JLC Opens Kaiwu Incubator as Yimu Technology Advances Optical Tactile Sensors for Robots

The open-source and closed-source tracks covered embodied intelligence, key robot hardware, AI edge computing and smart terminals. More than 50 chip and tool vendors exhibited alongside makers. In the closed-source track, AI projects accounted for more than 90 percent of entries, according to Leiphone. The gold award went to an AI dishwasher project that combines visual tableware recognition with high-pressure spray modules for small, high-frequency household cleaning and claims a five-second cleaning capability. The innovation award went to PeerSync, a portable device that combines speech transcription and AI pronunciation repair for two-way communication by hearing-impaired users; its team includes a 17-year-old middle school student. In the open-source track, the top prize in the Spark program's advanced category went to IronBen, a four-legged wheel-legged robot that combines wheeled drive and quadruped gait for complex terrain. The open-source track relies on the JLC open-source hardware platform to publish design drawings, source code and bills of materials so other developers can study, reproduce and modify designs.

JLC's existing services cover electronic design automation, component procurement, PCB manufacturing and electronics assembly, with additional 3D printing and CNC processing for structural parts. As of the end of June 2026, JLC EDA had 7.3688 million registered users worldwide; in the first half of 2026, JLC's platform had 979,100 paying users and processed 11.6394 million orders, Leiphone reported. The Kaiwu incubator will provide hardware R&D facilities for projects with stronger commercialization potential, supporting prototype development and iteration. JLC will link resources including the Hetao Shenzhen-Hong Kong industrial park and Meilin AI industry community, and rely on the Huaqiangbei electronics supply chain and ecosystem partners to connect projects with industrial scenarios and market channels. The support measures also include capital support across different stages of company development. Futian district and JLC jointly built the Greater Bay Area Intelligent Hardware Sci-Tech Innovation Center, which integrates agile manufacturing and pilot production, standard certification and scenario opening. Winning projects can receive free trial production and prototyping support and continue to connect with industrial and supply-chain resources.

In a separate interview published by iFanr on Sept. 29, Yimu Technology founder and CEO Li Zhiqiang said the company focuses on embodied intelligence tactile perception, spanning sensing, manipulation and deployment, to give robots deployable dexterous operation in the real world. Its SENTRA T0 optical ultra-thin bionic tactile sensor has been iterated to less than 3 millimeters thick, which the company describes as the world's thinnest mass-producible bionic visual-tactile sensor, solving the industry problem that dexterous hands cannot fit such sensors. Li told iFanr that Yimu calls its approach optical tactile rather than visual tactile because it uses light to detect deformation instead of taking photos for judgment. He said autonomous driving never wants contact, but the value of robotics lies in contact with the world.

Li argued that vision has advanced quickly, but robots still cannot know how much force they are using or whether an object will slip without touch. He described the typical consequence as "seeing but not doing": a robot can identify what an object is and where it is, but when it reaches out, it may crush, drop or misalign it. According to Li, VLA and world models without tactile feedback have low success rates in execution, sometimes only 50 to 60 percent. He said that in automotive manufacturing, only painting and welding workshops are highly automated, while final assembly still relies heavily on human hands for wiring harnesses, seats, rubber seals and small connectors because robots lack the final bit of hand feel.

Li identified three categories of tasks that require robot touch: objects whose appearance is deceptive, such as stone walnuts versus real walnuts; flexible objects whose nonlinear deformation is hard to predict, such as rubber strips on cars; and subtle manual skills, such as turning a poorly fitted key at a specific angle without breaking it. He cited the human process of re-grasping a glass, which requires correction to start in under 80 milliseconds, with only about 10 milliseconds available for tactile sensing, leaving vision too slow. Yimu's technical base began with light chips and micro-spectral sensing, Li said, and extends to optical touch because the underlying light-based capability is continuous.