Recently, Lingyan Technology completed tens of millions of RMB in seed and seed+ rounds of financing, with investors including Yingdong Capital, Yida Capital, and Plum Ventures. The company was established in Hangzhou in June 2026, founded by Zhao Changyi, with a core focus on bionic face (dexterous face) modules for humanoid robots. Simply put, it aims to enable the robot's face to make the right expression at the right time. While this may sound like an obvious statement, in robotics, it represents a narrow gate capable of stalling the entire industry's commercialization process.
1. A "Slow Path": Why Wire-Driven Mechanisms Are the Correct Solution for Robotic Faces
Bionic faces are not a new concept. Most players in the market choose linkage-driven mechanisms: motors drive linkages, which push facial components to simulate expressions through mechanical pushing and pulling. This approach offers short design cycles and rapid prototyping, but Lingyan founder Zhao Changyi believes linkages have three unavoidable critical flaws: high noise levels, large space requirements, and the quick-return characteristic of linkages causing sudden speed changes at the end of movements—resulting in expressions that "stutter," resembling mechanical devices rather than muscles.
Therefore, Lingyan chose wire-driven mechanisms. Using extremely thin cords routed internally across the face to mimic the arrangement of human facial muscles, they pull the skin to generate expressions. The force required for wire-driven systems is very small, and the driving form naturally aligns more closely with muscle mechanics. As a direct result, the facial degrees of freedom reach 32, the highest among comparable domestic products; by year-end, a fully wire-driven version with 46 degrees of freedom will be launched, capable of reproducing nearly all subtle human facial movements. In the field of dexterous hands, wire-driven mechanisms struggle to gain traction due to load-bearing and fatigue life issues, but since faces only need to pull skin for micro-movements, the shortcomings of wire-driven systems disappear here.
However, this is a "slow path." The position of every muscle and the routing pattern of every cord must undergo biomimetic mechanism testing in the laboratory. Linkages can quickly assemble a "functional" face using non-standard parts, whereas wire-driven systems require long-term accumulation to ensure every cord "pulls in the right place." Zhao Changyi is confident, stating firmly: "In the future, facial actuation will definitely converge on the wire-driven path."
More noteworthy is the origin of this technical path. Lingyan did not grow in isolation; it originated from the Key Laboratory of Engineering Bionics at Jilin University. Its peer, Quanzhibo, develops integrated joint modules, focusing on robot "movement"; Heiman Technology focuses on dexterous hands, targeting "grasping"; Lingyan focuses on dexterous faces, addressing "interaction." All three founders studied under Professor Liu Chunbao. Lingyan's wire-driven technology path stems from Heiman Technology's accumulation in dexterous hands, while its miniature drive motor module technology reuses Quanzhibo's expertise in robot joints. Joints, dexterous hands, and dexterous faces share not just supply chains, but also an engineering logic growing from underlying hardware upwards.
2. Two Halves of a Face: Hardware is the Skeleton, Algorithms are the Key to "Not Being Scary"
32 degrees of freedom only solve the problem of "being able to move." What truly transforms a bionic face from "uncanny" to "acceptable" is software.
Lingyan's core architecture is called FaceSoul Agent, divided into two halves: Face and Soul. Face handles high-fidelity physical presentation, including bionic skin, muscle transmission, expression temperature, and posture sensing; Soul carries the emotion engine, social cognition, and expression generation algorithms. Zhao Changyi deliberately avoided conversational large language models, locking the R&D focus strictly on "expression generation"—OEMs typically handle conversation and memory functions themselves, so the essential requirement for dexterous face suppliers is converting voice and visual inputs into precisely aligned mouth shapes, eye movements, and facial demeanor.
Traditional bionic faces rely on preset expression libraries: smiling is smiling, crying is crying, invoking fixed actions. Lingyan's self-developed amygdala emotional modeling model, trained on real emotional data, supports unlimited nuanced expression generation with latency controlled within 10 milliseconds. Robots can make continuous switches between different intensities such as "slight smile," "moderate happiness," and "high excitement" based on conversational context. Audio-lip-eye synchronization presents another challenge: mouth shapes must match pronunciation, eye contact must align with the speaker, and any misalignment of even fractions of a second causes "friendliness" to instantly collapse into the "Uncanny Valley." FaceSoul Agent achieves this through a low-latency distributed collaborative control system, ensuring facial control appears smooth rather than mechanically jerky.
3. Where Faces Are Used: Validating B-Side First, OEMs are the True Leverage
Lingyan's product matrix has already expanded along four lines: Lingshu M1 targets education and research, offering open interfaces for university labs to develop algorithms; Lingshu W1 targets commercial guidance, reception, scenic spot introductions, and cafe ordering; Lingyun series targets C-side emotional companionship in homes; and Lingneng series targets humanoid robot OEMs, providing plug-and-play standardized facial modules compatible with mainstream OEM mechanical and communication interfaces, enabling functionality simply by "changing the robot's head."
Zhao Changyi judges that the node for dexterous faces entering homes on a large scale is likely next year or the year after; this year, various companies are still iterating on technology, with the focus on validating scenarios on the B-side. Cultural tourism guidance, brand reception, and educational research have genuine demand for "a face that doesn't scare people" and allow early-stage products to iterate through real interactions. The significance of the Lingneng series leans more towards the industrial layer: pricing is expected to be controlled below 10,000 RMB, making it "affordable" for OEMs. Once bionic faces become optional standard equipment for humanoid robots rather than geek customizations, the mode of demand release will be completely different.
4. Filling the Missing Layer in Embodied AI
For companion robots, the most frequently mentioned term is the "Uncanny Valley." Humans derive most emotional information from faces, and tolerance for "faces" is extremely low; even minor unnaturalness is amplified. This means that if the interaction layer is not addressed, the step of humanoid robots moving from "tools" to "companions," and from factories to homes and commercial spaces, will always lack a foothold.
Lingyan Technology's value lies not in creating a more realistic silicone face, but in systematically addressing the "Uncanny Valley" as an engineering problem through a combination of wire-driven mechanisms and expression generation algorithms—from physical driving at the muscle level, to expression intensity control at the contextual level, to millisecond-level audio-lip-eye coordination.
In four months, raising tens of millions, led by a post-2000s team, choosing the narrowest door. Behind the narrow gate, there may not necessarily be a smooth road, but if robots are truly to enter the human world, that face must first make the right expression at the right time.
END