# Unitree Robotics

> Source: https://aiwiki.ai/wiki/unitree
> Updated: 2026-07-28
> Fact-checked: 2026-08-04
> Categories: AI Companies, Chinese AI, Robotics
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "Unitree Robotics." aiwiki.ai, 28 Jul 2026. https://aiwiki.ai/wiki/unitree
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

Unitree Robotics is the international brand of Yushu Technology Co., Ltd. (宇树科技股份有限公司), a Chinese [robotics](https://aiwiki.ai/wiki/robotics) manufacturer based in Hangzhou, Zhejiang. Wang Xingxing founded its predecessor company on August 26, 2016. Unitree develops, manufactures, and sells electric [quadruped robots](https://aiwiki.ai/wiki/quadruped_robot), [humanoid robots](https://aiwiki.ai/wiki/humanoid_robot), robot components, and software used to control or develop those machines. Its product history began with the Laikago quadruped in 2017 and expanded to humanoids with H1 in 2023.[1][2]

Unitree reports commercial scale across several customer categories rather than only prototype activity. The company's June 2026 registration prospectus reported 33,294 quadruped sales during 2023-2025 and 5,511 humanoid shipments during 2025. Those are issuer-reported figures. Unitree's exchange filings also caution that estimates of global robot shipments and market share use inconsistent definitions and are not objectively comparable, so claims that the company held a particular percentage of a global market should not be treated as a settled measurement.[1][3]

The company combines hardware manufacturing with motion-control software, developer interfaces, and research on [embodied AI](https://aiwiki.ai/wiki/embodied_ai). Its robots can be operated by a person, execute programmed or API-directed routines, or run learned control policies, depending on the model and configuration. Unitree told the Shanghai Stock Exchange in 2026 that its general embodied models remained in research, testing, and pilot deployment rather than large-scale use in products. This distinction matters: a robot that balances, runs, or follows a prescribed sequence without continuous teleoperation is not necessarily able to understand and complete an unfamiliar task on its own.[4]

This article covers the company and its product-platform business through August 5, 2026. Detailed specifications belong in the articles about individual robots, and the proposed public offering is covered separately.

## Corporate identity and organization

The company was incorporated in Hangzhou as a limited-liability company with registered capital of 100,000 yuan, wholly contributed by Wang Xingxing. It converted into a joint-stock company on May 28, 2025. The prospectus identifies the legal issuer as Yushu Technology Co., Ltd. and describes Hangzhou Yushu Technology Co., Ltd. as its predecessor. "Unitree Robotics" remains the public-facing English brand.[1]

Wang served as chairman, general manager, and chief technology officer when the prospectus was filed. Before the proposed offering, he directly held 23.8216 percent of the shares and, together with an employee incentive platform he controlled, controlled 34.7630 percent of the equity. A dual-class voting arrangement gave him 68.7816 percent of the voting power. These percentages describe the pre-offering structure in the filing, not a permanent ownership figure.[1]

Unitree reported 516 employees at the end of 2025. Of those, 184 were classified as research and development staff, 154 as sales staff, 123 as production staff, and 55 as management. The headcount had risen from 264 at the end of 2023. The figures show that Unitree is both a development organization and a manufacturing and sales business, rather than a research laboratory that only produces prototypes.[1]

The company describes its business as the development, production, and sale of humanoids, quadrupeds, components, and embodied-intelligence models. Its filings divide customers into three broad groups:

- **Research and education:** universities, research institutes, technology companies, and individual developers using robots for experiments, teaching, or secondary development.
- **Commercial and consumer:** buyers using robots for entertainment, exhibitions, stores, cultural events, or personal development.
- **Industry applications:** customers or integrators using robots for inspection, emergency response, manufacturing, logistics, and other operational settings.

