XuanChuang Robot makes teleoperation the deployment plan

A remote operator station linked to an industrial inspection robot.Illustration: generated for this article

XuanChuang Robot's integrated inspection-operation robot is already doing industrial work through teleoperation while it collects the data meant to make later autonomy possible. According to the report, the company is aimed at hazardous oil, gas, and chemical sites, and its orders on hand have exceeded 100 million yuan. It has also signed an order for more than 100 units with the China National Petroleum Corporation Xinjiang Oilfield system, completed the first batch delivery, and is continuing deliveries at 20 units per month.

That makes teleoperation part of the deployment plan, not a stage trick.

The pattern is practical. XuanChuang Robot is training from real-scene data gathered by deployed robots, owner-shared operation data, its own simulation platform, a VLA execution model, and tests based on an open-source JEPA world model, the outlet reported. Its A1 financing is slated for data pipeline construction, VLA model and JEPA world model training and deployment, plus standardized product inventory and production. The autonomy targets sit later: routine operations before 2030, complex-scenario operations before 2035.

Teleoperation is already part of the product, not a demo crutch

Teleoperation is not sitting outside XuanChuang Robot's product plan as a temporary showroom trick. According to the report, its integrated inspection-operation product already performs tasks through remote operators while it collects data, with autonomy targets set for routine operations before 2030 and complex-scenario operations before 2035.

2026 financing sequence for embodied-AI data and deployment infrastructure

February 2026Mifeng Technology was registered
10 days laterMifeng Technology announced a several-hundred-million-yuan angel round
March 2026Guanglun Intelligence completed 1 billion yuan in financing
April 2026Wuwen Zhike completed more than 100 million yuan in financing
June 2026Noitom Robotics completed several hundred million yuan in financing

That sequence matters because XuanChuang is not aiming at generic office corridors. The company focuses on special embodied-intelligence robots for hazardous chemical industrial sites, including oil, gas, and chemicals, according to the report. These are constrained commercial environments, but they are not easy ones. A remote human can keep work moving while the system records how real tasks unfold around valves, equipment, routes, and exceptions.

The deployment is already commercial.

The report said that XuanChuang's orders on hand have exceeded 100 million yuan. It has also signed an order for more than 100 units with the China National Petroleum Corporation Xinjiang Oilfield system, completed the first batch delivery, and is continuing deliveries at 20 units per month. In that context, teleoperation becomes an operations model: ship bodies, serve a site, collect real-scene interaction data, then feed the next training loop.

The funding plan points in the same direction. XuanChuang plans to use its A1 financing mainly for data pipeline construction, VLA model and JEPA world model training and deployment, and standardized product inventory and production, according to the report. Its current model work combines data from deployed robots, owner-shared operation data, its own simulation platform, a VLA execution model, and tests based on an open-source JEPA world model.

Lingyu Intelligent shows the same practical pattern from another angle. The report says its business covers robot body R&D, real-machine data collection, and cloud operation platform construction, and that it has achieved scaled commercial deployment in open retail scenarios. Its Pre-A funds are earmarked for mass production capacity, upgrades to the real-machine data pipeline, and iteration of the cloud operation platform.

The common idea is simple: autonomy is being built inside paid deployment, not before it.

The data fight is about volume, labels, and portability

The data argument inside embodied AI is less about whether robots need data than what kind can travel. Noitom Robotics is betting on human-centric collection: according to woshipm, it has built data factories in several cities globally, including Shenzhen, with total space of more than 10000 square meters. Its target is a cross-embodiment reusable data system for physical AI and humanoid robots, meaning data should survive the jump from one body design to another.

That portability is the hard part.

Dai Ruoli's example, cited by woshipm, makes the contradiction concrete: data collected on Unitree G1 still needs complex processing before it can be used on Unitree's own H1, and the problem grows when the target becomes Xpeng's robot. The woshipm author argues that human-collected data has stronger cross-embodiment generalization than real-machine data because it is less tied to one robot's joints, sensors, and control stack. Lingyu Intelligent is moving in the other direction. According to a media report, its business covers robot body R&D, real-machine data collection, and cloud operation platform construction, and its Pre-A financing is planned for mass production capacity, upgrades to the real-machine data pipeline, and cloud platform iteration.

