DEVOURED
China is training a robot future — one folded shirt at a time
China is gaining a scaling edge in robotics development by mobilizing large local workforces to collect massive, low-cost training data sets from real homes and factories.
What: Chinese tech companies like JD.com are rapidly collecting millions of hours of human movement data in homes and factories, paying residents and workers in cities like Suqian and Guangdong to film themselves performing chores and tasks, a strategy enabled by low labor costs and government support.
Why it matters: This localized, high-volume data collection model contrasts with the U.S. approach of outsourcing, potentially giving China an advantage in building robots better adapted to diverse real-world environments, accelerating physical AI development.
Deep dive
- Robotics development globally is bottlenecked by a shortage of complex visual and movement training data.
- Traditional teleoperation for robot training is costly and limited in preparing robots for diverse real-world environments.
- Chinese companies are using creative, low-cost methods to collect data in realistic settings, including egocentric (first-person) videos.
- JD.com aims to generate 10 million hours of data over two years by recruiting 100,000 employees and 500,000 external workers in a "data collection neighborhood" in Suqian.
- Workers in elderly care centers and kiwifruit farms are wearing head-mounted cameras.
- Data vendors in Guangdong are working with electronics and packaging factories to collect assembly-line data using head cameras and wrist sensors.
- This approach provides new job opportunities for people like Gao Bo, a stay-at-home mom earning $3/hour by filming chores.
- Analysts like Marco Wang from Interact Analysis believe China leads in hardware and data ecosystems for robotics, while the U.S. leads in AI talent and model research.
- Professor Alan Fern of Oregon State University notes that while the scaling logic from large language models is applied, it's still unproven whether this data will lead to truly intelligent robots capable of functioning in arbitrary environments.
Decoder
- Egocentric data: First-person video data, typically recorded from head-mounted cameras, showing a user's perspective and hand movements while performing tasks.
- Teleoperation: The control of a robot or machine from a distance by a human operator.
Original article
China's low labor costs, government support, and public enthusiasm for robotics development are helping the industry mobilize large populations to work in robotics data collection. Robotics developers globally have ramped up data collection in real households since the beginning of the year. US companies face high labor costs and are outsourcing their data collection to workers in developing countries. Chinese companies are able to collect massive data sets locally, helping them build robots better adapted to domestic environments.
Explore all 55 articles in today's digest
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi