Large language models can write poetry, yet cannot step beyond the factory fence: Yanxin Technology integrates the three technical routes of "Symbol + Behavior + Connection" into RISC-V chips to develop plug-and-play robot brains.
The robotics industry is currently at a somewhat contradictory juncture: the capabilities of large models continue to break new upper limits in dialogue and generation tasks, but once entering the physical world, the "brain" of robots still struggles to step outside the factory fence. A common problem facing the industry is that models perform well in a single scenario, but their success rate drops significantly when switched to a different environment or a new set of tasks. The difficulty of collecting data for long-tail scenarios, insufficient dynamic learning capability, and opaque reasoning process have kept "robots doing work at home" stuck at the demonstration level.
Beijing Yanxin Technology Co., Ltd. (hereinafter referred to as "Yanxin Technology") was founded in January 2025, with a registered capital of 1 million RMB, its registered address located in Changping District, Beijing, and its business scope covering basic and application software development for artificial intelligence, R&D and manufacturing of intelligent robots, etc. The solution proposed by this early-stage company is: instead of taking large models as the entire brain of robots, integrate the three technical routes of symbolism, behaviorism and connectionism into a set of "general embodied intelligent brain" system that can be pre-packaged in RISC-V chips, so that robot manufacturers can obtain scenario-based intelligence in a plug-and-play manner.
Industry Pain Points: Large Models "Cannot Enter" the Physical World
The industry judgment repeatedly stated by Yanxin Technology in multiple public roadshows is that traditional large models have bottlenecks in four dimensions: generalization, task success rate, long-tail scenario processing and dynamic learning capability. The environment that embodied intelligence needs to face is open and unstructured, which is different from the closed boundary of text generation tasks — objects at different orientations in the home, temporary obstacles in the workshop, and individual differences of users will all cause the pre-trained model to perform inaccurately. At the same time, the data backflow generated by robots running in the real environment is slow, making it difficult for the model to continue learning after leaving the factory.
This judgment is consistent with the current technical discussion direction of the industry. As an open-source, free instruction set architecture, RISC-V's open features allow enterprises to carry out in-depth independent design and optimization, and is regarded as a feasible technical path for China's embodied intelligent robots to achieve "autonomous and controllable" at the core processor level. Since 2026, the intersection field of RISC-V and embodied intelligence has become the focus of the industry. Organizations such as China Computer Federation (CCF) have also organized a series of technical activities themed "From Chips to Robot Applications: RISC-V Building an Open-Source Computing Power System for Embodied Intelligence" from August to September 2026.
Solution: "Industry Embodied Intelligent Brain" Integrating Three Technical Routes
Yanxin Technology's product idea is to integrate the three technical routes of symbolism, behaviorism and connectionism, build a general embodied intelligent brain system, pre-package and solidify it into RISC-V chips, form a plug-and-play standardized "industry embodied intelligent brain", and empower robot manufacturers to quickly realize scenario-based intelligence. According to the company's public roadshow information, the core capabilities of this system include continuous learning after factory delivery and human-computer interaction teaching — that is, after the robot is delivered to the user, it can still accumulate data and update behavior strategies through interaction, instead of being limited to the fixed capabilities set when leaving the factory.
At the "Venture Capital China" Science and Innovation Roadshow Hall Embodied Intelligence Special Session (June 2026) that the company participated in, the organizer's introduction to Yanxin Technology is: Focusing on the R&D of embodied intelligent brain, it has full-stack self-developed, safe and controllable embodied brain products, which can solve pain points such as difficult scenario generalization, robot hallucination, black box non-interpretability, greatly reduce data and computing power costs, support end-side deployment, and provide key technical support for low-cost, full-scenario implementation of robots. According to public reports under China Enterprise News, Yanxin Technology's technical path involves meta-learning and brain-inspired learning methods, which enables robots to have dynamic rapid learning and active learning capabilities, and adopts a dynamic operator group architecture to complete brain-like closed loop and new operator generation through decision operators, without cloud computing power support and can be deployed on the end side.
The above technical description is consistent with the industry trend that "robot intelligence should not rely on cloud computing power": end-side deployment means lower latency, stronger data privacy and more controllable computing power cost, which is also one of the prerequisites for embodied intelligence to move to household and small and medium-sized enterprise scenarios. However, the specific technical implementation details, performance indicators and mass production progress of this system have not yet been disclosed in public sources, and it is still in the early verification stage.
Progress and Team: Multiple Investment and Financing Roadshows, Founder Has Background in Display Industry and AI R&D
Up to now, the publicly verifiable progress of Yanxin Technology is mainly concentrated in the aspects of financing roadshows and industry activities. In April 2026, Yanxin Technology participated in the "Science and Technology China · Science and Technology Intelligence (Beijing Special Session)" investment and financing roadshow hosted by Beijing Venture Capital Association, Shenzhen Panorama Network Co., Ltd., and Beijing Zhongfu Investment Group; in May, its founder Liu Wenbin demonstrated the general embodied intelligent brain system based on RISC-V architecture at the AI Reception Hall Investment and Financing Matchmaking Meeting in Tsinghua Science Park (hosted by Tsinghua Yuan Sub-district Office and InnoWay); in the same month, the company participated in the "Science and Technology China · Science and Technology Intelligence Special Session" roadshow held in Beijing Fund Town, and carried out docking with investment institutions around technical barriers, market prospects and financing needs; in June, the company appeared at the "Venture Capital China" Science and Innovation Roadshow Hall Embodied Intelligence Special Session co-hosted by the Investment Promotion Agency of Ministry of Commerce of the People's Republic of China and other institutions, and the on-site comment guests included industrial and capital parties such as CCCC Capital, Shoucheng Capital, Zhaowei Group, iFLYTEK, and Changan Automobile.
In terms of the team, public activity materials show that Liu Wenbin, the founder and chief scientist of Yanxin Technology, once served as an image scientist at BOE Xinyihua Group and chief scientist at Hesi Thinking Technology Co., Ltd., and has been engaged in R&D and management of core theories of AGI general artificial intelligence, brain-inspired perception, and interpretable artificial intelligence for many years. The company's legal representative and major shareholder Liang Jian holds 88% of the shares, and Liu Wenbin holds 12% of the shares. The industrial and commercial information shows that the number of employees participating in social insurance in the company is 1, which is in the early stage of development.
Why It Is Worth Paying Attention To: The Intersection Window of RISC-V and Embodied Intelligence
The track where Yanxin Technology is located has two notable macro variables. First, embodied intelligence is generally regarded as the next implementation direction of artificial intelligence, and the robot body supply chain is rapidly maturing, but the "brain" is still the link where value and bottlenecks are concentrated; second, the adoption of RISC-V architecture in edge computing power and robot scenarios is moving from technical discussion to industrial practice. The advantages of open-source instruction set in independent controllability and cost structure make it a potential base for end-side embodied intelligence. Binding "brain software" with "open-source chips" is the differentiated position of Yanxin Technology that distinguishes it from general large model manufacturers and robot body manufacturers.
According to information from multiple roadshows, Yanxin Technology is advancing a new round of financing, and the expected use of financing is concentrated on product R&D and team expansion. The project is currently in the early financing stage, and the specific financing amount and valuation have not been publicly disclosed.