On May 23, the Embodied Interactive Intelligence Forum of the 2026 Global Artificial Intelligence Technology Conference (GAITC 2026) was held in Hangzhou.

Hosted by the Chinese Association for Artificial Intelligence, with support from Beijing Yunji Technology Co., Ltd., the School of Artificial Intelligence at Beijing University of Technology, and the Beijing Key Laboratory of Embodied Interactive Intelligence, the forum focused on key topics including embodied AI, edge intelligence, 3D active perception, multi-agent collaboration, and human-machine symbiosis. Experts, researchers, and industry representatives gathered to explore the technical paths and industrial value of bringing robots into real-world scenarios.

As a vision perception system provider for robotics and Physical AI scenarios, Union Image was invited to join the roundtable discussion with industry representatives from Yunji Technology, Yidao, Jiage Tiandi, Black Sesame Technologies, and other companies. The discussion centered on: From World Models to Value Models: Envisioning the Robot World in the Era of Proactive Service.

Robots Are Moving from Passive Execution to Proactive Service

For a long time, the core capabilities of service robots were mainly centered on task completion:

Can they navigate to a specified location?

Can they deliver items, guide users, perform inspections, and complete other actions?

Can they operate stably along predefined routes and workflows?

With the development of large models, embodied intelligence, spatial intelligence, and edge computing, robots are entering a new stage.

Future robots will not only understand instructions and execute actions. They will also need to understand their environment, assess human status, identify scene changes, and make more valuable service decisions at the right time.

This means that service robots are gradually evolving from task execution systems into scene understanding systems.

From World Models to Value Models: The Key Is Understanding the Real World

A world model can be understood as a robot's internal cognitive system for the real environment.

It needs to know:

Where people and objects are;

How the spatial structure is changing;

Whether the current scene is safe;

What may happen next;

And what outcome may result from a specific action.

But understanding the world is not enough. In real service scenarios, robots also need to determine which actions are valuable.

For example, in hotels, office buildings, hospitals, and commercial spaces, robots are not facing fixed workstations. They are working around dynamic crowds, complex lighting, open paths, and constantly changing user needs.

They need to know not only what is happening, but also how to decide:

Whether to proactively approach and provide service;

Whether to keep a proper distance;

Whether to yield to pedestrians;

Whether to prioritize the current task;

And which action can deliver a better user experience and higher operational efficiency.

This is the core shift from a world model to a value model.

A world model answers: how robots understand the world.
A value model answers: how robots create service value.

The Era of Proactive Service Requires a Stable and Reliable Vision Interface

For service robots, the first step toward understanding the real world is high-quality perception input.

In complex scenarios, the challenge is not only whether the robot's brain is smart enough, but also whether it can see clearly, accurately, and consistently understand the real environment.

For example:

Recognition capability in low-light environments,

Target perception in dynamic crowds,

Depth understanding in complex spaces,

Collaboration and calibration across multiple camera modules,

Long-term imaging stability,

And cost and consistency during large-scale deployment.

These factors determine whether robots can move from the laboratory into real-world applications, and whether they can provide continuous service over time.

As a result, the camera is no longer just an image capture component.

In the robotics era, vision is becoming a critical interface for robots to understand the world. It is also evolving from a single hardware capability into a system capability that supports spatial understanding, human-machine interaction, behavior recognition, proactive service decisions, and data feedback loops.

Union Image: Building Vision Infrastructure for the AI Era

Union Image has long focused on the AIoT vision field. Around camera modules, embedded vision systems, and complete imaging products, the company continues to build integrated capabilities from product definition and software-hardware development to mass production delivery.

For robotics and embodied AI scenarios, Union Image aims to move beyond traditional vision hardware supply and become a scenario-oriented perception system partner.

We care about more than helping robots see. We also focus on:

How robots can see stably under complex lighting;

How robots can maintain continuous perception in dynamic spaces;

How vision data can support world model construction;

And how perception systems can serve the final user experience and business value.

In the era of proactive service, robots need not only a stronger brain, but also more reliable eyes.

Union Image hopes to participate in building vision infrastructure for the AI era, helping robots understand the real world more stably, reliably, and cost-effectively.

Vision for All. Vision for Robotics.