Robot Data Collection Reaches Ordinary People

In Suqian, residents can wear JoyEgoCam to record their movements while folding blankets, wiping tables and preparing vegetables; the embodied-intelligence data-collection community jointly built by JD.com and Suqian began operating in May this year. Meanwhile, “Mifeng Pai,” an app owned by Mifeng Technology, has already launched. Once it formally opens to the public, users will be able to rent collection equipment, claim tasks, and receive cash payments after uploading videos and data that pass review.

Scale is the question this approach must answer. In a group interview on August 31, Yao Maoqing said that embodied intelligence’s data needs in the next phase could reach “tens or even hundreds of millions of hours,” acknowledging that it would be “extremely difficult” to replicate the existing model linearly. Traditional data-collection factories bring robots, teleoperation equipment, operators and simulated environments together to repeatedly perform tasks such as grasping, carrying and sorting. If demand suddenly expands, companies cannot rely solely on adding factories, robots and teleoperators at the same pace.

Figure in the United States is testing another organizational model. Its humanoid-robot data project, Index, had previously operated in secret for four months. Ordinary people can perform tasks such as cooking, cleaning, doing laundry and arranging shelves at home or in the workplace, then upload the data. As of August 25, Index had covered 108 countries, with more than 44,000 weekly active participants uploading over 16 million videos. Figure has paid data contributors $15 million and plans to invest more than $1 billion in data and computing over the next 12 months.

China’s approach is splitting into several paths: Mifeng is closer to platform-based crowdsourcing, JD.com is more focused on combining enterprise settings with community mobilization, while Huishui in Guizhou has added coordination by local government and regional resources. JD.com plans to mobilize more than 100,000 residents in Suqian alone, covering more than 100 specific settings across homes, offices, factories, logistics, stores and sanitation. Its broader goal is to mobilize more than 100,000 internal employees and 500,000 people from various industries externally, accumulating more than 10 million hours of real-world data within two years. Local authorities have disclosed that existing collection efforts cover eight major industries and more than 50 real-world settings, delivering more than 10,000 hours of high-quality data annually.

For ordinary people, what is currently on offer is cash paid for data that passes review, rather than a stable wage system. Projects in the industry have offered rates such as “15 yuan an hour” or “120 yuan a day,” but Yao Maoqing stressed that there is no standard pricing across different settings. Scarcity, difficulty of access and delivery deadlines can all change the price. For companies, the value of dismantling the factory model lies in bringing equipment into real work and daily life, rather than simply adding more data collectors; society itself could become a much larger data-production site.

The real barrier is therefore shifting from recruiting people to securing access to the physical world. Homes involve privacy, while companies involve commercial secrets; hospitals, factories, warehouses and elderly-care institutions also raise questions about the rights of third parties. A mechanic’s ability to repair cars does not mean the worker can wear a device into a workshop and upload the entire process. Where individual consent serves as the basis for processing personal information, China’s Personal Information Protection Law requires voluntary and explicit consent given with full knowledge of the circumstances; the Data Security Law requires data collection to use lawful and proper methods. The chain in which skills belong to workers, settings belong to companies, equipment is supplied by data firms, platforms process the data and robot companies buy it still has to answer who owns the data and how the people contributing the movements should share in its value. “Working for robots” is therefore better understood as a production model still taking shape than as an established profession.