These are Unitree's accounting classifications, not standardized industry definitions. A sale to a developer or integrator does not establish how the final system was deployed, and Unitree says it does not always know the ultimate use made by downstream customers.[3][4]

## History

### XDog and the 2016 founding

The technical starting point for Unitree was XDog, a quadruped prototype Wang built while completing a master's degree. According to the company, he designed the motor drive, electronics, mechanical system, control architecture, and software around relatively inexpensive external-rotor brushless motors. A video released around his 2016 graduation drew attention from media and prospective investors. Wang left a job in Shenzhen and formed the company with angel financing later that year.[2]

XDog was a prototype rather than a Unitree commercial product. After incorporation, the company rebuilt the design as Laikago, which appeared in September 2017. Unitree says it redesigned the motor, drive electronics, master control, mechanical structure, and control system rather than simply selling the student prototype. The prospectus describes Laikago as the company's first commercial delivery.[1][2]

### Quadruped development

AlienGo followed in 2019 with a new power system and integrated body design. Unitree introduced A1 in 2020 and Go1 in 2021, shifting part of the line toward lower-cost research and consumer use. B1, released in 2022, addressed industry customers. Go2 and B2 arrived in 2023 as successors in the consumer and industry branches, and B2-W added wheels to the industrial quadruped architecture in 2024.[1][2]

This sequence established a product pattern that Unitree continued to use: smaller Go-series machines for consumer, education, and developer demand, and larger B-series machines for payload, endurance, environmental protection, and industry integration. The categories overlap. A consumer-labeled robot may be used in a laboratory, and an industry model may still require a customer or systems integrator to add navigation, sensing, task software, or specialized equipment.[3]

Unitree added A2 in 2025 and As2 in February 2026. By July 2026 its official site also described a wheeled As2-W variant. The As2-W page labels published performance values as laboratory results, notes that configurations differ, and says that some illustrated functions were still being developed and would become available gradually. That caveat is important for all current product comparisons: a model-family name alone does not establish its sensors, computer, software access, autonomy, payload, or measured field performance.[2][5]

### Expansion into humanoids

Unitree released [Unitree H1](https://aiwiki.ai/wiki/unitree_h1), its first full-size humanoid, in August 2023. The company reused experience from electric actuators, compact body integration, estimation, and legged motion control, but a biped introduces different balance, fall, manipulation, and safety problems from a quadruped. H1 was initially positioned as a development and research platform, with later variants adding arms and other hardware.[1]

[Unitree G1](https://aiwiki.ai/wiki/unitree_g1) followed in May 2024 as a smaller platform. Unitree's filings distinguish base versions from developer-oriented versions. The EDU versions of G1, H1, and some other models expose secondary-development functions, while base configurations can be limited to supported commands and packaged behaviors. Consequently, demonstrations created with an EDU robot, added compute, or custom software should not be assumed to describe every retail configuration.[1]

In July 2025 Unitree introduced [Unitree R1](https://aiwiki.ai/wiki/unitree_r1), a lighter humanoid intended for developers, education, and consumer-facing uses. It announced [Unitree H2](https://aiwiki.ai/wiki/unitree_h2), a second full-size line, in October 2025. The prospectus also lists G1-D, a stationary or wheeled dual-arm form intended for data collection, model training, and work-cell experiments. These products broadened the range of body sizes and configurations, but they did not by themselves demonstrate general-purpose autonomy.[1][2]

Unitree used large public performances to demonstrate repeatable locomotion and multi-robot coordination. Company events included quadrupeds at the 2021 Spring Festival Gala and humanoids at the 2025 and 2026 galas. In its exchange response, Unitree explained that the 2026 group performance used a control system to send target positions and action sequences through APIs. On-board localization, perception, and control handled execution and adjustment, but engineers had prescribed the logic and sequence. The company itself contrasted this kind of structured automation with intelligent operation in unfamiliar environments, which it said still required further validation.[2][4]

## Product families

Unitree's catalog is better understood as a set of platform families than as a single ladder from least to most capable. Prices, sensor packages, computers, hands, batteries, environmental ratings, and software permissions vary by region and configuration. Published peak values are often measured in controlled conditions and should not be read as guaranteed continuous performance.

