AutoZYX
Open Research Agenda · AutoZYX
研究思路从来不缺，缺的是把每一个都做深、做实、做进产品里的精力。课题组的人手是有限的，但好问题不该因此被锁在抽屉里。所以——像开源数据集一样，把完整的研究思路也开源出去：每个开放话题都是一份可以直接上手的问题规划，背后接着可用的 DRIVEResearch 航测数据，让研究人员、企业工程团队和测评机构都能接着往前走，把问题做深，并把方法落到真实系统里。
Research ideas are never the bottleneck — capacity to pursue each one deeply and deploy it in real systems is. A small group can only chase so many; good questions shouldn't be locked in a drawer because of that. So, just as we open-source datasets, we open-source the research agenda itself: each open topic is a ready-to-use problem plan, backed by usable DRIVEResearch aerial data, for researchers, engineering teams and evaluation labs to deepen, validate and translate into products.
数据集可以开放，研究思路也应该开放。真正稀缺的不是“题目”，而是“把题目做深的人”。OpenTopic 把题目、数据基础、方法、指标和工作包一起公开，把一个想法从“只有我能做”变成“谁都能接着做”——这本身就是一种复利。
Datasets can be open; research ideas should be too. What's scarce is not the question but people to take it deep. OpenTopic makes the question, the data basis, the method, the metrics and the work breakdown public together — turning an idea from “only I can do this” into “anyone can carry it on.”
每个话题给出清晰的研究问题、为什么值得做、以及它在国内外现状中的空白位置——拿到就能判断要不要做、从哪切入。
Each topic states the research questions, why they matter, and where the white space sits in the literature — enough to decide whether and where to start.
方法、指标定义、收敛标准与数据来源都写清楚；已有线索用真实样本算出、可被重算，而不是漂亮的占位数字。
Methods, metric definitions, convergence rules and data sources are spelled out; pilot clues are computed from real samples and can be recomputed — not pretty placeholders.
可以整体认领，也可以只做一个工作包、一个场景或一个速度段；欢迎补充新假设、复现实验、工程试点和标准化建议。
Claim a whole topic or just one work package, scenario or speed bin; new hypotheses, replications, engineering pilots and standardization inputs are all welcome.
每个话题按同一套结构展开，确保“拿到就能做”。
Every topic follows the same structure, so it's actionable on arrival.
研究问题、为什么值得做、空白定位。Research questions, why it matters, the gap.
可用的航测数据、样本规模、已有线索。Usable aerial data, sample scale, pilot clues.
提取方法、指标定义、分层与收敛标准。Extraction, metric definitions, stratification & convergence.
可分阶段认领的工作包、测评工具、产品落地与标准化线索。Claimable work packages, evaluation tools, product paths and standardization leads.
目录会持续增加。当前三个话题已完整展开并开放认领，其余为提案中——也欢迎你提出新话题。
The catalog grows over time. Three topics are fully developed and open now; the rest are proposed — and new topic proposals are welcome.
加塞、前车制动、前车切出——商用车与乘用车的峰值制动减速度（最小加速度）分布、TTC 变化规律、最小 TTC 的值与时刻是否不同？高驾驶视野带来的预判优势能否量化，并转化为参数定义、系统设计与测试用例？含真实加塞样本的已有线索与完整方法。
Cut-in, lead-vehicle deceleration, cut-out — do peak-braking distributions, TTC trajectories and min-TTC value/timing differ? Can the high-eye-height anticipation advantage be quantified and turned into parameters, design and test cases? Includes real cut-in pilot evidence and full method.
查看完整研究规划 →View the full plan →
基于 2026 年国家标准项目《机器人预期功能安全实施指南》公示信号，系统梳理机器人 SOTIF 的标准地图、物理交互安全、LLM/VLA 决策安全与移动物理 AI 跨本体测评证据链。
Triggered by the 2026 national standard project notice for robot SOTIF, this topic maps standards, physical interaction safety, LLM/VLA decision safety and cross-embodiment evidence for mobile physical AI evaluation.
查看完整开放专题 →View the full open topic →
从赛车多车碰撞和无人驾驶事故现场交互出发，把多车连锁碰撞从单车碰撞的延伸，升级为可计算、可测试、可验收的生命周期目标。基于航测自然驾驶数据，研究多参与者交互度、风险传播、OpenSCENARIO 场景生成和事故后应急协同。
Starting from motorsport multi-car crashes and driverless-vehicle emergency-scene interactions, this topic turns multi-vehicle chain collisions into a computable, testable and verifiable lifecycle objective. Using aerial naturalistic driving data, it studies multi-participant interaction, risk propagation, OpenSCENARIO generation and post-crash coordination.
查看完整研究规划 →View the full plan →
用驾驶员视野模型 + 多指标碰撞时间图，从航测数据中自动提取以自车为中心的多车交互场景簇，并统计哪些多车分布模式在中国交通中真正常见且高风险。
Use a driver-vision model plus a multi-metric collision-time graph to auto-extract ego-centred clusters from aerial data, and rank which multi-vehicle layouts are common and high-risk in Chinese traffic.
把 P5–P95 分布、收敛检查与场景簇标签，连成代表性 / 边界 / 挑战三类可复现的 OpenSCENARIO 测试用例生成链路。
Connect P5–P95 distributions, convergence checks and cluster labels into a reproducible pipeline generating representative / boundary / challenge OpenSCENARIO cases.
选一个完整话题，或只做其中一个工作包、一个场景、一个速度段——粒度由你决定。
Take a whole topic, or just one work package, scenario or speed bin — your call on granularity.
基于 DRIVEResearch 航测自然驾驶数据复现实验、扩样、提出新假设；数据获取可联系。
Reproduce, scale up and form new hypotheses on DRIVEResearch aerial NDD; contact for data access.
引用本话题页，也欢迎把方法推进到测评工具、产品验证、标准提案或新的开放话题。
Cite the topic page, and help move the method into evaluation tools, product validation, standardization proposals or new open topics.
把研究思路开源，是为了让更多人拿着真实问题往前走。如果你想认领话题、获取数据或提出新选题，直接联系。
Open-sourcing research ideas is about letting more people move forward with real problems. To claim a topic, get data, or propose a new one, just reach out.