JD.com launches world's largest robot data campaign, targeting 10 million hours of human behavior data

Cao Peng, chairman of JD. com's Group Technology Committee and president of JD Cloud, announced at the forum of the 2026 World Robot Conference that JD. com will launch a data collection campaign of unprecedented global scale, mobilizing more than 100,000 participants. The initiative is designed to provide sufficient data support for training robot and embodied intelligence models, addressing current deficiencies in task understanding and generalization capabilities.

The campaign targets the collection of over 10 million hours of high-quality human behavior data from real-world scenarios, as well as more than 1 million hours of robot operation data, within a two-year period. Participants will include JD. com employees across various functions, external industry professionals, and city residents. The data collection process will strictly adhere to legal and compliance principles, with data security and reliability ensured throughout. JD. com hopes this initiative will establish a sustainable data infrastructure for the embodied intelligence industry.

Cao Peng noted that the industry currently trains models with data measured in the tens of thousands of hours, whereas a robot model capable of scaled replication and generalization requires data on the order of tens of millions of hours, leaving a significant shortfall. In certain specific standard tasks, robot efficiency and cost have already approached human levels, but insufficient data has prevented extension to more complex scenarios. This large-scale data collection effort will be carried out on a global basis to bridge the industry's data gap.