Both strategies exist because reusable robot data is still scarce. Woshipm cites a claim that as of early 2026, the world had only 500000 hours of usable high-quality real physical interaction data, far below a 10 million-hour entry threshold for general models. People's Daily, during WRC 2026, put China's compliant real-scenario data at only 500,000 hours and said the data gap for commercialization exceeded 99%. JD Cloud President Cao Peng's diagnosis fits that shortage: robot brains still lack generalization because real-scenario data is missing.

Few-shot models challenge the data-factory assumption

The hardest question for a teleoperation-led robot company is whether the operator is mainly doing useful work or feeding a data machine. The data-machine view has a brutal premise. woshipm cites a claim that, as of early 2026, the world had only 500000 hours of usable high-quality real physical interaction data, far below the 10 million-hour entry threshold for general models.

People's Daily, as relayed by woshipm during WRC 2026, framed the Chinese gap in similar terms: compliant real physical-interaction data in China totaled only 500,000 hours, while robot commercialization requires tens of millions of hours. The reported gap exceeded 99%. If that is the binding constraint, teleoperation becomes an industrial data collection strategy first and a labor-saving product second.

Zhongke Fifth Era attacks that assumption from the other side.

According to a Chinese tech outlet, the company builds products around embodied manipulation large models and launched the FAM series, described as an ultra-few-shot embodied manipulation model. Its claim is not simply that more robot data helps. It says the model can learn a task from only 3 to 5 real demonstrations, reach a basic task success rate of up to 97%, and cut data requirements to 1% of traditional methods.

That changes how a deployment plan is judged. In March, the outlet says Zhongke Fifth Era completed a real-world validation of general tote handling at the headquarters of a global manufacturing company in Germany. The robot continuously identified, grasped, and transported randomly placed totes of different specifications without human intervention. For a warehouse-like task, the useful question becomes narrower: how much fresh site data is needed before the robot behaves reliably enough there?

FAM-1.3 adds another clue. The outlet says it integrates three-dimensional spatial structure priors with video generation. BridgeV2W is described as translating complex action instructions into future images to simulate action outcomes, while the unreleased FlowWAM ranked first on the WorldArena global overall score leaderboard in April. The strategic split is clear: build bigger data factories, collect better task data, or bet that models can need far less.

Robot bodies are deployment assumptions in hardware form

Robot bodies turn a deployment thesis into metal. Zhongke Fifth Era's first self-developed wheeled embodied robot shows one end of that logic: according to media reports, it has reached scaled deployment in logistics. It has reached scaled deployment in industrial flexible production lines. It has reached scaled deployment in energy inspection operations. The body is still humanoid enough to work around human-designed tasks: 189cm tall, a 65cm arm span, 28 degrees of freedom across the whole body, with humanoid arms that have 7 degrees of freedom.

That is a bet on reach and manipulation without bipedal walking.

Lingyu Intelligent makes the opposite cost argument. Its TA series uses a wheeled chassis and gripper solution instead of bipedal legs and dexterous hands, according to media reports. It also uses a current feedback algorithm instead of six-dimensional force sensors. The claimed result is a robot priced below one-half to one-third of comparable products, and for many sites, that means the body is not trying to imitate a worker. It is trying to make the unit economics tolerable enough for deployment.

XuanChuang Robot's form factors start from safety boundaries. Its inspection line includes wheeled configurations. It includes rail-mounted configurations. It includes tracked configurations. Media reports say all have passed dual certification for explosion-proofing and intrinsic safety. The company has also added air-ground collaboration, using drones with base stations to inspect high points such as towers, plus two-wheeled-legged robots for narrow passages and low spaces.

That hardware variety matches its control philosophy.

XuanChuang Robot's core technology is the self-developed AEGIS architecture: an upper layer for task understanding and decisions. A middle layer handles action simulation. A bottom layer can intercept unreasonable large-model instructions within milliseconds. Dr. Tao Jin's view, as reported by media reports, is that large models are intelligent but still cannot directly satisfy high-risk industrial reliability requirements. Fu Zhe said integrated inspection-operation is the next main direction, with dual-arm actuators directly completing operation tasks.

