Pudu Robotics used the World Artificial Intelligence Conference in Shanghai to make a bigger claim than simply introducing another robot. Its message was that the next phase of commercial robotics will be won by companies able to connect intelligence, hardware, deployment and real-world data in one continuous system.

The centrepiece was the first offline public appearance of the PUDU D7, an industrial-grade semi-humanoid robot built for material handling, shelf picking and warehouse logistics. Yet the more important announcement was the architecture behind it: Pudu’s “One Brain, Multiple Embodiments” strategy, designed to let delivery robots, cleaning machines, industrial platforms and humanoids share a common intelligence layer.

The D7 had already been introduced online in June. WAIC was therefore less a conventional product launch than Pudu’s attempt to show how its commercial robot business could become a platform for Physical AI.

From Useful Machines to Learning Systems

Commercial robots are familiar in hotels, restaurants, shopping centres, hospitals and warehouses. Most are very good at a narrow job: moving meals, cleaning floors or transporting goods along predictable routes. Embodied AI promises something more adaptable—a machine that can interpret its environment, understand an instruction, plan an action and improve through feedback.

Pudu argues that the missing ingredient is not another foundation model in isolation. It is a reliable loop between the model, the robot body and thousands of real deployments.

That argument is commercially credible. The International Federation of Robotics reported that almost 200,000 professional service robots were sold in 2024. Transport and logistics accounted for more than half, while professional cleaning robot sales rose 34%. Physical AI is arriving in categories where customers already understand the operational value of automation.

Pudu also brings scale. The company says it has shipped more than 130,000 robots across over 85 countries and regions. Frost & Sullivan’s 2025 research ranks it first globally in commercial service robot revenue and shipments, although market-leadership claims should still be read alongside the methodology and scope of the underlying report.

One Brain, Different Jobs

Pudu’s architecture has three connected layers. The robot body is the physical interface, ranging from specialised delivery and cleaning machines to semi-humanoid and humanoid designs. PuduAgent OS acts as the coordinating system, combining visual-language models, memory, planning and safety controls. PuduFM provides the embodied intelligence intended to help robots perceive scenes, reason about physical outcomes and execute actions.

“By building a unified intelligence foundation, we enable different robot forms to share the same ‘brain,’ allowing robotics to evolve from standalone intelligence to a platform-based intelligence ecosystem,”

said Shawn Wu, General Manager of the General Embodied AI Product Line at Pudu Robotics.

READ MORE: Local 5G Just Got Easier with IIJ’s eSIM Model

The practical promise is skill transfer. A capability developed for one machine or environment could eventually be adapted across other robot forms without rebuilding the entire software stack. For customers operating mixed fleets, that could reduce integration work and make automation easier to scale.

The difficult part is proving that shared intelligence works outside controlled demonstrations. A cleaning robot, delivery platform and dual-arm warehouse machine have different safety limits, mechanical constraints and failure modes. Businesses will need evidence on uptime, intervention rates, task-completion accuracy and total operating cost—not only model architecture.

What the PUDU D7 Adds

The D7 is a wheeled semi-humanoid rather than a fully bipedal machine. That may appear less futuristic, but it is a sensible choice for warehouses and factories where stability, runtime and payload matter more than walking like a person.

Pudu says the robot can carry up to 14 kilograms and work at heights of up to two metres. Its dual arms are intended for high-reach picking, replenishment, material movement and intralogistics. The design positions it between conventional autonomous mobile robots, which mainly transport goods, and general-purpose humanoids expected to manipulate a wider variety of objects.

READ MORE: LG Brings World’s First CLOi ServeBot to U.S. Hospitality

It will not suit every operation. Facilities with fixed, repetitive workflows may achieve a faster return from conveyors, robotic arms or simpler mobile robots. The D7 becomes more interesting where layouts change, goods vary and human-oriented infrastructure makes conventional automation awkward.

Pudu has not yet provided enough public detail on commercial pricing, deployment timelines or independently verified performance in live customer environments. Those figures will determine whether the D7 becomes an operational product or remains an impressive expression of the company’s roadmap.

The Race Is Moving Beyond Hardware

Pudu is not alone in pursuing Physical AI. UBTECH’s Walker S2 targets industrial work with dual-arm manipulation and autonomous battery swapping. Agility Robotics has taken a narrower deployment-first route with Digit, placed into live GXO logistics operations through a Robots-as-a-Service agreement. Keenon and Bear Robotics are also extending established service-robot fleets toward humanoid or manipulation-led systems.

The alternatives reveal the market’s central debate. Should companies begin with a general humanoid body and find repeatable jobs, or start with proven commercial robots and gradually add intelligence and manipulation? Pudu is firmly choosing the second path.

Conclusion: Deployment Data May Be Pudu’s Real Advantage

Pudu’s strongest asset is not the D7’s shape. It is the possibility of turning years of commercial deployment into a data and distribution advantage. Robotics is shifting from selling isolated machines toward managing fleets, transferring skills and improving models through real operating experience.

Scale alone will not guarantee leadership. Agility can point to paid warehouse deployments; UBTECH is pushing industrial humanoids toward continuous operation, and Bear Robotics is combining a large service fleet with manipulation technology. Pudu now needs transparent customer results showing its shared brain can produce measurable gains across different bodies.

If it can demonstrate that, “One Brain, Multiple Embodiments” will be more than a polished AI slogan. It could become the operating model connecting today’s useful service robots with tomorrow’s more capable physical agents.