| Family | First Unitree release | Intended role in company materials | Scope and currentness boundary |
|---|---:|---|---|
| Laikago | 2017 | First commercial quadruped; research and development | Historical platform. It established the company's electric quadruped architecture. |
| AlienGo and A1 | 2019 and 2020 | Research, education, and mobility development | Earlier platforms that remain important in outside research papers; later Go models replaced much of their commercial role. |
| Go1, Go2, and Go2-W | 2021 onward | Consumer, education, and developer quadrupeds | Capabilities differ between retail and developer editions. Added customer code is not a factory feature. |
| B1, B2, and B2-W | 2022 onward | Industry-oriented inspection, emergency, and integration platforms | Industry deployment generally depends on payloads, application software, and site integration. |
| A2, As2, and [Unitree AS2-W](https://aiwiki.ai/wiki/unitree_as2_w) | 2025 onward | Medium-size and wheeled-legged industry platforms | The line was still changing in July 2026. Official pages warn that configurations and functions vary and some functions remain under development. |
| H1 and H2 | 2023 onward | Full-size humanoid research and application platforms | Separate articles track model specifications and revisions. Model demonstrations do not establish unsupervised general labor capability. |
| G1 and G1-D | 2024 onward | Compact humanoid development, interaction, data collection, and work-cell experiments | Base, EDU, hand, and compute options have different development access and degrees of freedom. |
| R1 | 2025 | Lighter humanoid for development, education, and interactive uses | Product materials describe multimodal interaction, but open-ended embodied task execution remained a research capability. |

The B2-W and As2-W illustrate another branch, the [wheeled-legged robot](https://aiwiki.ai/wiki/wheeled_legged_robot). Wheels can reduce energy use and increase speed on suitable surfaces, while articulated legs retain the ability to negotiate obstacles. That does not make a wheeled-legged platform equally capable on every terrain. Tire contact, ground conditions, payload, battery state, temperature, controller settings, and operator choices all affect the result.[1][5]

Unitree also sells or integrates joint modules, motors, reducers, manipulators, dexterous hands, cameras, and [LiDAR](https://aiwiki.ai/wiki/lidar). The company says it can design its own versions of many of these components. Its filings nevertheless disclose a mixed production model: core assembly and some components are handled internally, while mechanical parts, circuit-board assembly, injection molding, machining, and some sensor or hand configurations are purchased or outsourced. "Full-stack research and development" therefore describes the breadth of the company's design capability, not a claim that every physical part in every robot is manufactured in-house.[1][4]

## Engineering and control stack

### Mechanical and electronic integration

Unitree's engineering approach centers on electric joint modules, compact body integration, estimation, perception, and whole-body motion control. The prospectus identifies integrated joints, compact body structures, heat management, energy management, high-dynamic control, fall protection, and component design as core areas. These are company descriptions made in a securities filing. They support the scope of the work, but broad statements that a component or algorithm is "leading" remain issuer assessments unless an independent comparison supplies a common test.[1]

The production process combines internal assembly with a supplier network. Unitree says complete robots and core modules are assembled internally. It procures many custom mechanical parts and uses outside contractors for activities such as printed-circuit-board assembly, injection molding, and selected machining. For LiDAR, cameras, and dexterous hands, it offers some internally developed options and some purchased options to meet different customer requirements. This makes model and bill-of-material details necessary when evaluating vertical-integration claims.[1][4]

Electric actuation helped Unitree build machines that are smaller and easier to deploy than early hydraulic research robots, but actuator type alone does not determine overall capability. Gear design, cooling, batteries, structural strength, sensing, software, and the duty cycle all matter. Peak torque, top speed, maximum obstacle height, or a filmed backflip measures a narrow operating point; it does not report reliability, fall frequency, maintenance, task completion, or safe operation around people.

