Gewu Tongzhi Completes Nearly 100 Million Yuan Angel Round to Develop Robot Brain That Understands Physical Changes

Gewu Tongzhi has reportedly completed an angel funding round of nearly 100 million yuan. Investors include Hongfu Assets, Peak Investment, Hangzhou Capital, Hangzhou Talent Group, an unnamed provincial-level embodied AI investment firm and several leading publicly listed companies. The company says it will develop an “embodied brain” that enables robots to understand objects and physical changes during manipulation.

Gewu Tongzhi says it aims to develop an “embodied brain” that links vision with physical interaction, and plans to use the funding to build its training infrastructure and physical robot capabilities. Its approach to improving robot manipulation — linking data collection, model training and robot design — differs from developing software for a single function.

The focus of this funding round is not simply that Gewu Tongzhi has raised money, but that it plans to use the funds for large-scale pretraining of a “physical inverse world model,” hardware and robot development, proof-of-concept testing across multiple scenarios, and real-world data collection. This points to an approach that advances models, robots and data together. Rather than operating as a purely algorithm-focused company, it aims to develop both the training infrastructure for an embodied brain and the reproducibility of its physical robots. A defining feature is its treatment of robot action understanding not as an abstraction, but as a learning problem involving physical interaction. Placed alongside discussions of “robot brains” at companies such as Zibianliang Robotics, Kuawi Intelligence and Ruoyu Technology, covered in related articles by this newspaper, Gewu Tongzhi stands out for its emphasis on world models and physical interaction learning. The technical focus of a “brain” can vary, from strengthening vision-based recognition to estimating the mechanics involved in manipulation. This report points to the latter: modeling how objects deform, experience friction and make contact. The broader industry implication is that model training alone is not enough; real-world data collection and robot design must advance in parallel. The company says it will conduct proof-of-concept testing across multiple scenarios, suggesting that the focus is shifting from research demonstrations to reproducibility and sustained operation in industrial settings. Cai Jia, who previously worked at Huawei and has experience with mass-production delivery of ADS, can also be seen as part of a team mindful of the transition from research to implementation. The questions ahead are how accurately the system can estimate physical changes, what specifications the robots will have, and whether the proof-of-concept work will lead to mass-production designs. The report names investors and intended uses of the funds, but does not disclose the industries or customers involved in the proof-of-concept projects, the number of robots, the scale of the training data or model performance metrics. At this stage, the key issue is less the size of the funding round than whether an embodied brain centered on a world model can establish an integrated operation spanning data collection through robot development.

The fact that a former Huawei employee involved in mass-production delivery of ADS has founded a company pursuing physical understanding in robots can be seen as a shift of mass-production-focused experience developed in automotive AI into embodied AI. For Japanese companies, this could be a stage at which integrated proposals that include tactile and force sensing and robot design, rather than models alone, become essential.