LivSyn Robotics has raised a Series A of more than RMB 100 million to pursue a stubborn industrial-robotics problem: work cells often need their data, training, and control logic rebuilt when the hardware changes. Its RUDA platform is meant to give robot builders a common layer between physical machines and AI models. If the approach works outside demonstrations, factories could spend less time treating every new arm, sensor package, or robot body as a fresh integration project.
PEdaily reported the October 8 financing for the Beijing-based company, also known in Chinese reporting as Lingsheng Technology, naming Huafang Capital, Daohe Yuanqi, Suwen Electric Energy, and an unnamed embodied-AI strategic investor. TechNode independently reported a round of at least RMB 100 million, identifying Suwen Electric Energy and an undisclosed strategic investor. The money matters less as a vote on a robot brand than as financing for an integration bet: that the expensive part of putting AI into industrial work can become portable across hardware.
LivSyn’s pitch starts with an ordinary operational constraint. A robot cannot benefit from a capable model if its instructions, sensing, data format, and feedback loop do not connect cleanly to the machine on the floor. A team that changes equipment may face another round of collecting demonstrations, translating actions, testing control behavior, and repairing failures. RUDA, short for Robotics Unified Device Architecture, is LivSyn’s proposed answer to that break in continuity, still awaiting proof across embodiments.
RUDA Turns Integration Into the Product
On its English-language official website, LivSyn presents RUDA as a unified embodied-model architecture. The company places it inside a wider stack that begins with human-centered devices for capturing demonstrations, moves through DataCube and a data-management platform, then uses RUDA and PhiAgent for training before handing deployment and feedback to RoboAgent. That sequence reveals what LivSyn is trying to sell. The product is not simply a robot or an AI model. It is the connective tissue that is supposed to let information travel from a human action to a trained model and then back into a physical machine.
LivSyn describes this as a collect-train-deploy loop. Its devices capture demonstrations; the company processes and governs the data; RUDA and PhiAgent train vision-language-action or vision-action models across embodiments; and RoboAgent sends plans to robots while receiving feedback on execution. Each stage addresses a different failure point. Data can be hard to reuse, a model can be detached from a particular machine’s movements, and a plan can fail once it meets real hardware. By putting these stages in one architecture, LivSyn is attempting to make the handoff among them less bespoke.
That is a sharper proposition than calling a robot system intelligent. A factory buyer needs to know whether a task can keep working after a camera, arm, end effector, or robot body changes. An integrator needs to know whether a demonstration gathered for one job can lower the cost of the next. The architecture question sits inside the division of work captured by China’s embodied-AI financing tally: funding can favor the model layer, the machine layer, or the difficult connection between them. LivSyn is explicitly staking its round on that connection.
Industrial Data Has to Survive the Hardware Change
PEdaily describes PhiAgent as turning video of human operation into reusable robot data and motion trajectories. It describes RoboAgent as the layer that understands tasks, calls skills, controls robots, and adjusts from feedback. Those are PEdaily’s and LivSyn’s descriptions of how the system works, rather than independently validated measures of reliability. Still, they show why data collection is central to the company’s commercial logic. A useful record of a person doing an industrial task has value only if it can be turned into machine action without a long, hardware-specific rewrite.
The company and the reports say LivSyn has applied its technology to fitting and removing optical modules, assembling batteries, and checking electronics. Those are constrained jobs where repetition can make results observable, but the public information omits named production deployments, robot configurations, task conditions, failure rates, and operating cost. PEdaily reported a nearly 100% test success rate for optical-module insertion under unspecified conditions. The figure points to a task LivSyn believes its stack can handle, rather than audited, production-scale performance on another robot body.
That distinction goes directly to the kind of reuse RUDA promises. A near-perfect result on a narrow test may be useful to an engineering team, yet it says little about whether the same data, skill, and model can move between hardware without losing accuracy or demanding human intervention. The next customer will care about the full chain: what data was captured, what part was learned, what part was programmed, and what happened when the machine encountered variation. The value of reusable robot data is also at the center of Mifeng Pai’s crowdsourcing effort, because physical AI does not gain scale from a data set that remains locked to one controlled setup.
LivSyn’s public partner and client-logo page adds a further reason to distinguish visibility from validation. The site lists Tsinghua University, Tencent Cloud, Zhipu, BOE, COFCO, Foxconn, Transsion, and others, but offers no contract scope, deployed product, revenue, or timing. The logos may indicate a broad network around the company; they provide no proof that RUDA is operating in a named factory or that a customer has adopted the platform as a production standard.
Series A Puts Evidence Ahead of Architecture
The Series A gives LivSyn room to turn an architecture diagram into evidence that matters to industrial users. PEdaily identifies founder and CEO Yang Hongbing as a Tsinghua graduate with earlier Tencent and humanoid-robotics roles, and identifies co-founder and CTO Jiang Yuhua as a Tsinghua doctoral graduate. A funding round still leaves the company’s central operating problem unresolved.
The real question is whether RUDA can preserve useful work when hardware changes. An industrial buyer would need more than a claim that a model spans embodiments. It would need disclosure of the robot bodies involved, the task conditions, and whether operation is autonomous or assisted. It would need intervention and failure rates, a clear account of which skills or training data were reused, and an economic result that shows why the architecture is preferable to a conventional, hardware-by-hardware integration stack.
That is the next test LivSyn’s new capital makes possible. A disclosed deployment should show a concrete task moving through the company’s capture, data, training, and deployment loop, then identify what persisted when the hardware changed. If the company can document that reuse alongside the human assistance and operating cost required, RUDA will be more than a useful name for a stack. It will offer factories a practical basis for deciding whether a shared software layer can finally reduce the repeated work of making AI operate on different robots.