A practical deployment framework for embodied-AI robots

You need robots to create value now in high-risk industrial sites such as oil, gas, chemicals, energy inspection, or logistics, where reliability matters more than a flashy general-purpose demo. Favor constrained, purpose-built deployments with explicit safety architecture. XuanChuang Robot focuses on hazardous chemical industrial scenarios and uses AEGIS, a three-layer system in which the top layer handles task understanding and decisions, the middle layer simulates actions, and the bottom layer can intercept unreasonable large-model instructions within milliseconds. Beijing Humanoid Robot Innovation Center CTO Tang Jian also described industrial and special scenarios as the first stage of humanoid robot implementation.
You have a commercially meaningful task but autonomy is not mature enough for full closed-loop operation. Use teleoperation as a bridge, but treat it as both service delivery and data collection. XuanChuang Robot's integrated inspection-operation product currently performs tasks through teleoperation while collecting data, with goals of autonomous routine operations before 2030 and autonomous complex-scenario operations before 2035.
Your bottleneck is not robot hardware but the shortage or poor transferability of physical-interaction data. Optimize the data pipeline before scaling models blindly. XuanChuang Robot is collecting real-scene data from deployed robots, using owner-shared operation data, building its own simulation platform, deploying a VLA execution model, and testing an open-source JEPA world model. Noitom Robotics is taking a different route: Human-centric data collection and a cross-embodiment reusable data system, with data factories in several cities globally, including Shenzhen, totaling more than 10000 square meters.
You are choosing between gathering much more data and betting on models that need less data. Separate the two bets. Sources report that compliant real physical-interaction data in China totaled only 500,000 hours while commercialization requires tens of millions of hours, and that the data gap exceeded 99%. Against that backdrop, Zhongke Fifth Era claims its FAM series can learn a task from only 3 to 5 real demonstrations, achieve a basic task success rate of up to 97%, and reduce data requirements to 1% of traditional methods. Treat few-shot manipulation as a strategic alternative to brute-force data accumulation, not as a proven replacement across all scenarios.
The task can be solved by specialized automation rather than a human-shaped general robot. Do not over-specify embodied intelligence. In the barbecue skewer-making example, several robot manufacturers gave an optimistic estimate of "5 to 10 years" for robots to replace human workers, while a senior embodied-intelligence expert argued that robots could not perform full human replacement there, but automated meat-skewering equipment could. Use the simplest body and workflow that meets the job.
The robot works technically but creates a worse user workflow. Pause deployment and fix the product loop. The Yunji Technology apartment food-delivery robot could take the elevator autonomously, but the reported workflow had basic failures: it could not press the doorbell, had voice volume too low to be heard through a closed door, sometimes returned to the first floor with the food, did not show queue position, and peak-period delivery could take 1 hour. The lesson is to validate real demand, usable product, economic feasibility, and scalable replication before calling a deployment productive.

Product management decides whether autonomy matters

Product managers should treat autonomy as an ROI variable, not a badge. XuanChuang Robot is a useful example because its commercial story does not wait for full autonomy: according to the report, Fu Zhe said China National Petroleum Corporation has moved from last year's POC project procurement to batch procurement for XuanChuang products. He also said one chemical-enterprise deployment case could recover costs within one and a half years.

That is the bar: paid repetition and a payback clock.

The checklist starts with the fallback plan. XuanChuang's integrated inspection-operation product currently performs tasks through teleoperation while collecting data, according to the report, with goals of autonomous routine operations before 2030 and autonomous complex-scenario operations before 2035. For a buyer, that means the current product must be judged on remote labor cost, operator availability, response time, safety procedures and compliance fit. Autonomy upside matters only after the teleoperated service already works.

The second test is whether the scene is bounded tightly enough. Woshipm's author describes using a Yunji Technology food delivery robot in a Suzhou apartment: courier to front desk, front desk to robot, room number entry, autonomous elevator ride, door notification, resident pickup. The failure was mundane. The author says the robot could not press the doorbell, spoke too quietly through a closed door, sometimes returned to the first floor with the food, and did not show the resident the order queue. In peak periods, the trip from the first floor to the room could take 1 hour.

