Chinese Humanoid Robots

Overview: H1 is Unitree’s larger humanoid line and one of the robots that made Hangzhou visible in the global humanoid race. It is built as a full-size biped platform rather than a small desktop research toy.

Technical notes: Unitree lists H1 around 180 cm and about 47 kg, with 3D LiDAR plus depth camera perception and a quickly replaceable 864 Wh battery. H1-2 is listed separately with a heavier body and a higher degree-of-freedom configuration.

Main application: Full-size humanoid research, demonstrations and locomotion development.

Mobility: Bipedal full-size humanoid

Autonomy level: Controlled demos, developer operation and research use. Do not treat public locomotion videos as proof of daily factory autonomy.

Why it matters: H1 matters because it gave Unitree a full-size humanoid body while the company already had experience manufacturing legged robots and compact actuators.

Limitations: Locomotion demos do not prove autonomous industrial work. The page keeps H1 as an active platform, not a confirmed general-purpose worker.

Overview: G1 changed the economics of humanoid robotics because it brought a compact bipedal humanoid body into a price band that universities, labs and developers can discuss seriously.

Technical notes: Unitree lists G1 at 1320 mm standing height and about 35 kg with battery. The base configuration has 23 joint degrees of freedom while the EDU configuration can expand to higher joint counts and optional force-control three-finger hands.

Main application: Research, development, education, demonstrations and embodied AI experiments.

Mobility: Compact bipedal humanoid

Autonomy level: Developer platform and controlled task demonstrations. It is not presented here as a proven home worker.

Why it matters: G1 gives researchers a body for imitation learning, teleoperation, fall recovery, manipulation and whole-body control without the cost profile of large industrial humanoids.

Limitations: A low sticker price does not include integration, spare parts, safety work, supervision and task engineering. It should be treated as a platform, not a finished worker.

Overview: R1 is included because Unitree lists it inside its humanoid product family and because it fits the company’s strategy of making humanoid bodies cheaper and easier to access.

Technical notes: The currently crawled official R1 page exposes product navigation and app support more clearly than detailed hardware values. The article therefore avoids quoting height, weight, battery or payload figures for R1.

Main application: Lower-cost bipedal humanoid platform and developer entry point.

Mobility: Bipedal humanoid

Autonomy level: Not publicly confirmed for independent industrial work in this article. Treated as a platform under active iteration.

Why it matters: R1 matters if it continues Unitree’s lower-cost platform direction. That would put more humanoid bodies into schools, labs and small robotics teams.

Limitations: Public specifications are treated as limited in this page. Any exact performance claims should be checked against the latest Unitree product sheet.

Overview: Walker X represents UBTECH’s service humanoid branch. It is separate from Walker S industrial robots and should not be treated as the same deployment category.

Technical notes: The robot sits in UBTECH’s commercial humanoid products, where the focus is user interaction, service settings and public-facing demonstrations rather than heavy factory handling.

Main application: Commercial service, exhibition, interaction and human-facing applications.

Mobility: Bipedal humanoid service robot

Autonomy level: Service and interaction demos. Not presented as unsupervised public-space autonomy.

Why it matters: Walker X shows how UBTECH built a humanoid brand before the more industrial Walker S line became central to its factory narrative.

Limitations: Public service demos are not enough to claim unsupervised operation in hospitals, airports or malls.

Overview: Walker S is one of the most important Chinese industrial humanoid lines because UBTECH is pushing it into factories, logistics workflows and aerospace testing partnerships.

Technical notes: UBTECH describes Walker S2 as an industrial humanoid with battery swapping within three minutes, power management, ground-reaching movement and binocular stereo vision. S1 is positioned for multi-task industrial scenarios.

Main application: Industrial manufacturing, logistics, material handling and factory pilot tasks.

Mobility: Bipedal industrial humanoid

Autonomy level: Industrial pilot and task-specific automation. Airbus described the aerospace collaboration as early testing, so this is not written as mass replacement.

Why it matters: Walker S matters because it forces a harder question than stage demos: can a humanoid work repeatedly in manufacturing cells without blocking humans or production lines?

Limitations: Factory pilots and purchase agreements remain different from proven daily deployment at scale. The article labels the series as pilot/commercial, not fully autonomous factory labor.

Overview: UWorld U1 sits outside the industrial Walker S line. It belongs to the social, companion and lifelike humanoid branch of UBTECH’s work.

Technical notes: Because public discussion around U1 focuses heavily on emotional response, lifelike presentation and pricing, the article avoids treating it as a factory robot or a proven household assistant.

Main application: Lifelike companion robot series and emotional interaction concept.

