How Unitree Passed 5,000 Humanoid Shipments

Unitree has moved beyond viral demonstrations into measurable commercial scale. Omdia figures reported by the Associated Press put the company’s 2025 humanoid shipments above 5,000, while a new US restriction on foreign-made humanoid and quadruped robots has made market access a more immediate constraint.

The central explanation for Unitree’s rise still holds: it sells physical robots alongside unusually broad public development resources. But “open innovation” is only part of the answer. The company combines that accessible software layer with comparatively low entry prices, integrated components and enough production capacity to place a common robot body in many laboratories.

The shipment milestone is real, but it needs context

More than 5,000 humanoid shipments in one year is a significant result in an industry that remains at an early commercial stage. The figure applies to Unitree’s humanoid range, however, not specifically to the G1, and it measures shipments rather than revenue, active installations or autonomous work completed.

That distinction matters because spectacular motion is not the same as useful autonomy. A robot can walk, recover its balance or reproduce a trained routine while still requiring extensive engineering before it can perceive an unfamiliar workplace, choose a safe action and complete a valuable task without supervision.

Unitree’s achievement is therefore best understood as distribution of a hardware platform. It has put a large number of relatively consistent machines into the hands of researchers, developers and other buyers. Every additional installation can make shared control software, simulation models and training procedures more useful to the next team working on the same embodiment.

What Unitree has actually opened

Unitree’s official public repositories now cover far more than a basic device interface. They include C++ and Python SDKs, reinforcement-learning environments, MuJoCo and Isaac Lab simulation projects, XR teleoperation, ROS resources, robot descriptions, LeRobot-based manipulation workflows and models in the UnifoLM family.

This breadth lowers several practical barriers. A laboratory can inspect example code before buying hardware, begin controller work in simulation and use interfaces already recognized by robotics students and engineers. Researchers can also describe their setup more precisely when peers have access to the same robot model and at least some of the same software.

The result is a useful feedback loop, although it should not be overstated. Public tools can attract experiments; those experiments create documentation, integrations and trained policies; and that surrounding work makes the underlying robot easier to select for another project. Unitree benefits even when it did not write every application built on top of its interfaces.

This is an open development layer, not a fully open robot. Public SDKs and model files do not automatically expose proprietary firmware, actuator designs, manufacturing methods or every control mode. The accurate claim is that Unitree provides substantial open access around commercially produced hardware—not that the entire machine is open-source.

The headline price hides an important developer distinction

The G1 made Unitree conspicuous because its advertised entry price sits well below many advanced humanoid research systems. The current official G1 specification page lists the base model from $13,500 before tax and shipping, with a standing height of 1.32 metres, weight of about 35 kilograms and 23 joints.

That price does not describe every configuration a robotics laboratory may need. Unitree’s own comparison marks secondary development as unavailable on the base G1 and available on the G1 EDU, whose price requires contacting sales. The EDU configuration can also expand to as many as 43 joints and accept higher-compute modules and optional dexterous hands.

Consequently, the cheapest G1 should not be treated as a complete autonomous-research package. Computing, hands, support, shipping, safety infrastructure and engineering time can materially change the project cost. Even so, a visible entry price gives prospective users a concrete starting point and makes the platform easier to consider than systems sold only through lengthy private negotiations.

Why shared hardware can accelerate research

Robotics software is unusually dependent on the physical machine beneath it. A policy trained for one arrangement of joints, sensors and actuators does not simply transfer to every other humanoid. When many teams acquire the same embodiment, work on calibration, teleoperation, simulation and motion control becomes more reusable.

Unitree’s public stack supports this form of standardization. A team can collect demonstrations through teleoperation, train a policy using a supported workflow and deploy through documented robot interfaces. Another group can examine the method without first recreating the entire software connection to the machine.

This does not mean community experiments become official product capabilities. A research video may depend on a particular robot configuration, external computer, carefully prepared environment and custom policy. Buyers should separate what Unitree supplies from what a third-party laboratory has added, and they should not assume that a published demonstration will run on a newly unpacked base model.

Manufacturing is the other half of the advantage

Open repositories have little platform value if the corresponding machine is unavailable, inconsistent or prohibitively expensive. Unitree’s shipment volume indicates that it paired developer access with the less glamorous work of producing and delivering physical systems at scale.

The company also develops key components and integrates sensing, motion control and mechanical systems into finished robots. That vertical integration can support lower costs and quicker iteration, but it also concentrates control over repairs, firmware behavior and future compatibility. Openness at the application layer therefore coexists with dependence on a single hardware vendor.

Production scale and public tooling reinforce each other. More robots create a larger audience for reusable software, while a larger body of software makes the hardware more attractive. Neither factor alone fully explains why Unitree emerged near the front of the young humanoid market.

The next constraint is access, not developer interest

The US Federal Communications Commission’s July 2026 action changes the practical meaning of availability. The restriction applies to new versions of foreign-made humanoid and quadruped robots; reporting indicates that previously approved models may continue to be sold, so it should not be described as a blanket removal of every existing Unitree machine.

For US laboratories, procurement status, model approval and support continuity now deserve the same attention as SDK features. Open code can preserve knowledge and simulation work, but it cannot substitute for lawful access to replacement hardware, batteries, hands or other physical components.

Outside that market, Unitree’s underlying formula remains intact: make capable robot bodies comparatively attainable, expose enough of the development surface for serious experimentation and let external teams expand the range of research conducted on them. Its shipment record shows that the formula achieved scale. It does not yet prove that today’s humanoids are general-purpose workers—or that public software alone put Unitree there.

Unitree Passed 5,000 Humanoid Shipments—Open Tools Explain Only Part of Its Rise