Gewu Tongzhi
Category: Robotics / Embodied AI
Hangzhou embodied-AI 'robot brain' startup building tactile-aware dexterous manipulation intelligence for industrial robots, using a physics-inversion world model and a self-evolving learning loop. Gewu Tongzhi was founded in 2026. The company is led by Cai Jia (蔡佳). Based in Hangzhou, China. Total funding raised: $13.9M. Latest round: Angel. Key investors include Peak Investment (匹克投资), Hongfu Assets (鸿富资产), Hangzhou Capital (杭州资本), Hangzhou Talent Group (杭州人才集团), Hangzhou Runmiao Fund, Hangzhou Disruptive Achievement Transformation Fund, Hangzhou Talent Fund.
- Founded
- 2026
- Headquarters
- Hangzhou, China
- Total funding
- $13.9M
Value proposition
Gives robots a 'sense of touch' (手感) for fine, contact-rich manipulation such as fabric slippage, part jamming and insertion resistance. Robots keep learning on the production line, so experience from one workstation lowers deployment cost at the next.
Products and solutions
Physics-Inversion World Model (物理反演世界模型): infers contact and force information from video for large-scale pretraining, then calibrates it with real tactile and force feedback. PSI Physical Self-Evolution system: a closed loop of physics inversion, planning and execution, real feedback, evaluation and learning, and validated deployment. Real-to-Sim tactile data pipeline: reconstructs tactile signals from first- and third-person video. Deploy-Learn-Generalize industrial deployment methodology.
Unique value
Scales tactile training data to video scale by reconstructing touch signals from internet video (Real-to-Sim). It natively fuses touch and force with vision in a world model, and the system self-improves from on-site success and failure data.
Target customer
Industrial manufacturers that need flexible, high-precision manipulation, including footwear, textile/apparel, optical module and auto-parts factories.
Industries served
Footwear manufacturing, Textile & apparel, Optical modules, Automotive components
Technology advantage
The founder brings world-model and autonomous-driving mass-production experience from Huawei, including governance of tens of millions of data clips and training on clusters of nearly 10,000 cards. The data strategy puts quality first, using automated data-value scoring. Core members come from Huawei, Li Auto, Horizon Robotics and AgiBot. The company has had an industrial anchor partner (Peak) since its first round.
How they differentiate
Competitors mostly rely on wearable tactile gloves or crowdsourced data collection. Gewu Tongzhi instead reconstructs tactile labels from video at scale. It emphasizes in-factory, on-the-job self-evolution and treats customer repurchase rate, rather than demo success rate, as its core metric.
Main competitors
TARS (它石智航), NeoteAI (新智具身), Chuguang Yueqian (初光跃迁)
Key partnerships
Strategic cooperation agreement with Peak (匹克) for footwear-manufacturing deployment. Ecosystem partner of the National Embodied AI Application Pilot Base. Member of the Hangchuang Embodied AI Camp (杭创具身智能营).
Notable customers
Peak (strategic partner/customer for footwear manufacturing)
Major milestones
2026-08: Founded in Hangzhou as 格物通智（杭州）有限公司. 2026-10: Closed a nearly RMB 100M angel round, with Maple Pledge as exclusive advisor, and signed a strategic agreement with Peak. 2026-12 (target): First industrial PoC delivery.
Growth metrics
Registered in August 2026. Raised nearly RMB 100M in an angel round about 2 months after registration. Joint-innovation PoCs are underway, with the first PoC delivery targeted for the end of 2026.
Market positioning
Early-stage embodied 'brain' company (foundation model plus learning system) focused on contact-rich industrial dexterous manipulation in China. Its scenario-first strategy targets precision on narrow tasks first, then generalization.
Geographic focus
China (Hangzhou / Zhejiang manufacturing base)
About Cai Jia (蔡佳)
Ex-Huawei Car BU head of world-model foundation models, leading a team of 100+. Core contributor to ADS 2.0-4.0 R&D and mass-production delivery, and to ADS 6.0 pre-research. Member of the Huawei Scientist Committee. Previously at Huawei 2012 Laboratories and the Car BU 2030 Lab. PhD, School of Astronautics, Northwestern Polytechnical University (advisor Huang Panfeng), researching in-orbit dexterous manipulation by space robots. Received Huawei's top individual technical award.