### Software access and open repositories

Unitree publishes a version-two software development kit under the BSD 3-Clause license. The repository contains C++ libraries, headers, and examples for communicating with supported robots, and it directs users to the company's documentation center. An SDK exposes interfaces; it does not mean that firmware, trained models, hardware designs, or every behavior are open source.[6]

The company also publishes `unitree_rl_gym`, a [reinforcement learning](https://aiwiki.ai/wiki/reinforcement_learning) repository for Go2, H1, H1-2, and G1. Its documented workflow covers training in simulation, replay, transfer between simulators, and deployment to supported physical robots. The repository builds on Isaac Gym, MuJoCo, `legged_gym`, `rsl_rl`, and Unitree's own communication libraries. It provides a practical starting point for locomotion experiments, but successful execution still depends on the exact robot, firmware, configuration, network, trained policy, and safety procedures.[7]

[Sim-to-real transfer](https://aiwiki.ai/wiki/sim_to_real_transfer) is central to much of this work. Developers train a policy in a physics simulator, vary parameters to reduce sensitivity to modeling errors, test the policy in another simulator, and then deploy it to hardware. Simulation makes large-scale training less expensive and less destructive than learning every movement on a physical robot. It does not eliminate the reality gap. Contact, actuator delay, sensor noise, floor compliance, battery behavior, mechanical wear, and unmodeled disturbances can produce failures that were absent in simulation.

### Motion, automation, and autonomy

Unitree's 2026 exchange response separates three control modes:

| Mode | Where the immediate command originates | Typical use | Main limitation |
|---|---|---|---|
| Remote control | Human operator using a controller or application | Direct movement, demonstrations, emergency intervention | Requires an operator and does not show autonomous task understanding. |
| Programmed automation | Navigation, group-control, or application software using APIs and predefined logic | Routes, synchronized actions, structured inspection, repeated routines | Generalization is limited when the environment or task differs from the programmed case. |
| Model-directed operation | A language or embodied model interprets text, speech, vision, or other inputs and produces a task or action plan | Research on open-ended interaction and task execution | Reliability and stability remained under testing; Unitree reported no large-scale general embodied-model deployment during the filing period. |

All three modes can use automatic low-level stabilization. A remotely directed robot may still balance itself, and a choreographed robot may still localize and correct its footsteps. Conversely, language input does not prove that a general model safely controls every joint or understands all consequences. Articles and demonstrations should identify the control boundary instead of labeling the complete system simply "autonomous."[4]

Unitree reported that learned body-motion models were already used across its products and that some models connected to third-party language models for conversation or intent recognition. It separately said that its more general embodied projects, including world-model-action (WMA) and vision-language-action (VLA) models, had been tested or piloted rather than integrated at scale. This is the appropriate currentness boundary for claims that Unitree sells a general-purpose intelligent worker.[4]

## Use as a research platform

Outside research helps show what Unitree hardware can support, but the results belong to the researchers and their particular modifications. A 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems paper transferred a learned running policy to an A1 quadruped. The physical experiment reported bounding at 2 meters per second while carrying 5 kilograms. It demonstrated one learned locomotion task under the authors' test conditions, not the behavior of every A1 or a general navigation system.[8]

At the 2025 Conference on Robot Learning, researchers presented HuB, a framework for balance-intensive humanoid motions. They validated it on a G1 using reference-motion refinement, balance-aware policy learning, and sim-to-real training. The work is evidence that a commercially obtainable G1 can serve as a physical test platform for independent control research. It is not evidence that the same policy ships with the robot or that the robot can perform unrelated tasks.[9]

These examples explain part of Unitree's importance to the research ecosystem. A lab can buy a supported body, use an SDK or simulation model, and concentrate on control, perception, manipulation, or learning rather than designing a complete actuator and mechanical system first. The tradeoff is platform dependence. Results may rely on a specific edition, firmware release, sensor, payload, or proprietary low-level controller, and they may not reproduce on a lower-cost base model.

## Sales, applications, and financial development

### Reported unit scale

The prospectus gives product sales rather than only production goals. During 2023-2025, reported quadruped sales were 3,121, 7,136, and 23,037 units, totaling 33,294. Humanoid sales were 5 units in 2023, 412 in 2024, and 5,215 in 2025; the company reported 5,511 humanoid shipments in 2025.[1]

