At the recent 2026 World Artificial Intelligence Conference (WAIC), YUNJI unveiled its "World Value Model" and showcased products including an embodied intelligent single-arm collaborative robot and the UP200 industrial chassis.
According to YUNJI's definition, the World Value Model serves as an intelligent brain for human-machine collaboration in physical spaces. It goes beyond controlling a single robot to perform a specific action; it coordinates people, robots, equipment, and business systems within real-world environments, dynamically allocating resources based on task priority, user experience, and operational benefits.
This represents a layer of "upper-level decision-making system" that YUNJI is attempting to add to service robots.
Over the past decade, the service robot industry has primarily focused on enabling robots to operate stably in real-world settings. Capabilities such as positioning and navigation, obstacle avoidance, elevator use, automatic charging, fault recovery, and multi-robot operation determined whether robots could work long-term in hotel corridors, hospital floors, and office buildings.
However, as the number of robots increases and the connected devices and tasks become more complex, simply solving the problem of "whether a single machine can complete a task" is no longer sufficient.
Taking a hotel as an example, multiple demands such as delivering water, meals, laundry, and retrieving items may arise simultaneously. The number of robots is limited, and resources like elevators, passageways, and laundry equipment may be occupied. The system needs to not only know how a robot gets from a guest room to the laundry room but also judge which task is more urgent, which robot should be dispatched, and whether staff intervention is required if a task encounters an exception.
The World Value Model proposed by YUNJI primarily aims to address this layer of complexity.
Conceptually, it is not the same type of model as the frequently discussed "world model" in the industry.
World models typically focus on understanding and predicting the physical world. By learning the relationships between environmental states, actions, and outcomes, the model predicts changes in the world following an action and uses this to guide the robot.
In contrast, YUNJI's World Value Model does not primarily aim to simulate the physical world. It is closer to a value-based decision-making layer built atop world models, robot control systems, and business systems.
As the company's product vice president explained, they are optimizing across a longer task chain and a larger service scale. While world models employ a predictive approach to provide capabilities and guidance for the robot's execution layer, the focus of YUNJI's newly released model is "to optimize the entire task at a higher level."
In other words, the World Value Model seeks to answer the question: "Under current conditions, which action is more worthwhile?"
To achieve this, YUNJI has structured the World Value Model into a four-tier architecture from T1 to T4, corresponding to scene understanding, task planning, resource scheduling, and atomic execution, respectively.
T1 is responsible for understanding user needs and on-site status. T2 breaks down a service goal into a sequential task flow. T3 decides whether, when, and by whom a task should be executed. T4 then calls upon robots, robotic arms, elevators, access control systems, and business systems to perform specific actions and feeds the results back to the system.
Using hotel laundry as an example, when a guest makes a laundry request, the system first needs to identify garment information, pickup/delivery times, and special requirements, while also obtaining the real-time status of delivery robots, robotic arms, washing machines, elevators, and passageways. Subsequently, the task is decomposed into multiple steps: picking up the clothes, transporting them, handing them over, loading, washing, unloading, and returning.
During execution, the system must also decide whether to process this request immediately and which delivery robot and laundry equipment to dispatch. If elevators are congested, equipment fails, or a new urgent task emerges, the system needs to re-prioritize the sequence or delegate certain steps to human staff.
This differs from traditional robots following fixed rules to complete a single delivery task. It must handle not just a chain of actions but also conflicts between tasks, real-time changes in resource availability, and the division of labor between humans and machines.
Behind this model lies the operational data accumulated by YUNJI over time from numerous scenarios like hotels.
YUNJI's products have been deployed in over 40,000 hotels globally and approximately 200 hospitals, amassing 1.1 billion service interaction data points. Unlike internet model data, this information encompasses not just user clicks or text dialogues but also real operational details such as robot tasks, spatial status, device calls, exception handling, and manual interventions.
However, the ultimate test for the World Value Model's viability is whether it can consistently make more economical and demand-aligned decisions across diverse scenarios.
Comments