That is a product failure before it is a model failure.

Woshipm cites China Information Weekly at WRC 2026 on four tests for productivity: real demand, usable product, economic feasibility, scalable replication. Tang Jian's staged path puts industrial and special scenarios first, then commercial service scenarios, with home and generalized scenarios last. Xingdong Jiyuan's Xi Yue offered the deployment proof point: regular operations with SF Express and China Post in more than 10 logistics centers across 5 provinces and cities, with migration speed improved from two months to within one week.

So the practitioner question is narrow. What failure rate is acceptable, who takes over when autonomy stops, and does the constrained site produce repeatable work at a price the customer will buy again?

Deployment and data strategies across Chinese embodied-AI companies

Dimension	XuanChuang Robot	Lingyu Intelligent	Zhongke Fifth Era	Noitom Robotics
Primary deployment setting	Extremely hazardous chemical industrial scenarios such as oil, gas, and chemicals	Open retail scenarios	Logistics, industrial flexible production lines, and energy inspection operations	Data factories in several cities globally, including Shenzhen
Commercial status described in sources	Orders on hand have exceeded 100 million yuan; signed an order for more than 100 units with the China National Petroleum Corporation Xinjiang Oilfield system	Achieved scaled commercial deployment in open retail scenarios	Served more than ten core customers in China, including Sinopec Group, Leapmotor, and State Grid; overseas business has entered the actual delivery stage	Built data factories with a total space of more than 10000 square meters
Near-term autonomy pattern	Integrated inspection-operation product currently performs tasks through teleoperation while collecting data	Businesses cover robot body R&D, real-machine data collection, and cloud operation platform construction	Robot continuously identified, grasped, and transported randomly placed totes of different specifications without human intervention in a Germany validation	Focuses on Human-centric data collection and a cross-embodiment reusable data system
Data strategy	Collects real-scene data from deployed robots, uses owner-shared operation data, builds its own simulation platform, deploys a VLA execution model, and tests based on an open-source JEPA world model	Plans to continuously upgrade its real-machine data collection pipeline and iteratively develop its cloud operation platform	Claims its model can learn a task from only 3 to 5 real demonstrations and reduce data requirements to 1% of traditional methods	Argues for human-collected data with stronger advantages in cross-embodiment generalization than real-machine data
Robot-body choices	Inspection product line includes wheeled, rail-mounted, and tracked configurations; added air-ground collaboration and two-wheeled-legged robots	TA series robots use a wheeled chassis and gripper solution instead of bipedal legs and dexterous hands	First self-developed wheeled embodied robot is 189cm tall, has a 65cm arm span, has 28 degrees of freedom across the whole body, and has humanoid arms with 7 degrees of freedom	not covered
Safety or reliability mechanism	AEGIS uses a three-layer division of labor, with a bottom layer safety barrier that can intercept unreasonable large-model instructions within milliseconds	TA series robots use a current feedback algorithm instead of six-dimensional force sensors	FAM-1.3 deeply integrates three-dimensional spatial structure priors with video generation; BridgeV2W simulates action outcomes visually	not covered
Funding described in sources	Completed an A1 financing round of tens of millions of yuan	Recently completed a Pre-A financing round of several hundred million yuan; cumulative financing has reached several hundred million yuan	Completed A1 and A2 financing rounds with total financing exceeding 1 billion yuan; completed several billions of yuan in financing in less than two years since its founding	Completed several hundred million yuan in financing
Strategic bet implied by the source	Use constrained hazardous sites and teleoperation to do useful work now while collecting real-scene data for later autonomy	Lower-cost purpose-built bodies and retail deployment to scale real-machine data and cloud operations	Ultra-few-shot embodied manipulation models that need dramatically less data	Better reusable Human-centric data for cross-embodiment generalization
Longer-term autonomy target	Goals of achieving autonomous routine operations before 2030 and autonomous complex-scenario operations before 2035	not covered	not covered	not covered

Treat the autonomy claim as the least interesting part of a robotics pitch. Ask for the site boundary first: the aisle, dock, counter, or shift where the system is supposed to work, and what happens when it leaves that boundary.