The 2025 shipment figure counts units dispatched, while the sales table records units for which the company's sales-recognition conditions were met. The prospectus attributes the 296-unit difference to products that had been shipped but had not completed customer sign-off, acceptance, or other revenue-recognition procedures by year-end. Shipment, production, sale, delivery, active deployment, and robot-hours are different measures. Industry estimates also differ over whether to count bipedal, wheeled, dual-arm, research, and demonstration systems. Unitree's own first exchange response says there was no authoritative equivalent of vehicle-registration data and that published market-share percentages from different consulting firms could not be compared directly.[1][3]

The often-repeated figure that Unitree held 69.75 percent of the 2023 global quadruped market came from one consultancy's estimate of unit share. Another estimate cited in the same filing measured revenue share, and a later source used a different 2024 methodology. The exchange response explicitly says the datasets lacked a unified objective basis. A careful summary can say that several estimates placed Unitree first, but it should not combine their percentages into a trend or present one estimate as a census.[3]

### What customers were buying robots for

Unitree's 2025 interim revenue mix gives a more restrained picture of commercial use than product videos alone. For the first nine months of 2025, it classified quadruped revenue as 31.58 percent research and education, 42.30 percent commercial and consumer, and 26.12 percent industry applications. Humanoid revenue was 73.60 percent research and education, 17.39 percent commercial and consumer, and 9.01 percent industry applications.[3][4]

Within the humanoid industry category, Unitree estimated that 50 to 70 percent came from corporate reception or tour-guide uses. The remaining identified examples included manufacturing, inspection, and logistics. The company also said downstream developers frequently bought robots, added software or hardware for a particular application, and resold an integrated system. These disclosures show real commercialization, but they do not support the stronger claim that thousands of humanoids were already performing general manufacturing labor.[4]

Applications differ in what they demand from a platform. Research users value low-level access, compute expansion, durability, and repeatable interfaces. Commercial demonstrations emphasize interaction and programmed motion. Industrial inspection adds environmental protection, endurance, payloads, site mapping, application software, and support. A successful deployment in one group does not establish readiness in another.

### Revenue and profitability

Unitree's filing reports rapid growth as both quadruped volume and humanoid sales increased:

| Period | Revenue | Reporting note |
|---|---:|---|
| 2023 | 159.13 million yuan | Audited reporting period in the prospectus |
| 2024 | 392.77 million yuan | Audited reporting period in the prospectus |
| 2025 | 1.699 billion yuan | Audited reporting period in the prospectus |

For 2025 the filing gives net profit of 278.21 million yuan and profit attributable to shareholders after excluding non-recurring items of 590.75 million yuan. The unusually large difference primarily reflects 349.07 million yuan of share-based compensation classified in the filing as a non-recurring loss. These two profit measures answer different accounting questions and should not be substituted for each other.[1]

Product mix changed during the same period. Humanoids accounted for 1.88 percent of main-business revenue in 2023, 27.68 percent in 2024, and 51.78 percent in 2025. Average recognized selling prices fell for both quadrupeds and humanoids as volumes rose and the mix shifted toward different models. That supports a story of manufacturing scale and a changing catalog, but it does not by itself establish future demand, field reliability, or a durable margin.[1]

### Proposed public offering

The Shanghai Stock Exchange accepted Unitree's STAR Market application in March 2026. The listing committee passed the application in June, and the China Securities Regulatory Commission approved registration of the initial public offering on July 1, 2026. The approval was valid for 12 months and required the offering to follow the submitted prospectus and underwriting plan. Registration approval is one stage in an offering rather than a statement about robot capability or an assurance of future financial performance.[10]

The [Unitree Robotics IPO](https://aiwiki.ai/wiki/unitree_ipo_2026) article covers the proposed share count, planned use of proceeds, review history, capital structure, and offering status. Keeping that material separate prevents a fast-changing securities process from overwhelming the company article.

