To The Moon

The rise of embodied intelligence is creating a new job market for data collectors, with nearly a million people now working flexibly in this field. These part-time workers are teaching robots how to cook, clean, organize, and perform various household and industrial tasks through real-world demonstrations. As robots require vast amounts of behavioral data to understand and interact with the physical world, this new profession has quickly gained traction across China.

Robots remain in their infancy and require data collectors to serve as their teachers, with the embodied intelligence wave generating this new data collection role. The daily routines of ordinary people—from washing dishes and wiping tables to tidying clothes and working in supermarkets or hotel rooms—are becoming valuable training material for robotic systems. The industry has reached a consensus that training a general-purpose embodied intelligence robot requires 10 billion hours of real-world interactive data, yet currently only 500,000 hours of compliant, usable physical interaction data exist within China.

Where to begin Wang Xiaohua, a 62-year-old retiree from Baicheng, Jilin Province, stumbled upon this opportunity through a friend's sharing. After registering locally, she received a lightweight collection headband and began turning her daily chores into valuable data collection sessions. Initially nervous about whether her movements would meet standards, she eventually learned through online training that the job actually seeks the most natural, authentic human behavior rather than staged or artificial actions.

The workflow is straightforward: Wang registers on the platform, accepts daily home-scene tasks online, and the device automatically uploads collected data for quality inspection. By the next day, she can see her effective working hours clearly displayed, and at month's end, her salary arrives directly. She jokingly refers to this side gig as earning money while doing housework.

Similarly, a data collection community in Jiangsu Province has attracted numerous ordinary residents who have become AI's household chore instructors. Li Jie, a mother of two, eagerly signed up when the community launched its pilot program in April. As a stay-at-home mom, traditional nine-to-five jobs didn't fit her lifestyle, making this remote work opportunity a perfect match. After brief professional training, she now collects roughly six hours of data on weekdays, earning over 4,000 yuan per month while maintaining her weekends free.

The scale of opportunity This year is considered the first year of large-scale embodied intelligence data collection, with data collection centers springing up across regions and data companies rapidly emerging. The market's surging demand is rippling through the employment landscape, with job postings on social media and recruitment platforms showing a proliferation of data collection positions, most of which are flexible or part-time in nature, spanning homes, retail stores, factory floors, and hotel rooms.

Wang Xiaohua's household motion data eventually flows to Beijing-based embodied intelligence data company Jiyuan Zhihang. According to Chairman and CEO Gao Shaolong, the industry's development focus has shifted from basic movements like running and jumping toward understanding and perceiving the real world. He notes that specialized full-time collectors alone cannot meet the enormous demand, making part-time workers the backbone of this industry for years to come. Current robots resemble kindergarten children who need exposure to diverse real-world scenarios to build complete cognition and enhance their universal adaptability.

Why just nearly one million part-time collectors? The rapid market expansion has swelled the ranks of part-time data collectors to nearly one million, according to Gao. These workers are scattered across third and fourth-tier cities and townships in regions including the northeast, Xinjiang, Guizhou, Anhui, and beyond. Companies typically pay 70 yuan per hour of effective collection time, with some areas offering additional subsidies, making this a notably attractive income opportunity. During off-seasons for local hotels, factories, or orchards, residents can supplement their income through data collection, adding another dimension to this employment model.

However, the reality of this new profession often differs from the rosy picture painted by flexible employment and work-from-home promises. Investigative reporting reveals that equipment quality varies significantly across companies, and some workers complain that wearing heavier devices and mechanically repeating identical movements can be more exhausting than factory assembly line work. The industry lacks unified compensation standards, creating notable income disparities across different channels.

Young worker Xiaoyuan, who started collecting data in June, received not only a somewhat heavy head-mounted camera but also a mechanical gripper designed to simulate robotic hand movements. His daily three-hour sessions involve performing various hand gestures, which initially proved physically demanding. The head camera restricts excessive head movement, and maintaining extended arm suspension while repeating the same actions dozens or even hundreds of times leaves his limbs sore and numb. After mastering the basics, new challenges emerged—designing multiple coherent actions within task sequences rather than simply grasping and storing items. He admits the work proved more demanding than expected.

The income gap Some workers openly express disappointment with their compensation. While direct recruitment through human resources firms can offer daily wages of 200 to 300 yuan for five to eight hours of work, those who obtain assignments through social media intermediaries often face drastically reduced pay. Xiao Pan started working as a data collector through a social media intermediary in July with daily payouts of just 15 yuan per hour. To complete her tasks, she frequently works ten-hour days, yet after insurance deductions, her take-home pay amounts to only 147 yuan per day.

Gao Shaolong attributes these disparities to the industry's early-stage development and lack of unified standards. Different companies pursue different model training objectives, requiring different data formats and collection methodologies. No dominant industry leader has yet emerged, and most data processing companies were founded only within the past two years, resulting in considerable variations across hardware and collection procedures. However, the industry is gradually optimizing on both software and hardware fronts, with multiple companies developing lighter, more ergonomic collection products.

Jiyuan Zhihang, for instance, has developed lightweight headband devices weighing just over 100 grams to reduce the burden on workers' heads. Recognizing that video data is currently core to embodied intelligence training, the company's new headbands incorporate six cameras for 270-degree hand movement capture, eliminating the need for collectors to contort their bodies to record hand actions. This allows for more relaxed, natural movements while reducing physical fatigue during extended sessions. These lightweight headbands have already attracted significant interest from manpower service providers, with orders surpassing 10,000 units.

The path ahead Industry observers anticipate data collection will eventually expand into more specialized sectors. While robots remain in their cognitive development stage, they will eventually progress into professional skill learning. At that point, skilled workers such as auto mechanics, factory technicians, and agricultural specialists could become vital new contributors to the data collection workforce. Companies are already planning platform tools to accommodate these future professional participants.

Despite the challenges, industry leaders remain optimistic about the future of data collection as a profession. The path forward requires continued software and hardware iteration to ease physical demands, establishment of industry-wide standards to clarify responsibilities, and robust protections for flexible workers' rights. Only then can this emerging profession simultaneously support the growth of embodied intelligence while genuinely delivering on its employment promise.