JD.com Sets Embodied Intelligence Strategy, Aiming for Large-Scale Deployment in Commercial Scenarios Within 3-5 Years

“Embodied intelligence is one of JD Group's most important strategic priorities,” said Cao Peng, chairman of the JD Group Technology Committee, on November 4 at the 2026 World Internet Conference Wuzhen Summit.

Embodied intelligence is a hot topic at this year's World Internet Conference Wuzhen Summit. Today JD disclosed its progress in embodied intelligence data collection: JD's embodied data collection has covered more than 100 segmented production and service scenarios, collecting over 100,000 real work tasks. Within two years, the goal is to collect more than 10 million hours of high-quality real-world human scenario video data, and simultaneously collect more than 1 million hours of robot body operation data.

At present, JD's plans in the field of embodied intelligence mainly cover three directions: embodied data collection and data infrastructure construction, R&D and application deployment of equipment for real scenarios, and cutting-edge embodied large model R&D.

Cao Peng told a Yicai reporter that this year JD's embodied intelligence team is mainly focused on the overall layout and scale accumulation at the data level, the iteration and evolution of embodied models, and applications in real scenarios. The current challenge lies in the iteration and convergence of the entire technology scenario itself. Because the technology route has not yet converged, model vendors' demand for data keeps changing, and progress is not being made on a stable form, so there is considerable uncertainty.

Cao Peng also explained the necessity of developing embodied intelligence. He said traditional industrial robots are suitable for closed, fixed scenarios (such as automobile manufacturing), lack generality, and are hard-pressed to handle non-standardized scenarios such as logistics. The advantage of embodied intelligence is stronger motion capability and scenario generalization, which can solve the sorting challenges of non-standard items (such as clothing and general merchandise), improve stability and reduce costs.

In terms of large-scale deployment, robot deployment can be divided into commercial scenarios (B-end) and home scenarios (C-end). In Cao Peng's view, robots are expected to achieve large-scale deployment in commercial scenarios in the next 3-5 years, while it will take longer for robots to enter home scenarios on a large scale.

For example, warehouses, on the premise of high efficiency, can build specialized production processes for robots. In home scenarios, however, households usually will not remodel their home environment for robots. Limited by complex environments, high costs and safety and ethical issues, widespread adoption will come only after B-end technology matures, commercial scenarios are validated and costs drop significantly.

In terms of consumer scale, Cao Peng believes that for robots to enter ordinary households and reach the level of penetration of mainstream home appliances, it will still take a relatively long time, possibly five to ten years or even longer, but this time estimate carries considerable uncertainty.

After robots enter work scenarios in large numbers, companies will face issues of labor force transformation and employment.

Cao Peng said in this regard that JD is carrying out robot industry planning, providing maintenance and after-sales service for the robot industry; these are all new employment opportunities and forms generated in the process of industrial iteration. JD also plans to expand robot industrial parks to provide new employment opportunities such as maintenance and after-sales service, achieving skill upgrading and industrial iteration.

In addition, the threshold for robot maintenance will decrease with technological standardization and modular design, similar to car maintenance; in the future, a large number of trained personnel will be needed to do this work.