## Market position and competition

Independent estimates agree that Unitree ended 2025 as one of the two largest humanoid-robot makers by volume, but they disagree about the order because they count different baskets. Omdia's "General Purpose Embodied Robot Market Radar," as reported by the trade outlet Gasgoo in January 2026, put global humanoid shipments at roughly 13,000 units in 2025 and ranked [AgiBot](https://aiwiki.ai/wiki/agibot) first at about 5,100 units, Unitree second at about 4,200 units, and [UBTech](https://aiwiki.ai/wiki/ubtech) third. Counterpoint Research, cited in the same report, counted roughly 16,000 global humanoid installations in 2025, with China accounting for more than 80 percent of them and Chinese companies occupying four of the top five positions. In January 2026 Unitree said it had shipped more than 5,500 "pure" humanoid robots in 2025 and produced more than 6,500, a count that excludes wheeled, dual-arm, and other configurations; on that definition Unitree, not AgiBot, would rank first, and the statement is consistent with the 5,511 humanoid shipments reported in its prospectus. Shipments, installations, and recognized sales are different measures, the analysts' category definitions differ, and, as with the quadruped estimates discussed above, shares from different reports cannot be combined into a single ranking.[1][15]

The main Chinese rivals compete from different directions. Shanghai-based AgiBot, also known as Zhiyuan Robotics, said its 10,000th robot left the production line on March 30, 2026, roughly three months after its 5,000th, and described deployments in logistics, retail, hospitality, education, and manufacturing across several product series; the release, like Unitree's own announcements, is an issuer statement rather than an audited count.[16] UBTech, a Hong Kong-listed Shenzhen company, has concentrated on industrial humanoids: it announced in November 2025 that mass production and delivery of the Walker S2 had begun with a first batch of several hundred units, that Walker-series orders since early 2025 exceeded 800 million yuan, and that industrial partners included BYD, Foxconn, Geely Auto, and SF Express, with a stated target of 5,000 units of annual capacity in 2026.[17] [EngineAI](https://aiwiki.ai/wiki/engineai), a Shenzhen startup that develops its own powered joints, priced its 1.38-meter PM01 humanoid at 88,000 yuan in December 2024 and said it had delivered nearly 100 units across three models in five months.[18] [Galbot](https://aiwiki.ai/wiki/galbot_g1), based in Beijing, sells wheeled dual-arm robots that operate small unmanned retail stores, launched its first humanoid retail robot in March 2025, and raised 1.1 billion yuan (about 153 million US dollars) in a June 2025 round led by battery maker CATL, which Yicai reported brought its fundraising over the prior two years above 2.4 billion yuan.[19]

Price competition has been explicit. Unitree's R1, announced on July 25, 2025 at a starting price of 39,999 yuan (about 5,900 US dollars), undercut the company's own G1, which starts at 99,000 yuan, as well as rival machines such as EngineAI's 88,000-yuan PM01 and a 299,000-yuan research humanoid UBTech released in March 2025. The South China Morning Post described the launch as part of a race among Chinese makers to sell cheaper humanoid robots.[20] Unitree's filings show the same pressure from the revenue side: recognized average selling prices for both quadrupeds and humanoids fell during 2023-2025 as volumes rose and the model mix changed.[1]

Forecasts published in 2026 still placed Unitree in the leading pair. TrendForce projected in April 2026 that China's humanoid robot output would grow about 94 percent in 2026 and that Unitree and AgiBot together would take nearly 80 percent of shipments in that market.[21] These are projections built on the same kinds of inconsistent shipment definitions described above, not measurements, and they concern unit volume rather than revenue, capability, or reliability.

## Safety, security, and policy scrutiny

### Product-use boundaries

Legged robots combine mass, fast joints, batteries, cameras, microphones, network interfaces, and software that can affect the physical world. Safe operation depends on the robot, payload, firmware, environment, network, and human procedures. A model's ability to recover from a push or fall is not a certification for unrestricted operation near people.

Unitree's online disclaimer for its robot-dog series tells buyers to follow manuals, keep children and vulnerable people away, avoid unauthorized modification, comply with export controls, and not use the product for terrorism, nuclear facilities, biological or chemical weapons, or other military purposes that jeopardize national security. This is a user-policy statement. It does not independently verify where every unit is used or prevent a downstream buyer from modifying hardware.[11]

Unitree also operates a Security Response Center that accepts vulnerability reports, publishes a severity and reward framework, and provides a security contact and PGP key. The existence of a disclosure program is a useful security-process fact, but it is not equivalent to a published independent audit of every robot or cloud service.[12]