Mobility: Android-style humanoid / companion form factor

Autonomy level: Interaction-oriented AI claims, not independent household work.

Why it matters: U1 shows that China’s humanoid ecosystem is not only about factories. It also includes social robotics, companionship and the hard ethics of lifelike machines.

Limitations: Companion claims need caution. Emotional interaction is not the same as practical household manipulation or independent care.

Overview: GR-1 is the robot that made Fourier’s move from rehabilitation robotics into humanoids visible. It connected Shanghai’s rehab robotics base to embodied AI and full-body platforms.

Technical notes: Fourier’s official site groups GR-1, GR-2 and GR-3 under its GRx humanoid robot series. The page avoids model-specific numbers unless the official product source is clear.

Main application: General-purpose humanoid research platform and transition from rehabilitation robotics into humanoids.

Mobility: Bipedal humanoid

Autonomy level: Developer and research platform; not presented as independent mass-deployed worker.

Why it matters: GR-1 matters because Fourier did not come from internet software. It came from medical and rehabilitation robotics, where actuation, safety and human contact already matter.

Limitations: Research deliveries and demos do not mean generalized industrial deployment. Treat GR-1 as a platform.

Overview: GR-2 is presented as the upgraded middle generation in Fourier’s GRx series. It keeps the body format while shifting toward stronger developer access and more capable manipulation experiments.

Technical notes: Where official pages do not expose every parameter in crawlable text, this page keeps the hardware note short and points readers back to Fourier’s product pages.

Main application: Humanoid platform for research, service experiments and embodied AI development.

Mobility: Bipedal humanoid

Autonomy level: Developer and pilot-oriented. Not presented as fully autonomous.

Why it matters: GR-2 matters because China’s humanoid race is not only Shenzhen and Hangzhou. Shanghai has a cluster around Fourier, AgiBot and Kepler.

Limitations: Do not read marketing videos as evidence of high-duty industrial reliability.

Overview: GR-3 moves Fourier’s humanoid story toward care, assistance and human-facing interaction while still sitting inside the GRx hardware family.

Technical notes: The current site headline frames GR-3 as a caring and capable companion. The article treats that as positioning, not proof of independent caregiving.

Main application: Care, companion and assistance-oriented humanoid positioning with research and institutional interest.

Mobility: Bipedal humanoid

Autonomy level: Care and interaction positioning. Not written as independent eldercare replacement.

Why it matters: GR-3 shows how humanoid companies test different markets: factory work, research platforms, care settings and service interaction.

Limitations: Eldercare requires reliability, safety, consent, cleaning, privacy and supervision. The page does not claim those problems are solved.

Overview: AgiBot A2 is one of the central names in Shanghai’s humanoid push. The company connects robots, data collection, teleoperation and embodied AI infrastructure.

Technical notes: AgiBot’s own product navigation separates A2 Ultra, A2 Lite and A2-W. A2-W is framed around flexible manufacturing, while A2 variants cover full-size humanoid use and performance positioning.

Main application: Full-size humanoid platform, flexible manufacturing and embodied AI product family.

Mobility: Bipedal and wheeled variants depending on model

Autonomy level: Controlled demos, teleoperation data collection and task-specific industrial work. Not presented as full autonomy.

Why it matters: AgiBot matters because it is building the data side of humanoid robotics as aggressively as the hardware side.

Limitations: Public videos and Guinness-style endurance feats are not the same as verified factory productivity. The page separates demonstrations from deployment claims.

Overview: A3 is included as a recent AgiBot humanoid platform because the company has publicly shown it alongside A2 and X2 systems in international B2B contexts.

Technical notes: Until a stable official A3 product page exposes full technical details, this article avoids hard numbers and focuses on platform role.

Main application: Full-size humanoid platform for interaction, locomotion and manipulation demos.

Mobility: Bipedal humanoid

Autonomy level: Public B2B demonstrations. Not described as fully autonomous worker.

Why it matters: A3 shows that AgiBot is expanding its humanoid lineup rather than relying on one form factor.

Limitations: Current public demos should be treated as demonstrations unless a customer deployment and task scope are documented.

Overview: X1 and X2 are included because they connect AgiBot’s hardware work to open-source tooling and embodied AI experimentation.

Technical notes: The company’s own navigation separates X1 and X2 from the A2 family. This page does not merge them into the industrial A2 line.

Main application: Open-source robot work, smaller humanoid or agile embodied AI development depending on model.

Mobility: Humanoid / agile robot platform depending on model

Autonomy level: Research, developer ecosystem and controlled tasks; exact autonomy varies by model.

