China's Embodied AI Data Sector Shifts Focus to High-Quality Supply Amid Scarcity
具身智能数据赛道“变天”
At the 2026 World Internet Conference, experts highlight a severe supply-demand gap where effective robot data costs 500-1000 yuan/hour and lacks standardization due to fragmented hardware and OS protocols. Industry analysts note that demand is shifting from raw volume to high-quality, diverse datasets covering failure cases and complex interactions, prompting a market transition from capacity expansion to effective screening.
Source: Chinese robotics — Robot learning research · Read original article ↗
English translation unavailable. Choose the original article to read it.
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
robot hours
500–1,000
View original evidence
乌镇峰会上，工业和信息化部电子第五研究所人工智能中心副主任李帅道出了具身智能数据面临的严峻现实：核心矛盾在于供需差两个数量级，有效数据存量为百万小时，而头部厂商判断需求为亿小时；真机数据成本500-1000元/小时，远高于互联网语料采集成本；大模型有互联网语料可爬，自动驾驶有海量路测，真机数据却没有“家底”。
Open source S10Reported success rate
76.8%
View original evidence
真机路径是行业公认的“黄金标准”，据李帅介绍，真机数据具有小样本效应：仅用约10%的真机数据即可明显提升成功率；全量真机数据的情况下，成功率则可达76.8%，远高于仅用互联网或通用数据的效果。
Open source S12robot cost
200,000–700,000
View original evidence
不过，真机数据的短板在于重资产，单台设备约20万-70万元，且有效产出并不高，其整体成本甚至是仿真合成路径（该路径的主要问题在于仿真到现实的鸿沟，在真实家庭任务中落差甚至在77个百分点）的100倍。
Open source S13
- paper: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
乌镇峰会上，工业和信息化部电子第五研究所人工智能中心副主任李帅道出了具身智能数据面临的严峻现实：核心矛盾在于供需差两个数量级，有效数据存量为百万小时，而头部厂商判断需求为亿小时；真机数据成本500-1000元/小时，远高于互联网语料采集成本；大模型有互联网语料可爬，自动驾驶有海量路测，真机数据却没有“家底”。
Open source S10
真机路径是行业公认的“黄金标准”，据李帅介绍，真机数据具有小样本效应：仅用约10%的真机数据即可明显提升成功率；全量真机数据的情况下，成功率则可达76.8%，远高于仅用互联网或通用数据的效果。
Open source S12
不过，真机数据的短板在于重资产，单台设备约20万-70万元，且有效产出并不高，其整体成本甚至是仿真合成路径（该路径的主要问题在于仿真到现实的鸿沟，在真实家庭任务中落差甚至在77个百分点）的100倍。
Open source S13
Source：Chinese robotics — Robot learning research · finance.sina.com.cn