IDG Capital leads the angel round of financing. Alibaba-related and Tsinghua embodied projects have raised over 100 million yuan in half a year | First exclusive report by 36Kr

36Kr has learned that Guangxiang Technology, an industrial embodied intelligence enterprise, has completed multiple rounds of financing, including the seed round, angel round, and angel + round, with a cumulative amount exceeding 100 million RMB. The financing was jointly led by financial investment institutions IDG Capital and Orient Fortune Capital, and followed by robot industry capital EFORT, Zero One Capital, Datai Capital, and L2F Entrepreneurs Fund.

It is reported that the enterprise's financing funds will be mainly used for the R & D of core technologies of the company's embodied robots, the promotion of productization, and the commercial delivery work.

Guangxiang Technology was founded in April 2025 by Zhang Tao, the former technical director of Alibaba's AutoNavi, and Li Shengbo, a professor at Tsinghua University and an expert in the field of artificial intelligence. At present, Guangxiang Technology has become a strategic cooperation partner for embodied intelligence of many global automobile manufacturers. Based on the embodied model that enables "robots to learn by themselves" and the tool platform that enables "large - scale implementation of embodied intelligence", Guangxiang Technology hopes to "help industrial manufacturing scenarios such as the automotive and 3C industries build a general industrial embodied brain".

Starting from the automotive manufacturing scenario and gradually transitioning to general - purpose robots

During the period when Zhang Tao decided to engage in the entrepreneurship of embodied intelligence, there was such a voice in the industry - robot companies that enter vertical scenarios first will be covered by companies that directly develop general - purpose robots in the future.

However, Zhang Tao holds a different view. He compares robots in vertical scenarios such as industry with general - purpose robots to L2 and L4 in the field of autonomous driving respectively. "If the technology develops fast enough, L4 can indeed cover all L2 scenarios", but we believe that the robot industry will go through a long development cycle like autonomous driving. Therefore, starting from vertical scenarios and gradually transitioning to general - purpose robots for all scenarios is a more feasible business path."

Based on this thinking, Zhang Tao targeted the direction of wheeled industrial robots at the beginning of his entrepreneurship.

In his opinion, industrial operations belong to "standard environment + complex operations, which is a challenging but quickly implementable scenario at present". In the industrial field, automotive manufacturing is the most typical track with a large enough market space. Guangxiang Technology has estimated that just the intelligentization of the final assembly process in automotive manufacturing has a market scale of tens of billions, and it can be quickly replicated and extended to almost all industrial manufacturing scenarios.

Once the application scenario is selected, the form of the robot can also be determined accordingly.

When explaining to 36Kr why he decided to develop wheeled robots, Zhang Tao said: "The greatest advantage of bipedal humanoid robots is their ability to overcome terrain obstacles. However, in the standardized environment of a factory, the advantages of bipedal humanoid robots cannot be reflected, and the defects of high energy consumption and inaccurate positioning may be magnified. Wheeled robots have low energy consumption and more accurate positioning, which are more suitable for the factory environment and requirements."

The market for automotive manufacturing robots has great potential, but it may not be easy to obtain an entry ticket.

Zhang Tao told 36Kr that industrial robots are not like the demonstration - type robots on the Spring Festival Gala stage. For demonstration - type robots, people are more concerned about whether they can complete actions. However, the measurement standards for the operation tasks of robots in industrial scenarios are actually quite strict. "For example, industrial robots need to balance various operation indicators such as action accuracy, time rhythm, and smoothness of actions."

Moreover, there is a strong interaction between industrial robots and the environment. "Robots need to perceive the environmental state and the state of the operation object in real - time, plan and execute actions in real - time, and avoid collisions during the operation process." All these pose challenges to the construction of operation - type models.

Committed to enabling industrial embodied robots to learn and evolve by themselves

One of Guangxiang Technology's coping strategies is to "build a self - learning intelligent model for the industry".

In terms of model structure, Guangxiang Technology has developed a highly smooth neural network structure specifically for industrial operations, aiming to enable robots to achieve high - precision, high - reliability, and high - smoothness action output. When training the model, Guangxiang Technology abandoned the more easily achievable imitation learning and adopted reinforcement learning, which has more potential but also greater challenges.

Zhang Tao said that although imitation learning can "quickly achieve a seemingly good operation effect with a small amount of data, such as achieving a 90% - 95% success rate in simple PnP tasks", it cannot guarantee a success rate close to 100% as required by the industry, nor can it meet the multi - dimensional performance requirements such as efficiency and accuracy at the same time. These requirements are precisely the key to ensuring high - quality automotive manufacturing.

Therefore, the Guangxiang team hopes that through the model training method of reinforcement learning, robots can have "sustainable self - learning ability to evolve", so as to open up a technical path for robots to continuously improve their performance and ultimately meet a series of strict requirements of the automotive manufacturing scenario for robots.

The training of the model depends on data, but in the field of embodied intelligence, the scarcity of real - machine data is a problem that troubles most players in the industry.

In response, Guangxiang Technology proposed to increase the proportion of simulation data in model training, and rely on the high - precision scene modeling ability and the advantages of high - precision digital models of industrial customers to narrow the gap between simulation data and real - machine data, so as to open up the model training link from simulation to real - machine.

When explaining why to increase the proportion of simulation data, Zhang Tao said: "If we are just making a demo, setting up a fake workstation and collecting some real - machine data may work. However, if we want to achieve large - scale implementation in the future, the massive real - machine data required to achieve an extremely high success rate may be unbearable for us."

The GOPS platform is another preparation made by Guangxiang Technology for the large - scale implementation of robots.

In Zhang Tao's words, this platform modularizes the design, development, training, and even debugging of the embodied intelligence model in industrial scenarios. It can build a stable and efficient link. Under the premise of clear scenario tasks, it can achieve high - quality end - to - end model development for any industrial scenario, enabling the enterprise to have "large - scale delivery capabilities".

Currently, Guangxiang has reached cooperation with several automotive enterprises and completed the first - phase POC verification for real production workstations. Looking to the future, Guangxiang Technology proposes that within the next three years, it will enter at least ten automotive manufacturers and deploy thousands of intelligent robots that meet the factory's needs. At the same time, Zhang Tao hopes that the company's products will be widely applied to the production and manufacturing workstations in other industrial scenarios.

In Zhang Tao's longer - term plan, Guangxiang Technology also intends to enter other large - scale industrial and commercial scenarios outside of manufacturing and move towards general embodied intelligence through a gradual path.

Author's WeChat: luckg17305264638