Embodied Emergence reports that Jiangsu Cell Wall Intelligence, a Kunshan hard-tech firm, has closed a several-hundred-million-yuan angel-plus round led by Oriental Fortune Capital, with Legend Star and Tianqi Capital following and FlowCapital as long-term adviser. The money funds CyboPal ONE volume delivery and its embodied interaction system.
After a million-yuan angel round last year, this Tsinghua-linked startup took a large cheque and posed a real question to the noisy embodied-intelligence field: while the sector crowds into humanoid prototypes, could the desktop productivity scene be the first place consumer embodied AI closes a business loop?
Why skip the humanoid track
Cell Wall was founded in March 2023 in Kunshan, grown from Tsinghua’s core technical circle. Founder Peng Tianfang, a 1992-born Tsinghua precision-instrument PhD, worked on MEMS micro-electro-mechanical chips, the sensory base of robot perception. After a stint at Qiyuan Lab on tech transfer, he formed a pragmatic view: hard-tech must face cash flow and real user need, and products should fit people, not the other way round. That steered the firm away from the hot general-humanoid track to the overlooked desktop office scene.
The team serves commercial delivery: the algorithm lead comes from Didi and Baidu autonomous driving, the chief AI expert from Huawei, Alibaba DAMO and Baidu, the Agent team built a GitHub project past 17,000 stars, and supply draws Foxconn and listed robot-firm veterans. Many embodied teams make dazzling lab prototypes but lack volume and supply control; the prototype’s win is not the product’s win.
The industry increasingly sees humanoids stuck on high cost, weak generalisation in messy spaces and long home-deployment cycles, with most products stuck at trade-show demos. Cell Wall picked the high-frequency desktop office scene, needing no human-like body, to reach a paid need faster.
Where the desktop barrier sits
CyboPal ONE breaks from companion toys: a six-degree arm holds a light 4K screen and, with industrial motion control, follows the user in millimetres. Fixed screens force users to bend to them; this one locks the visual centre when focused and pulls back when the user leans back, with gesture and voice links. MEMS sensing feeds posture, distance and gesture to a multimodal model and agent that drive the arm, closing the perceive-decide-act loop.
QYResearch puts the 2025 desktop-robot market at about 618 million US dollars, rising to 969 million by 2032. Most products lean to entertainment; mature office-productivity tools are scarce. The hard part is not the arm but the timing balance between millisecond control and slower model inference, the gap that turns demos into stalled experiences.
Two rounds in a row show capital endorsing this different route and a shift in the field’s winds. Humanoids face three gates of cost, scene and generalisation that are hard to clear fast; Cell Wall’s desktop scene needs no human form and reaches a daily, high-frequency environment where real payment shows sooner. The newest test is volume delivery: can lab-grade millimetre tracking survive mass output, with supply consistency and failure rate deciding the product’s fate.
Cell Wall’s probe matters: embodied intelligence need not be a human-like robot. The real core is a machine that senses human state and adapts to human habit, not one that copies the human shell. Anchoring to high-frequency need and balancing skill, cost and real pain is the more realistic path to the mass market.
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.
Translated and adapted from OFweek Robot (https://robot.ofweek.com/2026-09/ART-8321201-8460-30701465.html).