What Unitree’s “Three Bases, Two Centers” Plan in Chengdu Means

Unitree Robotics’ new wholly owned Chengdu subsidiary, Chengdu Yuchen Technology, is more than a regional office on paper. Registered on September 2, 2026, the company is part of a proposed local footprint that links research and pilot-scale testing with application development, data services and industry–education collaboration. 10

The central idea is straightforward: a robot maker needs a repeatable path from prototype to dependable product. Chengdu is meant to provide a place where Unitree can develop and test machines, place them in selected real-world settings, capture operational data and train the people who will deploy and maintain them. The facilities are announced plans, rather than evidence that all capacity is already operating. 1012

The “three bases, two centers” model

Reporting on the initiative describes three bases and two centers for embodied intelligence. 12 Together, they point to an effort to cover the full chain from hardware development to use-case validation.

1. A western pilot-production base

The pilot-production base is intended to support humanoid, quadruped, wheeled and robotic-arm products as they move beyond engineering prototypes. In practice, that transition requires reproducible assembly, supply-chain coordination, reliability testing, quality control and serviceability—not just a convincing demonstration.

Chengdu’s policy direction is compatible with that goal: the city’s AI innovation plan explicitly includes support for small-batch trial production to help innovations move from technical development toward market use. 27

2. Film, tourism and digital-content robotics

A film-and-television robotics base would connect robots with cultural tourism, entertainment and digital content. These are potentially useful early environments because routes, venues and staff oversight can be more controlled than in homes or open-ended workplaces.

That does not make these deployments proof of broad utility. They can, however, help operators learn how visitors interact with robots, how frequently machines need intervention, and whether a venue can generate recurring revenue from a robot-led service or attraction.

3. Education robotics

The education component is intended to span schools, vocational education and higher education. Its value is not only equipment sales: it can create curricula, expose students to robotics development and help form a local technician and developer pipeline.

For embodied AI, practical deployment skills matter alongside model research. Organizations need people who can operate machines safely, integrate them into workflows and maintain them when conditions differ from the lab.

4. An application-innovation center

The application center is aimed at converting general-purpose hardware into deployable solutions in areas such as public safety, education and cultural tourism. Chengdu’s broader AI policy emphasizes selecting, supporting and scaling application scenarios, a model that gives robotics companies potential test venues and institutional partners. 28

5. A real-world data-collection center

The data center is the strategic core of the proposal. Embodied AI systems need data on perception, movement, manipulation and human interaction in physical environments. That data can support training and iteration of the models that govern robotic behavior.

Unitree founder Wang Xingxing has said robot “brains” may be approaching a “ChatGPT moment.” 3 Whether or not that inflection arrives soon, the comment explains why data collection matters to Unitree’s strategy: better mechanical hardware alone will not solve the difficult task of reliable operation in changing, unstructured settings.

Why Chengdu matters to Unitree

The plan reportedly makes Unitree the fifth prominent Chinese embodied-AI company to establish a Chengdu presence, alongside firms including UBTECH, AgiBot and Galbot. 12 The importance is not the ordinal ranking; it is the prospect of clustering manufacturers, component suppliers, AI developers, universities, deployment sites and public-sector partners in one region.

There is evidence that Chengdu is already trying to support production as well as demonstrations. AgiBot launched a southwest embodied-intelligence industrial base in Chengdu in May 2026 and said its first 200 new-generation humanoid robots had rolled off the line there. 41 That is not proof that Unitree will achieve comparable scale, but it shows the city can host an operating robotics manufacturing effort.

Chengdu also says its core AI industry exceeded 150 billion yuan in 2025, with a target above 260 billion yuan by 2027. 35 For a robotics company, that ecosystem can matter because embodied AI depends on more than robot bodies: it needs perception systems, compute, simulation, systems integration, application developers and customers willing to test new workflows.

GD01’s role: attention, not commercial validation

Unitree’s GD01 mecha is a useful illustration of the difference between technical spectacle and commercialization. The rideable machine can move in bipedal and quadrupedal configurations; Unitree has described it as a mass-produced transformable mecha. 48

Such a product can be valuable for brand awareness, content creation and controlled cultural-tourism experiences. It can also demonstrate mobility hardware and attract students, developers and potential partners. But it is not evidence that humanoid robots have solved the practical problems of factory work, hospital integration or unsupervised public operation.

That distinction matters because reporting still characterizes commercial robot use as limited, with many deployments focused on research or display. 49

The real test: scene-led businesses

Unitree’s Chengdu plan will be judged less by how many facilities are announced than by whether those facilities create viable, repeatable deployments.

• In factories, robots need to improve throughput, quality, safety or labor availability at a cost that beats alternatives.
• In hospitals, they need safe workflows, maintenance support and integration that genuinely saves staff time.
• In shopping streets, attractions and cultural venues, operators need ongoing revenue from services such as guiding, entertainment, patrols, delivery or sponsored experiences—not one-off promotional events.

The most important metrics will be paid utilization, uptime, maintenance expense, safety performance and the speed at which real-world data improves robot capability. Chengdu’s ecosystem and Unitree’s proposed footprint can make those experiments easier to run. They do not, by themselves, establish sustainable robot economics.

Bottom line

Unitree’s Chengdu subsidiary is best read as a bid to build a localized embodied-AI flywheel: pilot production feeds deployments; deployments create data; data improves software; education and local partners expand the capacity to deploy again. 1012

It is a credible strategic response to the gap between impressive prototypes and reliable scaled products. But the decisive outcome will be commercial, not ceremonial: can Unitree turn Chengdu’s test scenarios into recurring, measurable customer value?