Use China Information Weekly's WRC 2026 test as a checklist: real demand, usable product, economic feasibility, scalable replication. Then press on the data loop. woshipm cites People's Daily saying China's commercialization data gap exceeded 99%, while JD Cloud's Cao Peng tied weak robot generalization to missing real-scenario data. That makes every deployment a data contract as much as a labor contract.

Watch the narrow cases. Xingdong Jiyuan pointed to regular work with SF Express and China Post in more than 10 logistics centers across 5 provinces and cities, with migration moving from two months to within one week.

For readers outside China

Availability: The source material covers Chinese startups and some overseas activity, but it does not provide general purchasing availability for most systems. Zhongke Fifth Era has obtained several hundred million yuan in overseas orders, has overseas business in actual delivery, and plans to expand in Europe, Australia, New Zealand, Japan, and South Korea in the next six months. Hypershell is explicitly global: North America, Europe, and China are its three core markets, and Halo is scheduled to be officially released on September 3 at IFA in Berlin. Availability outside China for XuanChuang Robot, Lingyu Intelligent, Noitom Robotics, Xingdong Jiyuan, and Yunji Technology is not disclosed in sources.
Pricing: Most product pricing is not disclosed in sources. XuanChuang Robot has orders on hand exceeding 100 million yuan and signed an order for more than 100 units with the China National Petroleum Corporation Xinjiang Oilfield system, but unit pricing is not stated. Lingyu Intelligent claims its TA series robots are priced at below one-half to one-third of comparable products, but no actual price is given. Hypershell, Zhongke Fifth Era, Noitom Robotics, and the Yunji Technology robot examples do not include product prices in the source material. Financing figures are disclosed for some companies, including XuanChuang Robot's A1 round of tens of millions of yuan, Lingyu Intelligent's Pre-A round of several hundred million yuan, Zhongke Fifth Era's A1 and A2 rounds exceeding 1 billion yuan, Noitom Robotics' several hundred million yuan financing, and Hypershell's $50 million Series B+ financing round.
Closest Western equivalents: For XuanChuang Robot: industrial inspection and intervention robots for oil, gas, chemicals, and other hazardous sites, closer to specialized industrial robotics than to a general-purpose humanoid.; For Noitom Robotics: motion-capture and robot-data infrastructure, closest to a robotics data foundry or mocap-driven physical-AI data provider.; For Lingyu Intelligent: cost-optimized commercial service robots for open retail scenarios, using wheeled chassis and grippers rather than humanoid legs and dexterous hands.; For Zhongke Fifth Era: embodied manipulation-model and robot-stack companies working on warehouse, industrial production, logistics, and energy inspection manipulation.; For Hypershell: consumer wearable assistive exoskeletons for mobility and outdoor use.
Data residency: Data-residency and compliance details are mostly not disclosed in sources. Noitom Robotics has built data factories in several cities globally, including Shenzhen, with more than 10000 square meters of total space, but the sources do not say where customer data is stored or processed. XuanChuang Robot collects real-scene data from deployed robots and uses owner-shared operation data, but the sources do not disclose storage location, retention policy, or cross-border transfer rules. The broader Chinese discussion emphasizes scarcity of compliant real physical-interaction data: People's Daily reported during WRC 2026 that compliant data from real physical interaction scenarios in China totaled only 500,000 hours and that the data gap exceeded 99%.

Sources

woshipm 具身智能风口里，没有自由人 https://woshipm.com/embodied/6456049.html
36kr 硬氪独家 | 清华，中科院具身大脑企业融资10亿，已获得数亿海外订单 https://36kr.com/p/3947288204770693
woshipm 在具身智能行业，产品经理能干什么？ https://woshipm.com/embodied/6455670.html
36kr 拿下中石油体系超百台订单，这家公司为特危化场景提供"巡操一体"机器人丨36氪首发 https://36kr.com/p/3959929849642113
36kr 清华系具身智能企业获数亿元融资，机器人已在京东超市规模化落地岗｜硬氪首发 https://36kr.com/p/3948044825279621
geekpark 一年卖出 3 万台后，极壳将在 IFA 发布下一代外骨骼 https://geekpark.net/news/369478

The evidence: 92 facts from 6 Chinese articles

Each line below was extracted from the article it sits under, in Chinese, before any of this was written. The writing is done from these and never from the source prose - that separation is structural, not a promise. How we work.

