Beijing’s market regulator has published a filing for a new joint venture between China National Petroleum Corporation’s equipment arm Kunlun Manufacturing and its investment vehicle Kunlun Capital on one side, and Unitree Robotics and Beijing Fanjie Intelligent Technology on the other. The business is industrial embodied robot solutions. The split is 60, 20 and 20, with the CNPC side in control.
Three months ago Kunlun Capital took part in Unitree’s STAR Market strategic placement with RMB 136 million. Moving from a financial stake to a controlled operating entity in one quarter is fast by the standards of state-owned equipment groups.
Why 60 per cent is the number that matters
An energy major has put a hardware company and a model company inside a single legal entity that it controls. The purpose is not research leadership. It is to decide what kind of machine is allowed into a CNPC field station, and who signs the acceptance report that follows.
Unitree is the known quantity. Its B2-Ex explosion-proof quadruped already works gas gathering stations and production areas on route inspection and fault screening. Fanjie is the unfamiliar one. Incorporated in April 2026 with registered capital of RMB 1.11 million, it was incubated by the team of Qiao Hong at the Chinese Academy of Sciences Institute of Automation, who holds 9 per cent directly. Its chief executive Han Jinchi comes from the same group. The company’s stated direction is the integration of world models with embodied foundation models, constraining model output with mechanistic models rather than relying on data alone.

That fits the operational problem in oil and gas. Real leaks, fires and emergency states cannot be enumerated in a training set, so a purely learned model has a structural weakness in exactly the rare conditions that matter most.
The logic holds. The field validation does not exist yet. Fanjie has almost no measured hours in an oilfield. Putting it inside the same entity as the buyer is a way to compress the distance between a prototype and the signature on an acceptance form.
CNPC already had robots. What it lacked was a standard part
Kunlun Digital Intelligence, the group’s digital arm, has an explosion-proof inspection robot called Xiaozhi with Ex d IIC T6 certification. It ran 216 continuous hours at minus 40 degrees Celsius over the Spring Festival in Xinjiang, replacing more than 350 person-shifts of manual inspection, with recognition accuracy above 99.5 per cent.
The pipeline bureau’s explosion-proof quadruped completed the first four-legged inspection deployment in oil products at the Xigu depot in May 2026, solving signal shadowing, trench crossing and intrinsically safe circuit retrofit along the way. Those are operating records, not demonstrations.
The problem is dispersion. Xiaozhi is wheeled and struggles where trenches are dense and terrain undulates. The pipeline bureau pilot is a project, with hardware, algorithms and sensors coming from different suppliers, which makes system integration and unified scheduling expensive. Kunlun Digital Intelligence is strong in software, models and systems integration but is not an equipment manufacturer. Kunlun Manufacturing, set up in Xiongan in October 2023, is the equipment platform and handles intelligent upgrades of products.
Neither is positioned to hold hardware, model and acceptance criteria in one product definition. That is the gap the venture fills.
The first competitor is inside the group
Multiple robot routes are already running inside CNPC, and they compete for the same procurement budget. Kunlun Digital Intelligence’s wheeled units, the pipeline bureau’s quadruped pilot and external suppliers such as Qiten, which has passed eight million accumulated operating hours in industrial explosion-proof inspection, are all in the same queue.
Procurement in oil and gas is driven by operating procedures. A robot that passes a pilot is still years of stability assessment away from a framework purchase. A technology has genuinely entered routine procurement only when it is written into the formal operating and maintenance procedure.
Which means 60 per cent control buys product definition and does not buy the accumulation of operating hours. Unitree’s hardware reliability and Fanjie’s model interpretability have to run enough site hours inside the same entity before Kunlun’s acceptance standard becomes a reusable procurement basis.
There is a physical constraint waiting in the first winter. Unitree’s B2-Ex is rated from minus 20 to 55 degrees Celsius. Xinjiang oilfields routinely drop below minus 40, and Xiaozhi only manages that with a low-temperature battery and full-chain cold-weather hardening developed for the purpose.
Editor’s note: This is an adapted translation of the original OFweek report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://robot.ofweek.com/2026-11/ART-8321202-8420-30710796.html.