Why it matters: Open-source bodies and tools can widen China’s developer base for embodied AI, especially when paired with datasets and teleoperation.

Limitations: Exact form factor, customer status and deployment claims should be checked model by model before reuse.

Overview: IRON is XPeng’s humanoid bet. It matters because it ties robotics to an EV company with batteries, motors, factories, AI chips and retail environments.

Technical notes: IRON should be described through dated sources because XPeng has shown different public iterations. The article labels it as prototype / pilot rather than a mature factory worker.

Main application: Physical AI, retail, factory and corporate environment pilots linked to XPeng’s EV ecosystem.

Mobility: Bipedal humanoid

Autonomy level: Prototype and pilot strategy. Not independent mass deployment.

Why it matters: XPeng can test humanoids in places it already controls: showrooms, factories and vehicle supply chains. That gives IRON a plausible pilot path.

Limitations: A lifelike walk and onstage reveal do not prove uptime, manipulation reliability or return on investment. Public balance failures and staged conditions should remain part of the analysis when relevant.

Overview: Kuavo is Leju’s full-size humanoid line and a key part of the Shenzhen humanoid cluster alongside UBTECH and EngineAI.

Technical notes: Leju’s older portfolio includes small education robots and bipedal platforms, while Kuavo represents the larger humanoid direction.

Main application: General-purpose humanoid research, education and industrial exploration.

Mobility: Bipedal humanoid

Autonomy level: Research, demos and industrial exploration. Not claimed as autonomous factory replacement.

Why it matters: Leju matters because it has experience with educational humanoids and is moving toward larger industrial bodies.

Limitations: The article does not quote exact Kuavo specs unless the official product sheet is available.

Overview: AELOS and PANDO show the education side of Leju’s history. They are not the same category as a full-size industrial humanoid.

Technical notes: The page keeps these robots in the research and education class because small humanoids are useful for teaching gait, control and robotics programming.

Main application: Education, research, competitions and smaller humanoid development.

Mobility: Small bipedal humanoid robots

Autonomy level: Education and programmable research use.

Why it matters: Educational humanoids build the talent base that later supports larger platforms.

Limitations: Small education robots should not be used as evidence for industrial humanoid deployment.

Overview: EngineAI is part of Shenzhen’s second wave of humanoid startups. Its systems became visible through unusually smooth walking and bipedal motion demonstrations.

Technical notes: SA01 is best treated carefully as a bipedal robot platform. SE01 is the fuller humanoid entry. PM01 is included only as a platform in the public lineup, with exact details left to official sheets.

Main application: Bipedal and humanoid robot platforms for motion, demonstration, research and emerging service scenarios.

Mobility: SA01 bipedal robot; SE01 full-size humanoid; PM01 compact humanoid class if confirmed by product material.

Autonomy level: Controlled demos and platform development. Not described as proven autonomous labor.

Why it matters: EngineAI matters because walking quality, balance recovery and motion control remain core bottlenecks for humanoids.

Limitations: High-motion demos, boxing demos and public stunts do not prove useful work.

Overview: Astribot S1 became known for fast, smooth manipulation videos. The important part is not just speed, but the data and teleoperation pipeline behind the work.

Technical notes: The available research frames Astribot around whole-body manipulation, human demonstrations and policy learning. This page does not treat viral clips as final deployment proof.

Main application: Whole-body manipulation, teleoperation, household-style tasks and embodied AI data collection.

Mobility: Upper-body / humanoid manipulation system depending on configuration

Autonomy level: Teleoperation and learned policies. Not full autonomy.

Why it matters: Manipulation is where humanoids become useful or fail. Astribot sits directly in that problem.

Limitations: Public videos can hide resets, selected takes and human control. The page labels Astribot as manipulation-focused research, not as a home robot.

Overview: Kepler is included because it is one of the Shanghai humanoid companies often discussed alongside Fourier and AgiBot.

Technical notes: The article keeps Kepler in a watchlist/active development label and avoids hard specs when official public data is not stable in the current crawl.

Main application: Humanoid robot development for research, service and industrial scenarios depending on model.

Mobility: Bipedal humanoid platform

Autonomy level: Watchlist. No broad deployment claim in this page.

Why it matters: Kepler adds to the density of Shanghai’s humanoid cluster.

Limitations: Model names and exact specs should be verified from current official product sheets before reuse.

Overview: CyberOne matters as a signal that Xiaomi, one of China’s biggest consumer electronics companies, tested humanoid robotics publicly.

Technical notes: CyberOne should be kept separate from Xiaomi’s current robotics software and VLA research work. It was a visible humanoid prototype, not a sold worker.