36kr拿下中石油体系超百台订单，这家公司为特危化场景提供“巡操一体”机器人丨36氪首发

XuanChuang Robot has completed an A1 financing round of tens of millions of yuan.
The investors in XuanChuang Robot's A1 financing round include Qianhai Ark, Guangyang Co., Ltd., and Westlake Sci-Tech Innovation Investment.
XuanChuang Robot plans to use its A1 financing funds mainly for data pipeline construction, VLA model and JEPA world model training and deployment, and standardized product inventory and production.
XuanChuang Robot was founded in December 2022.
XuanChuang Robot focuses on the development and application of special embodied-intelligence robots for extremely hazardous chemical industrial scenarios such as oil, gas, and chemicals.
XuanChuang Robot's core team mainly comes from Harbin Institute of Technology.
XuanChuang Robot founder and CEO Fu Zhe has educational backgrounds at Harbin Institute of Technology, the University of Birmingham, the University of Sheffield, and Imperial College London.
XuanChuang Robot's core technology is its self-developed AEGIS system architecture.
AEGIS uses a three-layer division of labor: the upper layer understands tasks and makes decisions, the middle layer simulates actions, and the bottom layer acts as a safety barrier that can intercept unreasonable large-model instructions within milliseconds.
XuanChuang Robot is currently training models by collecting real-scene data from deployed robots, using owner-shared operation data, building its own simulation platform, deploying a VLA execution model, and testing based on an open-source JEPA world model.
XuanChuang Robot's product system has expanded from single inspection to parallel inspection and integrated inspection-operation product lines.
XuanChuang Robot's inspection product line includes wheeled, rail-mounted, and tracked configurations, all of which have passed dual certification for explosion-proofing and intrinsic safety.
XuanChuang Robot has added two configurations, air-ground collaboration and two-wheeled-legged robots; the former uses drones with base stations to inspect high points such as towers, and the latter targets narrow passages and low spaces.
XuanChuang Robot's integrated inspection-operation product currently performs tasks through teleoperation while collecting data, with goals of achieving autonomous routine operations before 2030 and autonomous complex-scenario operations before 2035.
XuanChuang Robot's orders on hand have exceeded 100 million yuan.
XuanChuang Robot has signed an order for more than 100 units with the China National Petroleum Corporation Xinjiang Oilfield system, completed the first batch delivery, and is continuing deliveries at a pace of 20 units per month.

36kr清华系具身智能企业获数亿元融资，机器人已在京东超市规模化落地岗｜硬氪首发

Lingyu Intelligent recently completed a Pre-A financing round of several hundred million yuan.
Lingyu Intelligent's recent Pre-A financing round was jointly invested by Zhuzhou Industrial Investment, Feitu Venture Capital, Future Frontier Venture Capital, Xineng Venture Capital and others.
Lingyu Intelligent plans to use the Pre-A financing funds mainly for expanding capacity for scalable mass production of robot bodies, continuously upgrading its real-machine data collection pipeline, and iteratively developing its cloud operation platform.
Maple Pledge Capital has long served as Lingyu Intelligent's private equity financing adviser.
Lingyu Intelligent has completed three financing rounds within this year.
Lingyu Intelligent's previous investors include Innoangel Fund, ChinaEquity Capital, Galaxy Innovation Capital, Futian Capital, Leaguer VC, and Tianying Capital.
Lingyu Intelligent's cumulative financing has reached several hundred million yuan.
Lingyu Intelligent was founded in early 2025.
Lingyu Intelligent focuses on building underlying infrastructure for embodied intelligence, with businesses covering robot body R&D, real-machine data collection, and cloud operation platform construction.
Lingyu Intelligent has achieved scaled commercial deployment in open retail scenarios.
Lingyu Intelligent co-founder and chief scientist Mo Yilin is a tenured associate professor in Tsinghua University's Department of Automation.
Lingyu Intelligent co-founder and chief scientist Mo Yilin studied under Richard M. Murray, a member of the U.S. National Academy of Engineering and a pioneer in robotic manipulation.
Lingyu Intelligent co-founder and chief scientist Mo Yilin has more than 10,000 Google Scholar citations.
Lingyu Intelligent co-founder and CEO Jin Ge holds a bachelor's degree from Tsinghua University's Department of Automation and an MBA from Tsinghua University's School of Economics and Management.
Lingyu Intelligent co-founder and CEO Jin Ge previously served as managing partner at Vision Plus Capital and vice president at Orbbec Photonics.
Most of Lingyu Intelligent's core management team members have undergraduate backgrounds from Tsinghua University.
Lingyu Intelligent's main team members have work experience at ByteDance, Kuaishou, Tencent, Meituan and other major technology companies.
Lingyu Intelligent's TA series robots use a current feedback algorithm instead of six-dimensional force sensors.
Lingyu Intelligent's TA series robots use a wheeled chassis and gripper solution instead of bipedal legs and dexterous hands.

