When algorithms begin to understand the texture of shelves and the rhythm of breathing, the coordinates of technology quietly shift from the virtual world. No longer fixated on the stacking of parameters and the noise of traffic, the true winner is shifting towards mastering the rules of physics. The real scenes deeply rooted in the capillaries of the supply chain are far more capable of nurturing intelligent workers than pure network data.

If AI breaks away from screens, understands emotions, recognizes environments, and controls robots to serve warehouses, factories, and homes, it will bring about a whole new revolution. At the 2026 World Artificial Intelligence Conference, JD.com fully showcased the JoyAI model matrix, open-source industry's top first perspective dataset, and launched the JoyInside whole house AI model room AI Home. Viewers can wear JoyEgoCam to experience real-life collection. Relying on models, data, and real-life scenarios, JD.com promotes AI intervention in reality perception, decision-making, and practical operations.

The competition in the AI industry is shifting from digital to physical space in the second half. Traditional big models are good at generating graphics and text, while the new generation of AI also needs to understand space, memory tasks, and coordinate devices to solve practical problems in industries and people's livelihoods. JD does not follow the route of piling up parameters and grabbing traffic, relying on years of supply chain, real-life scenarios, and industry data to iterate AI value in real scenarios. JD's output is not just cloud, model, and hardware, but a complete set of physical efficiency enhancing AI solutions.

Multi class models empower industries

A voice command, smart devices can recognize the speaker and provide adaptive feedback based on the environment and habits. The conference released two major models, JoyAI voice and video editing, with emotion recognition, low latency interaction, and real-time editing capabilities. By 2026, JD.com has launched 7 basic models, forming a complete system of voice, visual, and embodied intelligence. The model empowers various industries with intelligent logistics scheduling to quickly plan distribution routes; Upgrade multiple AI diagnostic products in the medical field; Industrial online intelligent procurement manager, forming the AI technology foundation of service entities. JD continues to open up its model capabilities to assist in the intelligent upgrading of various industries.

Open source real-world training data

The model is the brain of the robot, and real-world data is its growth experience. Relying solely on online graphics and text is not enough to support robots in home and factory operations. The high cost and poor universality of real-time data collection and annotation are bottlenecks in the development of embodied intelligence. JD.com collaborates with 600000 people to collect data, establishes a large-scale data collection center, accumulates millions of hours of human operation materials in two years, and builds a data trading platform. This open source EgoLive first perspective dataset complements real-life learning materials for robots. The exhibition showcases the complete learning chain of robots, with real people collecting images and annotating them to construct digital twin simulation training. Finally, the physical cargo handling robot completes its tasks. JD builds a full chain data infrastructure for procurement, standardization, training, and testing, reducing the threshold for obtaining industry data.

Whole house intelligence is close to life

AI is finally landing in daily homes. JD AI Home model room covers the entire house space: AI desk lamp for answering and correcting questions, tea bar machine for active water supply, kitchen appliances to match home tastes, mattress monitoring for sleep data. Whole house appliances and robots are interconnected and provide on-demand collaborative services. The home environment is noisy and has multiple instructions crossing. JoyInside implants independent AI into the hardware to distinguish interaction objects and has long-term memory and active service capabilities. At present, we cooperate with nearly 200 brands, and by 2026, we will have access to over 10 million hardware units. Keywords: JoyAI, embodied intelligence, dataset

The model, data, terminal, and cloud base form the closed loop of JD AI landing. JD.com has thousands of application scenarios in retail, logistics, healthcare, industry, and more, with drones and humanoid robots in storage appearing simultaneously at the exhibition. The industry focuses more on AI content generation, while JD.com focuses on AI entity implementation applications. JD will build a world-class intelligent operation center for the physical world, allowing AI to deeply participate in the operation of real production and life.Editor/Gong Ziwei