Main application: Prototype humanoid demonstration linked to Xiaomi’s robotics and consumer electronics ecosystem.

Mobility: Bipedal humanoid prototype

Autonomy level: Prototype demonstration, not commercial humanoid deployment.

Why it matters: It connected humanoid robotics to consumer electronics supply chains, cameras, batteries, edge AI and brand-level R&D.

Limitations: No current large public deployment path is assumed.

Overview: XR-1 belongs to an earlier wave of cloud-connected humanoid service robots. It is included for context, but its current commercial status is not treated as active without confirmation.

Technical notes: The page uses a careful legacy/watchlist label because CloudMinds’ wider work includes cloud robot infrastructure and service robotics, not only a single humanoid body.

Main application: Cloud-connected service humanoid, reception and interaction scenarios.

Mobility: Humanoid service robot

Autonomy level: Cloud-assisted service robotics, not independent general autonomy.

Why it matters: XR-1 shows that China’s humanoid ecosystem had service and cloud-robot concepts before the 2023-2026 bipedal wave.

Limitations: Current availability, fleet status and deployment details are not assumed.

Overview: Tiangong is important because it connects Beijing’s public robotics strategy to humanoid platforms and open ecosystem work.

Technical notes: Humanoid races and endurance events reveal both progress and limitations: battery swaps, falls, resets and human support remain important details.

Main application: Open research platform, public robotics programs and endurance demonstrations.

Mobility: Bipedal humanoid

Autonomy level: Research and event demonstrations; not generalized deployment.

Why it matters: It shows how China uses public platforms and events to push robotics visibility beyond individual startups.

Limitations: Race completion does not equal useful industrial deployment.

Overview: Booster is included in watchlist status because it appears in China’s humanoid and robotics competition ecosystem, but the page avoids classifying every model as a main humanoid without stable source details.

Technical notes: The table keeps Booster outside the confirmed main industrial humanoid group until official product material is reviewed model by model.

Main application: Research, education, robotics competitions and humanoid-related development if product documentation confirms classification.

Mobility: Humanoid or bipedal platform depending on model

Autonomy level: Watchlist only in this page.

Why it matters: Competition and education robots can become important developer platforms even when they are not factory humanoids.

Limitations: No unverified robot specs or deployment claims are added.

Overview: PaXini appears in humanoid and tactile robotics discussions, but this page keeps TORA in watchlist status because product status and humanoid classification need current official confirmation.

Technical notes: The page intentionally avoids specs and autonomy claims here.

Main application: Humanoid-related robotics and tactile intelligence watchlist.

Mobility: Humanoid status should be confirmed by official material before full classification.

Autonomy level: Watchlist only.

Why it matters: Tactile intelligence and dexterous manipulation are major humanoid bottlenecks, which makes PaXini worth tracking.

Limitations: Watchlist means not treated as a confirmed deployed humanoid product.

Overview: Galbot is included because embodied AI and mobile manipulation overlap with humanoid robotics, especially around data and dexterity.

Technical notes: The page labels it as watchlist rather than a confirmed humanoid robot company to avoid mixing non-humanoid mobile manipulators into the humanoid table.

Main application: Mobile manipulation, embodied AI and robot data pipelines if confirmed by product line.

Mobility: Mobile manipulation / humanoid-related watchlist

Autonomy level: Watchlist only.

Why it matters: Mobile manipulation may solve some useful tasks before bipedal humanoids do. That makes Galbot relevant but not identical to Unitree or UBTECH.

Limitations: No humanoid specs are claimed.

Overview: TRON is tracked because humanoid development often starts with legs, balance and locomotion before full hands and upper-body manipulation arrive.

Technical notes: The page keeps TRON as bipedal/legged unless an official full humanoid upper body is confirmed.

Main application: Legged locomotion research and robot mobility platform.

Mobility: Bipedal / legged platform; not listed as full humanoid unless upper-body configuration is official.

Autonomy level: Watchlist only for humanoid relation.

Why it matters: Locomotion platforms influence humanoid control stacks and reinforcement learning work.

Limitations: Not counted as a confirmed full humanoid robot in the main classification.

Overview: Xianxingzhe is included to avoid making China’s humanoid story look as if it began in 2023. It belongs to an earlier university and defense-research period of bipedal humanoid work.

Technical notes: This page keeps the entry historical and does not add modern performance claims.

Main application: Early Chinese bipedal humanoid research.

Mobility: Bipedal humanoid research platform

Autonomy level: Historical research platform.

Why it matters: It gives the timeline a deeper base before the current commercial rush.

Limitations: The page does not treat Xianxingzhe as a current product.