Embodied Emergence has learned that Lieon (Shenzhen) Robotics Technology has closed a tens-of-millions-of-yuan strategic round, invested by Shenzhen Lifud Technology, with Yunyue Capital as exclusive financial adviser. At the same time the company reached a cooperation intent with associate professor Zhang Meiying’s team at the School of Artificial Intelligence, Southern University of Science and Technology, to jointly tackle embodied intelligence, spatial perception, world models and high-quality embodied data.
This round folds a real manufacturing line into the R&D iteration loop, offering a new solution sample for an industrial-embodied-intelligence sector starved of data.
How the team read the industry rhythm
Many robot startups polish a hardware prototype first, then hunt a scenario. Lieon took the reverse path. Its core team comes from commercialisation teams at UBTECH and Pudu, carrying both algorithm R&D and factory-delivery experience. Founder and CEO Zhang Jian has over a decade in robot system architecture and project management; co-founder and CPO Zhang Ping has taken multiple robots from R&D through mass production to field deployment.
CTO Jiang Chenchen holds a PhD with a cross-disciplinary background across semiconductor materials, 3D perception, multimodal large models and robot motion control, with over 40 related invention patents.
Zhang Jian has argued the real gap is not the large model nor the robot body, but the middleware infrastructure connecting model, robot and the real world. That judgment set Lieon’s route: from day one it bet on data capture, skill training and scenario iteration rather than one hardware product. Lifud’s investment is not a normal financial bet.
Lifud runs two industrial parks, three R&D centres, serves over 70 countries and 10,000-plus enterprise customers. It opens all production lines and overseas channels to Lieon, giving not just cash but a steady stream of real industrial samples.
Can data capture break the data drought?
The sector agrees the biggest bottleneck in embodied AI is not the model algorithm but the chronic shortage of high-quality real interaction data. Traditional teleoperation capture costs hundreds of yuan per hour, while simulation data suffers reality-to-sim decay and collapses on real lines.
Lieon’s LA Capture System is a wearable, body-free capture device that simultaneously records the worker’s first-person view, hand motion, spatial position and tactile multimodal signals. Workers capture data in normal operation without an expensive robot body. A companion platform auto-splits action sequences, aligns multimodal timestamps, scores quality and versions the data, turning work into training assets. The device is already running in Lifud’s factory.
The long-term goal is an “embodied-intelligence skill factory” storing reusable skills like grasping, assembly and packing. Versus mainstream schemes, this route never interrupts production rhythm or rewires the line.
Is industrial capital enough with just money?
CVC investment in robotics is no longer rare, but most stops at writing cheques and rarely opens the workshop as a test bed. Lifud’s value is turning its complex light-industrial lines into Lieon’s lab, full of sorting, plugging and assembly tasks plus failure samples like slipping parts, exactly the scarcest training material.
Lifud’s 70-plus-country sales network also paves Lieon’s path from domestic polish to overseas copy. Some peers bet everything on simulation; others build costly teleoperation centres, but both dodge the missing physical detail of the real world.
But Embodied Emergence warns wearable capture has limits: quality is disturbed by worker habits, and heavy later engineering on cleaning and screening remains. Hardware only lowers the bar, it does not replace data governance.
Where is the real breakthrough?
The sector is stuck in a contradiction: demo videos lift expectations while few industrial projects close a real loop. Prototype buzz, cold scale-up, has become a chronic disease.
Zhang Jian’s point hits the pain: the middleware is the true short board. Bodies and models iterate fast; only a data link and skill system rooted in the real world build a lasting moat. This round shows a logic switch, from “smartest model” to “infrastructure that works”. The divide is no longer the most dazzling demo but the lowest-cost, sustainable data loop.
In the next two to three years, survivors may not be the highest paper-metric teams but those who root in the line and turn field experience into reusable robot skills. Lieon holds a rare industry ticket; whether its skill transfers beyond a single factory remains the market’s question.
Editor’s note: This is an adapted translation of the original OFweek Robot report. It has been trimmed and restructured for readability for an international business audience.