### United States government actions

In May 2025, members of the U.S. House Select Committee on the Chinese Communist Party sent a letter to the Departments of Defense and Commerce and the Federal Communications Commission. The lawmakers alleged military-civil-fusion, procurement, and cybersecurity risks and asked the agencies to investigate Unitree and consider several designations. The document is evidence of the committee's allegations and requests; it is not an adjudication of every factual assertion in the letter.[13]

The U.S. Department of Defense subsequently included "Hangzhou Yushu Technology Co., Ltd. (Unitree)" in a Section 1260H notice published in the Federal Register on June 10, 2026. The notice states that the Deputy Secretary of Defense determined that Unitree met the statutory definition of a "Chinese military company" operating directly or indirectly in the United States. For Unitree, the department stated that it found indirect ownership by and affiliation with China's State-Owned Assets Supervision and Administration Commission and treated the company's "Little Giant" designation as military-civil-fusion assistance.[14]

That is a U.S. government determination under a specific statute. It should not be rewritten as proof that every Unitree robot is military equipment or that every customer is a military user. The same notice provides a process by which listed entities may request reconsideration. Unitree's prospectus, meanwhile, identifies Wang as the controlling shareholder and provides a detailed pre-offering shareholder table but does not describe SASAC as the controlling owner. The two records use different legal and analytical frameworks, and the public materials reviewed here do not resolve the Department of Defense's asserted indirect-ownership chain.[1][14]

Unitree's civilian branding and product-use restrictions do not erase the designation, and the designation does not erase the company's civilian sales, research customers, or stated restrictions. A neutral account keeps all three records distinct: what the company says its products are for, what lawmakers alleged, and what the Department of Defense formally determined.

## How to interpret Unitree claims

Unitree occupies several roles at once: robot manufacturer, component developer, research-platform vendor, consumer seller, systems supplier, and embodied-model researcher. Statements about one role do not automatically transfer to another. The following distinctions prevent common errors:

- **Platform capability versus shipped feature:** a research team may add a policy, computer, sensor, hand, or payload that is not included in the standard product.
- **Peak demonstration versus operating envelope:** top speed, torque, jump height, and complex choreography do not report sustained duty, failure rate, or safety.
- **Balance autonomy versus task autonomy:** learned low-level control can stabilize a machine while a person or fixed program still chooses the task.
- **Shipment versus deployment:** a shipped research robot may never enter a production workflow, while one deployed robot can accumulate many operating hours.
- **Market estimate versus audited result:** revenue and recognized sales in a securities filing have a different evidentiary basis from consulting estimates of a loosely defined global market.
- **Company claim versus independent result:** Unitree's statements describe its position; academic experiments, regulatory records, and customer deployments answer different questions.

As of July 28, 2026, the strongest supported characterization is that Unitree had achieved substantial commercial scale in legged-robot hardware and that independent researchers used its commercially available machines as development platforms. Its 2025 interim disclosures also show that humanoid revenue was concentrated in research, education, and commercial presentation; general embodied-model operation was not deployed at scale; product functions varied by configuration; and market-share comparisons remained methodologically uncertain.[1][3][4]

