Hangzhou's GEWUAI raises nearly $14M to give robots a sense of touch
The two-month-old 'embodied brain' startup won industrial and state backing for technology that reconstructs tactile signals from video.
GEWUAI (Gewu Tongzhi), a Hangzhou startup building an 'embodied brain' for robots, has closed an angel round of nearly 100 million yuan, about $14 million, just two months after its founding in August 2026. Investors include Peak Investment, Hongfu Assets, Hangzhou Capital, Hangzhou Talent Group, a provincial-level embodied AI industrial fund and several leading listed companies, with Maple Pledge serving as sole financial adviser.
GEWUAI's bet is that vision has hit its limits in manipulation research: seeing a screw is not the same as driving it. When robots face contact-based tasks — fabric slipping, parts jamming, insertion resistance — cameras cannot reveal how much force to apply or which way friction is acting. The company's answer is what it calls a 'physical inversion world model': inferring implicit physical information such as contact and applied force from visible actions for large-scale pretraining, then calibrating against real force and tactile feedback.
The core of the approach is Real to Sim — using real physical interaction data to reverse-engineer simulation environments, then reconstructing tactile signals from the massive volume of first- and third-person video on the internet. The idea is to give tactile data the scaling properties of visual data, sidestepping the slow, expensive collection of glove-based tactile datasets. Whether the reconstructed signals can meet industrial precision requirements has no public third-party validation so far.
On top of the world model sits a PSI physical self-evolution system that turns every deployment into a data flywheel: successful runs, failure trajectories and human corrections from real work are filtered and fed back to improve the model. The company says it is running joint innovation proofs of concept with customers in textiles and apparel, optical modules and auto parts, with the first PoC delivery expected by the end of 2026. Founder and CEO Cai Jia says the metric the team watches most is customer repurchase rate, not accuracy benchmarks.
The round's industrial anchor is Peak Investment, a leading domestic sports technology group, which signed a strategic cooperation agreement with GEWUAI around footwear manufacturing — flexible manufacturing, flexible materials and sub-millimeter precision — where touch sensing is decisive. The two sides aim for large-scale orders in the company's early stage.
Cai earned his PhD in astronautics from Northwestern Polytechnical University, researching dexterous on-orbit manipulation, then joined Huawei's 2012 Labs and the Car BU 2030 Lab, where he worked on ADS 2.0 through 4.0 and million-unit mass-production delivery, leading a 100-person team and joining Huawei's Scientist Committee. The core team draws from Huawei, Li Auto, Horizon Robotics and AgiBot.