Gewu Tongzhi raises nearly RMB 100 million angel round to reconstruct robot touch from video

Hangzhou embodied-AI "brain" company Gewu Tongzhi has completed an angel round of nearly RMB 100 million, 36Kr's hardware desk reported on Oct 18. Peak Investment, Hongfu Assets, Hangzhou Capital, Hangzhou Talent Group, an unnamed provincial-level embodied-AI industry fund and several unnamed listed companies invested, according to 36Kr, and Maple Pledge was the exclusive private-equity financing advisor.

The bet behind the round is that a robot's sense of touch can be learned mostly from tactile readings reconstructed out of ordinary video, skipping glove collection for the bulk of the data. Gewu Tongzhi says its physical inversion world model infers hidden physical information such as contact and force from visible motion for large-scale pretraining, a route the company says it proposed first. Real force and tactile feedback then calibrate it, adding physical labels to action data. Founder and CEO Cai Jia says only that the reconstructed readings are expected to be accurate enough for pretraining, with real on-site data to align them further, so the first PoC delivery the company targets by the end of 2026 is where that bet first meets a factory.

Gloves are the bottleneck the company wants to skip

Cai said collecting touch data only with tactile gloves is expensive and slow, and that gloves get in the way of workers' normal jobs, leaving glove data volumes very limited. He holds a PhD from Northwestern Polytechnical University and previously led the world-model foundation model at Huawei's Car BU, 36Kr reports.

36Kr notes tactile readings usually carry only dozens of dimensions against millions in an image, so putting both into one model stays hard.

Cai puts factory requirements between ±0.1 mm and 1-2 mm accuracy, cycle times from milliseconds to minutes, success rates of 90% to 99% and reliable running of 8 to 24 hours, and says the current stage is "30% model and 70% engineering."

Peak has signed a strategic cooperation agreement under which Gewu Tongzhi will work toward footwear manufacturing's "dual flexibility plus one high precision" requirement of flexible manufacturing, flexible materials and sub-millimeter precision, aiming to win volume orders early and jointly push deployment and industrial validation. Gewu Tongzhi is running joint-innovation PoCs with customers in textiles and apparel, optical modules and auto parts, and expects to complete its first PoC delivery by the end of 2026; footwear is not among the PoC sectors 36Kr lists. Which of those lines will the reconstructed touch prove itself on first?