Pudu Robotics moves towards "operation": a data migration experiment

```

On July 17, during the 2026 World Artificial Intelligence Conference (WAIC), Pudu Robotics (hereinafter referred to as "Pudu") unveiled the Physical Agent architecture of "Body + System + Skills".

According to Pudu's vision, the robots of the future will be physical intelligent agents composed of robot bodies, intelligent systems, and reusable skills. Robots of different forms can share a set of underlying models and software architecture, and then activate corresponding abilities based on specific tasks—this is known as "one brain, multiple forms".

At the body layer, Pudu has already deployed three types of products: dedicated robots, humanoid-like robots, and humanoid robots.

Among them, dedicated robots mainly handle high-frequency, standardized tasks; humanoid-like robots enter complex industrial and commercial environments; humanoid robots explore more general scenarios of human-robot collaboration.

At the system layer, Pudu has launched Pudu Agent OS, whose core value is to enable robots to share the same "brain". Through a unified system, different robots can reuse existing abilities, achieving cross-body and cross-scenario migration.

The skills layer determines what specific work robots can accomplish.

Compared to the architecture, what draws greater market attention this time is the real-world data Pudu has already accumulated.

Insufficient data has always been a core challenge faced by the robotics industry. Large language models can obtain vast amounts of text and image data from the Internet, but the data needed by robots must come from real or highly replicated physical environments, including spatial changes, movement trajectories, object interactions, execution feedback, as well as failure and correction processes. Data collection is costly and slow.

Long-term commercial deployment has provided Pudu with a relatively stable source of data.

According to Pudu, its robots have entered numerous commercial scenarios worldwide, and so far the cumulative shipments have exceeded 130,000 units, generating over 50 million hours of real operation data each year.

Although Pudu covers a wide range of scenarios, the tasks currently undertaken by its robots are still mainly focused on delivery and cleaning.

The next stage in embodied intelligence is solving whether robots can truly accomplish their tasks upon arrival.

Wu Xiang, General Manager of Pudu Robotics' Embodied Intelligence Business Line, told Wall Street Insights that in the next stage, the company will explore operational tasks in retail supermarkets, industrial manufacturing, and warehousing logistics, enabling robots to further perform actions such as picking up, transporting, sorting, restocking, and assembling upon arrival at designated locations.

Pudu's proposal of "one brain, multiple forms" is also aimed at enabling these skills to be reused among different robots.

Following this logic, the environmental perception and navigation abilities accumulated by delivery robots can be transferred to humanoid-like and humanoid robots; task comprehension and operational skills trained on a certain type of robot can also be adapted and deployed on other bodies.

This approach may lead to a data flywheel: robots of different forms enter real-world scenarios and generate data, the data is used to train a unified model, the model develops new skills, which are then deployed to more robots and tasks.

Whether this data flywheel can effectively operate in real commercial scenarios is now under close watch.

Risk Warning and DisclaimerThe market entails risks, and investment needs caution. This article does not constitute individual investment advice and does not take into account the specific investment objectives, financial condition, or needs of individual users. Users should consider whether any opinions, viewpoints, or conclusions in this article suit their particular circumstances. Investing based on this is at your own risk. ```