In the sweltering heat of a garment factory in Suqian, Jiangsu province, workers are stitching more than just fabric. Some wear sleek black headgear, a proprietary wearable device designed by e-commerce giant JD.com to record the minutiae of physical labor. These workers are the vanguard of a massive human-in-the-loop operation aimed at bridging the gap between digital intelligence and the physical world, a field known as embodied AI.
While Large Language Models like ChatGPT revolutionized the digital realm, the robotics industry faces a critical 'data drought.' Training a robot to perform household chores or warehouse logistics requires millions of hours of first-person video, haptic feedback, and spatial data. JD.com has launched an ambitious initiative to mobilize 100,000 employees and up to 500,000 external contributors to collect 10 million hours of real-world interaction data within two years.
In Suqian, JD's hometown and a major logistics hub, the recruitment drive has reached deep into the community. Stay-at-home mothers and local residents are paid between 3,000 and 3,500 RMB ($415 to $485) per month to record six hours of daily chores. Tasks range from folding laundry and cleaning tables to more complex industrial actions like precision sewing and fruit picking, providing the 'data fuel' necessary for AI to understand human intent and physical constraints.
The technical objective is to train Vision-Language-Action (VLA) models. These models allow robots to generalize knowledge; by learning the hand-eye coordination of task A, the AI may eventually perform task B without specific prior training. JD’s self-developed JoyEgoCam captures upper-limb trajectories and force distribution, translating human motion into algorithmic building blocks for the next generation of service and industrial robots.
Industry experts note that while the robotics hardware is maturing, the 'brain' remains the bottleneck. Current industry datasets hover around 500,000 hours, a fraction of what is needed for true commercial utility. JD’s strategy leverages China's vast labor pool and its unique 'company town' ecosystem in Suqian to create a data-gathering machine that few international competitors can replicate at such a low cost.
As the race for humanoid robotics intensifies, the battleground has shifted from lab-grown simulations to the messy, unpredictable reality of everyday life. By turning domestic chores and factory shifts into training sets, JD.com is betting that the path to the 'General Artificial Intelligence' of the physical world starts in the living rooms and workshops of provincial China.