## References

1. Yushu Technology Co., Ltd. "Prospectus for the Initial Public Offering and Listing on the STAR Market (registration draft)." Shanghai Stock Exchange, June 2, 2026. https://static.sse.com.cn/stock/disclosure/announcement/c/202606/002178_20260602_SU4G.pdf
2. Unitree Robotics. "About Unitree." Company profile and development timeline, accessed July 28, 2026. https://www.unitree.com/about/
3. Yushu Technology Co., Ltd. "Response to the First-Round Inquiry on the Application for an Initial Public Offering and STAR Market Listing." Shanghai Stock Exchange, March 20, 2026. https://static.sse.com.cn/stock/disclosure/announcement/c/202603/002178_20260320_OE27.pdf
4. Yushu Technology Co., Ltd. "Response to the Second-Round Inquiry on the Application for an Initial Public Offering and STAR Market Listing." Shanghai Stock Exchange, March 20, 2026. https://static.sse.com.cn/stock/disclosure/announcement/c/202603/002178_20260320_BLK4.pdf
5. Unitree Robotics. "Unitree As2-W." Official product page, accessed July 28, 2026. https://www.unitree.com/mobile/As2-W/
6. Unitree Robotics. "unitree_sdk2." Official source repository, accessed July 28, 2026. https://github.com/unitreerobotics/unitree_sdk2
7. Unitree Robotics. "unitree_rl_gym." Official reinforcement-learning repository, accessed July 28, 2026. https://github.com/unitreerobotics/unitree_rl_gym
8. Guillaume Bellegarda, Yiyu Chen, Zhuochen Liu, and Quan Nguyen. "Robust High-Speed Running for Quadruped Robots via Deep Reinforcement Learning." IEEE/RSJ International Conference on Intelligent Robots and Systems, 2022. https://arxiv.org/abs/2103.06484
9. Tong Zhang et al. "HuB: Learning Extreme Humanoid Balance." Proceedings of the 9th Conference on Robot Learning, PMLR 305, 2025. https://proceedings.mlr.press/v305/zhang25b.html
10. China Securities Regulatory Commission. "Approval of Yushu Technology Co., Ltd.'s Registration Application for an Initial Public Offering." July 1, 2026. https://www.csrc.gov.cn/csrc/c105906/c7642867/content.shtml
11. Unitree Robotics. "Disclaimer and Warning in Use of Unitree Robotics Robot Dog Series Products." Accessed July 28, 2026. https://shop.unitree.com/en-fr/pages/disclaimer
12. Unitree Robotics. "Unitree Security Response Center." Accessed July 28, 2026. https://security.unitree.com/
13. U.S. House Select Committee on the Chinese Communist Party. "Letter to Secretary Hegseth, Secretary Lutnick, and Chairman Carr." May 6, 2025. https://selectcommitteeontheccp.house.gov/sites/evo-subsites/selectcommitteeontheccp.house.gov/files/evo-media-document/Unitree.pdf
14. U.S. Department of Defense. "Notice of Availability of Designation of Chinese Military Companies." Federal Register document 2026-11571, published June 10, 2026. https://www.federalregister.gov/d/2026-11571
15. Gasgoo. "Annual Champion Changes Hands? Unitree Announces Over 5,500 Humanoid Robot Shipments." January 24, 2026. https://autonews.gasgoo.com/articles/news/annual-champion-changes-hands-unitree-announces-over-5500-humanoid-robot-shipments-2014711162140991489
16. AgiBot (Zhiyuan Robotics). "AGIBOT Reaches 10,000 Units as Real-World Demand for Robots Accelerates." March 30, 2026. https://www.agibot.com/article/231/detail/53.html
17. UBTECH Robotics. "UBTECH Humanoid Robot Walker S2 Begins Mass Production and Delivery, with Orders Exceeding 800 Million Yuan." PR Newswire, November 17, 2025. https://www.prnewswire.com/news-releases/ubtech-humanoid-robot-walker-s2-begins-mass-production-and-delivery-with-orders-exceeding-800-million-yuan-302616924.html
18. The Robot Report Staff. "EngineAI releases PM01 humanoid robot for commercial, educational use." December 27, 2024. https://www.therobotreport.com/engineai-releases-pm01-humanoid-robot-for-commercial-educational-use/
19. Yicai Global. "CATL Heads Up USD153 Million Investment in Chinese Robotics Startup Galbot." June 24, 2025. https://www.yicaiglobal.com/news/chinese-robotics-startup-galbot-bags-usd153-million-in-latest-fundraiser
20. South China Morning Post. "China's Unitree debuts US$5,900 humanoid robot in race to make cheaper products." July 25, 2025. https://www.scmp.com/tech/tech-trends/article/3319637/chinas-unitree-debuts-us5900-humanoid-robot-race-make-cheaper-products
21. TrendForce. "China's Humanoid Robot Output to Surge 94% in 2026; Unitree and AgiBot to Capture Nearly 80% Market Share." April 9, 2026. https://www.trendforce.com/presscenter/news/20260409-13007.html