordination design	信息交换字段、角色流程、验证场景information-exchange fields, role flows, verification scenes
WP5	评价指标研究Evaluation metric research	多车风险放大评价、测试通过准则multi-vehicle risk-amplification evaluation, pass criteria

成果可以落到测评工具、企业试点、标准建议或论文，也可以只认领其中一个工作包。 Results can enter evaluation tools, enterprise pilots, standardization proposals or papers. You may claim only one work package.

参考文献References

[1] FIA, Safety Charge: How the FIA is leading the way in high voltage racing, 2026-02-04. https://api.fia.com/news/safety-charge-how-fia-leading-way-high-voltage-racing

[2] FIA University, Working for Safety: Medical and Research-Driven Advances in Motor Sport Safety, 2026-04-15. https://api.fia.com/news/fia-university-launches-working-safety-white-paper-medical-and-research-driven-advances-motor

[3] FIA / DMSB Academy, International Medical Training Days at the Nürburgring, 2017-02-21. https://api.fia.com/news/international-medical-training-days-nurburgring

[4] FIA, Breaking the barriers of safety, 2013-05-28. https://api.fia.com/news/breaking-barriers-safety

[5] Formula 1 car testing regulations and load cells, 2024-03-30. https://www.loadcellshop.co.uk/load-cell-uses/how-to-test-your-new-f1-racing-car/

[6] TecPro Barriers: F1 Track Safety, 2026-05-14. https://f1briefing.com/tecpro-barriers-improve-f1-track-safety/

[7] FIA Formula E 2026-2027 Technical Regulations, wheel tether requirements. https://www.fia.com/system/files/documents/2026-2027_season_13_formula_e_techregs_wmsc_10.12.2025_eng_fr_v17.pdf

[8] NASCAR A-post flap safety update, 2025-08-21. https://thesportsrush.com/nascar-news-everything-about-the-a-post-flap-that-nascar-will-debut-at-daytona/

[9] NHTSA, An Automated Vehicle That Cannot Safely Interact With First Responders is a Danger to the General Public, 2026-07-08. https://www.nhtsa.gov/press-releases/av-developers-automated-vehicle-that-cannot-safely-interact-first-responders-danger

[10] SAE AVSC, AVSC Best Practice for First Responder Interactions with Fleet-Managed ADS-DVs, 2024-04-04. https://avsc.sae-itc.org/Global/FileLib/SAE_International/AVSC-I-01-2024_pd_04.02.24_JG_FINAL_4.03.24_bam_clean.pdf

[11] AVL, Euro NCAP 2026: What's Changing and How to Stay Compliant, 2025-10-22. https://www.avl.com/en/blog/euro-ncap-2026-whats-changing-and-how-stay-compliant

[12] MarkLines, NCAP（新车评价规程）的概要及趋势（1），2022-11. https://www.marklines.com/cn/report/rep2396_202211

[12] MarkLines, NCAP (New Car Assessment Program) Overview and Activities (1), 2022-11. https://www.marklines.com/cn/report/rep2396_202211

[13] 上汽奥迪 E7X 完成行业首次十字路口多车随机碰撞挑战，公开汽车媒体报道，2026-04-14. https://www.toutiao.com/article/7628873761420263979/

[13] Audi E7X completes the industry's first multi-car random crossroad impact, public automotive media report, 2026-04-14. https://www.toutiao.com/article/7628873761420263979/

[14] 中国智能网联汽车产业创新联盟，标准预研｜智能网联汽车高度自动驾驶车辆运营管理异常事件处置规范，2026-05-12. https://mp.weixin.qq.com/s/hQiBeE37qpUqKCRHsJZHjQ

[14] China Intelligent and Connected Vehicles Industry Innovation Alliance, Standard pre-research: abnormal-event handling for highly automated vehicle operations, 2026-05-12. https://mp.weixin.qq.com/s/hQiBeE37qpUqKCRHsJZHjQ

[15] ISO 34502:2022, Road vehicles — Test scenarios for automated driving systems — Scenario based safety evaluation framework. https://www.nen.nl/norm/pdf/preview/document/303466/

[16] UL Standards & Engagement, UL 4600 Ed. 3-2023 scope summary, 2023-03-16. https://webstore.ansi.org/standards/ul/ul4600ed2023

[17] Speedcafe, FIA concludes investigation into Grosjean crash, 2021-03-05. https://speedcafe.com/fia-concludes-investigation-into-grosjean-crash/

[18] ISO 21448:2022, Road vehicles — Safety of the intended functionality. https://www.iso.org/standard/77490.html

如何参与How to Contribute

1 · Claim

认领一个工作包Claim a work package

可以只做场景挖掘、传播建模、OpenSCENARIO 生成或事故后协同设计。

Take scenario mining, propagation modeling, OpenSCENARIO generation or post-crash coordination design.

2 · Use the data

使用开放数据Use the open data

基于 DRIVEResearch 航测自然驾驶数据复现实验、扩样并提出新假设。

Reproduce, scale and form new hypotheses on DRIVEResearch aerial naturalistic driving data.

3 · Translate

验证与落地Validate & translate

把方法推进到测评工具、企业试点、标准建议或新的开放话题。

Move the method into evaluation tools, enterprise pilots, standardization proposals or new open topics.

好问题不该被锁在抽屉里Good questions shouldn't stay in a drawer

如果你想认领话题、获取数据或提出新选题，直接联系。

To claim a topic, get data or propose a new one, just reach out.

联系合作Get in touch DRIVEResearch ← AutoZYX 主页← AutoZYX home

多车交互与自动驾驶连环碰撞安全 Multi-Agent Interaction and Chain-Collision Safety for Automated Driving