Fangqi Technology Raises Angel Round for General-Purpose Robot Brain

The financing will support Fangqi's effort to build a general-purpose skill operating system that allows robots to learn continuously and transfer skills across different embodiments, the report said. The company aims to develop a general robot brain and gradually build infrastructure for embodied capability learning, accumulation, and distribution across multiple robot bodies and scenarios.

Founder Dr. Wang Xinzhou argues that the weak generalization of embodied robots stems from world models lacking human knowledge and embodied brains lacking physical common sense. In short, he says, physical knowledge and semantic knowledge remain separated between world models and embodied models. To address this, Fangqi has proposed a semantic world model, aiming to build robots that learn, work, and grow like humans.

The semantic world model deeply couples an embodied large model with a world model and organizes vision, language, and physical states into a unified three-dimensional semantic space. This allows a robot to understand the world, predict actions, and make autonomous decisions in a human-like way, according to the company. Fangqi says Turing Learning and the semantic world model together form an evolution flywheel that drives the continuous growth of a cloud-native embodied brain.

Founded in 2026, Fangqi describes itself as a developer of general-purpose embodied intelligence brains. Its core team members come from Tsinghua University and cover world models, VLA algorithms, cloud-native systems, and industrial commercialization, combining frontier research with real-world deployment experience. The company says it was among the first in the industry to propose the Turing Learning paradigm, which uses curriculum learning, task learning, practice learning, and reflection learning. The approach is intended to break the limitations of traditional teleoperation-based robot learning, in which robots know what to do but not why, and to let robots continuously learn from human data and real-world practice in a scalable and replicable way.

After the funding, Wang said the company will continue to iterate its cloud-native embodied brain technology and recruit top R&D talent. It will also accelerate pilot validation of its products in benchmark scenarios and accumulate a general skill library that can be reused across embodiments. In the short term, Fangqi plans to refine deployable productivity tools; in the long term, it aims to build embodied intelligence infrastructure for the physical world, giving robots the ability to learn autonomously and improve continuously.

Fangqi said its core technology has completed technical validation for Turing Learning and the semantic world model in commercial service scenarios. It has applied for multiple technology patents and ranked second globally in the Stanford Household Challenge BEHAVIOR 2026. The company has formed strategic partnerships with Youdi Robotics (stock code 03231.HK) and Jiangsu Longhuan to advance pilot deployments. Using three-dimensional cleaning as a proving ground, Fangqi plans to gradually roll out pilots in commercial service scenarios and says it has a clear commercialization value and payback model.

Liu Bo, general manager and managing partner of TusStar Venture Capital, said the core competition in embodied intelligence lies in whether companies can move beyond the old path of piling up massive real-machine data and find a low-cost, scalable path for general robot learning. He said Fangqi's semantic world model and Turing Learning paradigm offer a new approach and that TusStar sees the company as having a chance to build a foundational software platform for next-generation embodied intelligence.

Editor's Summary

Fangqi Technology has raised an angel round in the tens of millions of yuan to develop a general-purpose embodied intelligence brain. The company is betting on a semantic world model and Turing Learning to improve robot generalization and cross-embodiment skill transfer, while pursuing commercial pilots in service scenarios with partners. TusStar Venture Capital and founder Wang Xinzhou frame the effort as an alternative to data-heavy robot learning.