Robotics Training Data Gap and Emerging Data Collection Infrastructure
机器人学会干活需要多少数据燃料 训练数据缺口超99%
As of early 2026, global robot training data amounts to about 500,000 hours, far below the 10 million hours needed for practical embodied intelligence. Data collection is expensive and labor-intensive, with new dedicated hardware and infrastructure emerging to address these challenges. The article details recent data sets, training facilities, and policy support for the field.
The article highlights the significant gap in real-world data for training embodied intelligent robots, emphasizing the need for extensive data collection and the challenges of cost and quality. It outlines the current state of data collection efforts, including new hardware and infrastructure developments, and notes the growing importance of data in the industry.
Source: Chinese robotics — Dataset and collection discovery · 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
training_data_hours
≈500,000 hours
training_data_hours
10,000,000 hours
- dataset: not_reported
What remains unknown
Dataset licensing and commercial-use rights are not established here.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
全球能用的机器人训练数据，截至2026年初，只有约50万小时。
Open source S2
要训练一个实用的具身智能机器人，至少需要1000万小时的数据，缺口超过99%。
Open source S3
此外，2024年8月，鹏城实验室发布了ARIO数据集，包含300万条数据，结合了真机和仿真数据。
Open source S10
2025年1月，上海张江投用了全国首个异构人形机器人训练场，同年3月，北京石景山也投入使用了一个训练中心，一期就达到了百万条数据级别。
Open source S11
上海张江国地中心训练场超过5000平方米，拥有100多台不同构型的机器人，累计形成超100万条、2.5PB的高质量真机数据。
Open source S24
Implications for data suppliers
RoboSignal interpretation and collection questions, not statements of buyer demand.
- Confirm the required data type and collection setting with the buyer; this source does not establish a complete collection specification.
- Compare the reported units and scope before using these quantities in a budget. Recording hours, sensor-hours and trajectories are different measures.
- Validate demand and acceptance criteria with a buyer before scaling. Publication, popularity and a research result do not establish a purchase commitment.
Source：Chinese robotics — Dataset and collection discovery · 3g.china.com