36kr硬氪独家 | 清华，中科院具身大脑企业融资10亿，已获得数亿海外订单

Zhongke Fifth Era recently completed A1 and A2 financing rounds with total financing exceeding 1 billion yuan.
Zhongke Fifth Era's A1 financing round was jointly invested by Xigao Investment, Bank of China's BOC AIC fund, and Sanfeng Investment.
Zhongke Fifth Era's A2 financing round was jointly invested by Xinneng Venture Capital, BOC AIC fund, Zhongshan Venture Capital, Hongruida Investment, Beyondsoft, and Jinchuan Investment (Zhejiang Province New Energy Vehicle Sub-Fund).
Zhongke Fifth Era was founded in 2024.
Zhongke Fifth Era has completed several billions of yuan in financing in less than two years since its founding.
Zhongke Fifth Era builds products and solutions around embodied manipulation large models.
Zhongke Fifth Era has served more than ten core customers in China, including Sinopec Group, Leapmotor, and State Grid.
Zhongke Fifth Era has obtained several hundred million yuan in overseas orders, and its overseas business has entered the actual delivery stage.
Zhongke Fifth Era plans to focus on expanding overseas markets including Europe, Australia, New Zealand, Japan, and South Korea in the next six months.
In March, Zhongke Fifth Era completed a real-world validation of general tote handling at the headquarters of a global manufacturing company in Germany.
During Zhongke Fifth Era's tote-handling validation in Germany, a robot continuously identified, grasped, and transported randomly placed totes of different specifications without human intervention.
Zhongke Fifth Era founder and CEO Liu Nianfeng holds a doctorate from the Institute of Automation, Chinese Academy of Sciences, and studied under Chinese Academy of Sciences academician Tan Tieniu.
Zhongke Fifth Era's recently upgraded FAM-1.3 version deeply integrates three-dimensional spatial structure priors with video generation.
Zhongke Fifth Era's unreleased next-generation world model FlowWAM ranked first on the WorldArena global overall score leaderboard in April.
Zhongke Fifth Era's first self-developed wheeled embodied robot has achieved scaled deployment in logistics, industrial flexible production lines, and energy inspection operations.
Zhongke Fifth Era's first self-developed wheeled embodied robot is 189cm tall, has a 65cm arm span, has 28 degrees of freedom across the whole body, and has humanoid arms with 7 degrees of freedom.

