NextFin News --  Industrial embodied intelligence startup Guangxiang Technology has completed multiple rounds of funding—seed, angel, and angel in half a year since its inception, raising a cumulative total of more than RMB 100 million.

The financing was co-led by top-tier financial investors IDG Capital and Dongfang Fuhai, with participation from robot-industry strategic investor EFORT, as well as 01VC, Datai Capital, and the L2F Entrepreneurs Fund (Guangyuan). Proceeds from this round will be primarily used for R&D of core embodied-robot technologies, accelerating productization, and commercialization and delivery.

Guangxiang Technology was founded in April 2025 by Zhang Tao, former Technical Director at Alibaba’s Amap, together with Li Shengbo, a professor of Tsinghua University, a leading expert in artificial intelligence. It is an embodied intelligence company jointly incubated by Tsinghua University’s School of Vehicle and Mobility and School of Artificial Intelligence. The company has already become an embodied-intelligence strategic partner to multiple leading global automotive OEMs. With its embodied model built around enabling “robots to learn on their own” plus its embodied platform aimed at enabling “large-scale deployment of embodied intelligence,” it is helping industrial manufacturing scenarios such as automotive and 3C build a general-purpose industrial embodied brain, driving intelligent upgrades across industry.

The industrialization team led by founder and CEO Zhang consists of technical executives from major tech companies including Alibaba, Tencent, Huawei, and Geek+. They have successfully led the mass production and deployment of world-leading spatial perception and localization technologies across millions of in-vehicle terminals, delivered 56,000 autonomous mobile robots worldwide, and achieved a 10x efficiency improvement in lean production of high-definition maps.

Co-founder and Chief Scientist Li is an internationally recognized expert in AI. He has published more than 200 papers with over 22,600 citations, won 12 best-paper awards from leading academic venues in China and abroad, and was selected as an Elsevier China Highly Cited Researcher for four consecutive years. The DSAC series of reinforcement learning algorithms he led has reached internationally leading state-of-the-art (SOTA) performance, and he also spearheaded the development of China’s first fully neural-network, end-to-end autonomous driving system (IDrive).

The core technical team is composed entirely of PhDs from top universities such as Tsinghua University and Zhejiang University, spanning the full embodied-intelligence stack—including reinforcement learning, visual perception, and optimal control—and has earned top-tier honors such as three consecutive championships in an international localization competition, a gold medal in the CVPR IMC Challenge, and first place on the Argoverse autonomous driving leaderboard across both datasets.

When Zhang first decided to jump into embodied-intelligence entrepreneurship, a common refrain in the industry was that robot companies that prioritized vertical, scenario-specific applications would eventually be eclipsed by general-purpose robot companies. Zhang, however, came to a very different conclusion: he likened vertical industrial robots and general-purpose robots to L2 and L4 autonomous driving respectively, arguing that the robotics industry—like autonomous driving—will go through a long development cycle. Starting with vertical scenarios and then transitioning progressively toward all-scenario generalization is the more viable commercial path.

Based on this belief, Guangxiang Technology set its sights on wheeled industrial robots, focusing on automotive manufacturing scenarios. In Zhang Tao’s view, industrial operations combine a “standardized environment + complex manipulation,” making them both highly challenging and capable of rapid deployment today. Within industry, automotive manufacturing is the most representative track with ample market headroom—Guangxiang Technology estimates that simply intelligentizing the final assembly process in automotive production alone represents a market on the scale of hundreds of billions of yuan, and it can be quickly replicated and extended to nearly all industrial manufacturing scenarios.

As for the robot’s form factor, Zhang’s reasoning is equally straightforward: the biggest advantage of bipedal humanoid robots is their ability to overcome terrain obstacles. But in factories—highly standardized environments—this advantage has little room to play out, while shortcomings such as high energy consumption and less precise localization may be magnified. By contrast, wheeled robots consume less power, localize more accurately, and align far better with the practical needs of factory environments.

Faced with industrial scenarios’ stringent multi-dimensional requirements for operational precision, cycle time, and motion smoothness, Guangxiang Technology’s core strategy is to “build self-learning intelligent models for industrial use.”

On the model-architecture side, Guangxiang Technology developed a high-smoothness neural network architecture purpose-built for industrial manipulation, enabling robots to output motions that are highly accurate, highly reliable, and exceptionally smooth. On the training side, the company moved away from easier-to-implement imitation learning and instead adopted reinforcement learning, which is harder but has greater upside. Zhang noted that while imitation learning can quickly reach a 90%–95% success rate with a small amount of data, it cannot meet industrial scenarios’ requirement of near-100% success—precisely what is essential to ensuring high-quality automotive manufacturing.

To address the industry pain point of scarce real-robot data in embodied intelligence, Guangxiang Technology proposed increasing the proportion of simulated data used in model training. Leveraging the team’s years of accumulated high-precision scene-modeling capabilities and industrial customers’ high-precision digital modeling resources, the company continues to narrow the gap between simulation and real-robot data, building a complete training pipeline from simulation to real-world deployment.

In addition, GuangxiangTechnology independently developed the GOPS platform, fully modularizing embodied-intelligence model design, development, training, and even debugging for industrial scenarios. This creates a stable, efficient, end-to-end model development pipeline, giving enterprises true large-scale delivery capabilities.

At present, Guangxiang Technologies has forged in-depth strategic partnerships with several internationally renowned automotive companies and has successfully completed the first-phase POC validation on real production stations. This means that Guangxiang ’s technical solution has moved beyond the lab and has truly withstood the test of industrial operations—progressing from theoretical feasibility to engineering deployment. In less than a year, Guangxiang completed a pivotal step that many comparable companies need several years to cross.

Looking ahead, Guangxiang Technologies plans to enter at least ten automakers over the next three years, deploy more than a thousand intelligent robots that meet factory requirements, and extend its product capabilities horizontally to a broader range of industrial manufacturing scenarios such as 3C and heavy industry. In its longer-term strategic blueprint, industry is only the starting point—not the destination: as its embodied-model capabilities continue to iterate and the GOPS platform is replicated at scale, Guangxiang Technologies will gradually expand into large commercial scenarios. Following a progressive path of “deep vertical specialization—full industrial coverage—general-purpose embodiment,” it will steadily move closer to true general-purpose embodied intelligence.