JD.com details progress in embodied AI data collection, covering over 100 scenarios with 100k+ tasks

During the Robotics and Embodied AI Forum at the 2026 World Internet Conference in Wuzhen, Cao Peng, Chairman of the Technology Committee at JD.com, presented on the company's advancement in moving AI from the digital to the physical realm. JD.com's embodied data collection currently spans over 100 distinct production service scenarios and has gathered more than 100,000 real-world operational tasks. Cao Peng emphasized that competition in physical AI is a systemic capability contest, not just a point competition between models and robot hardware. To scale data resources, JD.com is building an embodied AI data collection center, aiming to accumulate over 10 million hours of high-quality human real-world video data and over 1 million hours of robot operational data within two years. The company utilizes its extensive ecosystem, including over 3,000 application scenarios across retail, logistics, and health, and possesses 3,800 warehouses and over 200,000 offline retail pharmacies. Technologically, JD Cloud has developed a comprehensive embodied data service system, including the JoyEgoCam collection terminal and the JoyAI-Sim physics simulation engine, which achieves about 90% consistency between simulation and real-world testing, with 3.2 times the evaluation efficiency. Furthermore, JD.com's logistics division plans to procure 3 million robots, 1 million autonomous vehicles, and 100,000 drones over the next five years, with its Zhiyuan embodied robots already operating in JD Logistics' Zhilang warehouse.