geekpark一年卖出 3 万台后，极壳将在 IFA 发布下一代外骨骼

Hypershell has recently completed research and development for a multi-joint lower-limb exoskeleton product.
Hypershell's next-generation flagship product is named Halo and is scheduled to be officially released on September 3 at IFA in Berlin.
Hypershell was founded at the end of 2021.
Hypershell began mass production and delivery of its first-generation Hypershell X series in December 2024.
Hypershell delivered a cumulative total of 30,000 units worldwide in 2025, its first year of mass production.
North America, Europe, and China are Hypershell's three core markets.
Hypershell completed a Series B financing round in November last year, reaching a post-money valuation of nearly 2.7 billion yuan.
Hypershell announced a $50 million Series B+ financing round in May led by Ant Group and Meituan Long-Z.
Halo uses an integrated architecture with 4 assistive motors on the left and right sides of the hip and knee joints to provide full-leg assistance.
Halo's power unit has nearly doubled its power density compared with the previous-generation hip-joint power unit.
Halo has a folded form close to the Hypershell X series, making it suitable for storage and carrying.
Hypershell achieved a new end-to-end algorithm breakthrough in the first half of this year.
The first-generation mass-produced Hypershell X series can autonomously identify 12 different motion states using sensors and algorithms.
IDC data shows that domestic exoskeleton product shipments exceeded 26,000 units in 2025, and the overall market size exceeded 1.6 billion yuan.
IDC data shows that consumer assistive exoskeletons sold 19,000 devices in 2025, accounting for 73% of all exoskeleton shipments.
JD.com data shows that more than 10 exoskeleton robot brands have joined the platform, and transaction volume for the exoskeleton robot category grew by more than 15 times year on year in the first half of this year.
During the full 618 shopping festival cycle, Hypershell ranked among the top three smart robot brands and consistently ranked first in the wearable exoskeleton category.
Hypershell will open its first directly operated offline experience store in China at Shenzhen Bay MixC by the end of this month.

woshipm具身智能风口里，没有自由人

Dai Ruoli and Liu Haoyang founded the motion capture technology company Noitom Technology 14 years ago, and Dai Ruoli served as CTO.
Noitom Technology worked on film and television productions including Game of Thrones, Logan, and Star Trek: Discovery.
Noitom Technology reduced its team from 300 people to around 70 people after the pandemic.
Noitom Robotics has built data factories in several cities globally, including Shenzhen, with a total space of more than 10000 square meters.
Noitom Robotics focuses on Human-centric data collection and a cross-embodiment reusable data system for physical AI and humanoid robots.
In March 2026, Guanglun Intelligence completed 1 billion yuan in financing.
In April 2026, Wuwen Zhike completed more than 100 million yuan in financing.
In June 2026, Noitom Robotics completed several hundred million yuan in financing.
Mifeng Technology, a subsidiary of AgiBot, was registered in February 2026 and announced a several-hundred-million-yuan angel round 10 days later.
In autumn 2023, two well-known U.S. robotics companies contacted Noitom Technology to buy more than 100 sets of motion capture equipment for robot data collection.
Meta's Reality Labs division lost $13.7 billion in 2022.
Meta laid off nearly 10000 people in 2023.
Global shipments of VR headsets were about 4.75 million units in 2025, down about 35% year on year.
In December 2023, Mi Liangchuan, then head of Xpeng's robotics business unit, had a 2-hour phone call with Dai Ruoli and tried to persuade him to become a supplier for Xpeng's robot.
After the Spring Festival, Dai Ruoli formed a 19-person WeChat group for a robotics team to study the new business.
Noitom Robotics was spun out as an independent company from Noitom's robotics team, with Dai Ruoli serving as CEO.

woshipm在具身智能行业，产品经理能干什么？

The woshipm author experienced a Yunji Technology food delivery robot in an apartment in Suzhou during an entrepreneurial period.
The Yunji Technology apartment food delivery robot was designed for a process in which a courier handed takeout to the front desk, the front desk put the takeout into the robot and entered the room number, the robot took the elevator autonomously, and the robot notified the resident at the door to pick up the food.
People's Daily reported during WRC 2026 that compliant data from real physical interaction scenarios in China totaled only 500,000 hours, while robot commercialization requires tens of millions of hours of data.
People's Daily reported during WRC 2026 that the data gap for robot commercialization in China exceeded 99%.
Beijing Humanoid Robot Innovation Center CTO Tang Jian presented a three-stage path for humanoid robot industrial implementation at WRC 2026: industrial and special scenarios break through first, commercial service scenarios accelerate penetration, and home and generalized scenarios ultimately land.
During WRC 2026, Xingdong Jiyuan co-founder Xi Yue said the company's embodied logistics solution had achieved regular operations with SF Express and China Post in more than 10 logistics centers across 5 provinces and cities.
During WRC 2026, Xingdong Jiyuan co-founder Xi Yue said the company's embodied logistics solution had exceeded human operational efficiency in some scenarios and improved scenario migration speed from two months to within one week.