Ruoyu Technology has raised fresh capital for a robot brain meant to work where mistakes have immediate physical consequences. The Shenzhen company’s newly reported Pre-A round, described only as a nine-figure RMB or hundred-million-yuan-level financing, is intended to advance Jiutian, its cloud-and-local-control system, and the data loop around a refueling-robot trial in Foshan. The material question is not the round’s undisclosed price tag. It is whether an embodied-AI system can plan and carry out a demanding physical sequence while sensing and responding at the site.
The investor group puts Ruoyu in a hardware-minded corner of China’s robotics market. 36Kr, which first reported the round, identified Sunway Communication as lead investor, with Jinmu Capital, Qindao Capital, Huanghai Capital, and Zhongchuang Capital participating. TechNode also reported the financing while crediting 36Kr. Neither report disclosed a precise amount, valuation, ownership stake, check sizes, or closing date. Ruoyu says the new backing is intended to support development of the system around a real machine and field trial.
The company dates to 2023 and emerged from Harbin Institute of Technology, Shenzhen, according to its profile. Founder and CEO Sun Teng is pursuing a strategy that joins the robot’s brain, body, and scenario data. The company calls the system Jiutian; the hardware vehicle in the Foshan work is its Lanyue 01 robot. For an operator considering automation, that pairing is what counts: software cannot be separated from the sensors, local compute, and physical machine asked to do the job. It is also a useful point of comparison with EastFrontier’s wider embodied-AI reporting.
Jiutian’s cloud-edge-end split keeps control near the robot
Ruoyu describes Jiutian as a cloud-edge-end architecture rather than a cloud-only service. In the company’s account, the cloud manages a world model and broader task planning; the edge layer handles millisecond-level sensing and safety control; and the terminal robot uses heterogeneous compute and multiple sensors for precise execution. 36Kr reported that division of labor, while TechNode summarized the premise as cloud planning coupled with local perception and control. This is Ruoyu’s stated design, not an independently benchmarked robot platform, but it shows where the company expects each decision to be made.
That split is especially relevant in refueling. A remote planning layer can frame the task, but a robot at a fuel island must make its next move through on-site perception, local control, and its own sensors. Ruoyu says selected, de-identified work data may return asynchronously to the cloud for model and strategy improvement, subject to authorization and security compliance. Its stated loop is meant to connect experience at a live site to later system changes, rather than treat each deployment as a fixed installation. Sun Teng told 36Kr the new money is intended for full-stack robot-brain research and development, real-scenario data loops, and industrial implementation.
Ruoyu’s design may be built for the transition from a stated architecture to dependable operation, but the public material does not establish that Jiutian has made it across sites or robot fleets.
Foshan refueling gives Ruoyu a high-consequence proving ground
Ruoyu chose intelligent refueling as its first high-value application, and 36Kr reported that trial operation began at a Foshan demonstration site in April. Ruoyu’s own announcement says Lanyue 01 entered a gas station for pilot operations that month. The setting links planning, perception, robot hardware, and site conditions in a task where vehicle positioning, the refueling sequence, and the fuel-island configuration cannot be treated as abstract inputs.
There is independent confirmation that the project exists as an application case, though not as a performance scorecard. Shenzhen’s Industry and Information Technology Bureau placed Shenzhen Ruoyu Technology among 48 selected 2026 Robot+ demonstration cases. The official case list identifies its energy entry as a special-robot refueling application based on a multimodal large model. The listing confirms the selected demonstration case; it does not report revenue, deployment scale, safety outcomes, or model accuracy.
36Kr also reported Ruoyu’s statement that the Foshan task recorded a high success rate for fixed vehicle types in an open environment and required manual takeover only in abnormal cases. That company-reported observation has no published completion percentage, task count, intervention count, operating duration, or independent testing record. It should not be read as a general performance result. The consequential question is whether a robot can keep doing the job when vehicle types, lighting, weather, layouts, and unusual events change.
Foshan matters because gas-station operators and customers need a system that can complete a long physical sequence safely in a high-consequence setting. The pilot gives Ruoyu a place to collect operating experience where missed perception or poor execution have real costs. It remains a pilot, not a network-wide rollout or proof of unit economics. Readers can place it beside EastFrontier’s broader embodied-AI coverage, while Ruoyu still has to show that its system works beyond its reported fixed-vehicle trial.
The Pre-A sets up a multi-site evidence test for Ruoyu
The lead investor adds relevant systems context without revealing a business arrangement. Sunway Communication is a Shenzhen-listed maker of RF-related electronic components and modules, with humanoid robots among the application areas on its official profile. Its involvement is notable for a company whose product depends on hardware as well as software. The reports do not disclose a supply agreement, technology transfer, board seat, or commercial partnership between Sunway and Ruoyu, and the investment should not be treated as evidence of any of those outcomes.
Ruoyu says the capital will support robot-brain research, real-world data loops, and industrialization or commercialization. The next meaningful measure is the record that plan produces. A useful extension would disclose completed refueling cycles, completion rates, task times, manual-takeover rates and causes, availability, and maintenance requirements over a stated period.
The operating setting matters, too: how many sites are involved, whether the deployment is paid, who operates the machines, and what safety events or near misses occur. Clear operator or regulator sign-off, where applicable, would carry more weight than another unmeasured autonomy claim. None of those measures has been publicly supplied for the Foshan pilot. Their absence is not proof that the project has failed; they are the evidence needed to judge whether a controlled trial can become a repeatable service.
For Ruoyu, the Pre-A is a chance to close the distance between a stated cloud-edge-end design and operations that can be checked outside a single reported trial. For investors and robotics buyers, a multi-site record would show whether local sensing and control, cloud planning, and field-data improvement reinforce one another under changing conditions. The important development is fresh backing around a defined test: a Shenzhen startup is working to show that its existing refueling pilot can earn trust cycle by cycle.