Qingyan Precision raised hundreds of millions in funding, as a state-backed equipment manufacturing team invested.

Investment界 AI learned that today (July 13), Qingyan Precision announced that it has quickly completed two rounds of funding worth several hundred million yuan within June; with this, the company’s Series B round has been officially locked in.

“A state-owned team + half of the auto-circle” emerges: the several-hundred-million-yuan B2 round is led by Xingyuan Capital, with FAW Fusheng participating; the following B3 round is led by BAIC Industry Investment, with Yulon Group participating. In addition, the China National Machinery Industry Industrial Fund was newly added.

As early as June 2026, the Ministry of Industry and Information Technology and the State-owned Assets Supervision and Administration Commission jointly launched the “Special Action for Humanoid Robots and Embodied Intelligence Practical Training in Real Scenarios,” requiring that embodied intelligence cannot only run in laboratories—it must enter real factory workstations,开启“作业模式” (work mode).

Before that, Qingyan Precision had already positioned itself around the physical AI engineering base. Through eight years of accumulation in industrial sites, embodied robots “learn to work” in real, complex, and harsh industrial scenarios—truly enabling deployment.

Rare moves by central SOE capital

Looking around, the industrial resources involved in Qingyan Precision’s current round of financing are extremely rich.

Among them are central SOE funds—such as the China National Machinery Industry Industrial Fund.

Even rarer is the formation of a car-industry capital matrix that is not common: the entire Series B round gathers six automakers—BAIC Industry Investment, Xingyuan Capital, FAW Fusheng, Great Wall Capital, Shaanxi Automotive Industry Capital, and Yulon Group. Dense capital injections from automakers indicate that Qingyan Precision’s physical AI engineering base and test/verification system have already been embedded into domestic mainstream automakers’ core supplier chains. This is recognition from both upstream and downstream of the auto industry chain.

An investment lineup that is highly vertical and strongly industrial in nature proves that the investment logic in the second half of embodied intelligence has already shifted: capital is no longer blindly chasing demo videos of humanoid robots; instead, it is heavily betting on physical AI infrastructure companies that master real industrial scenarios, have high-quality data closed-loop capability, and possess engineering deployment ability.

For physical AI to truly land, it must cross key links such as product development, supply chain, on-site delivery, customer service, and ongoing operations and maintenance. In other words, it needs real trials—able to make production lines usable.

Only by tightly binding capital with business can it ensure a continuous, stable entry point to real industrial scenarios, thus forming a virtuous cycle.

As mentioned in the “Special Action for Humanoid Robots and Embodied Intelligence Practical Training in Real Scenarios,” by end of 2026, humanoid robots and other key products will first complete application verification and routine deployment in a batch of representative scenarios,开启作业模式; and will refine and form more than one hundred high-value application scenarios, further enriching the application spectrum of embodied intelligence and driving the formation of deployment capability at the ten thousand-unit scale.

Qingyan Precision can be said to have precisely positioned itself. These two rounds of financing both come with key turns: starting from running through the closed loop of new energy physical intelligence, it gradually moves toward broader industrial scenarios, aiming to build an industrial physical AI engineering base and deeply cultivate the embodied intelligence field.

From this perspective, its breakthrough is not merely a single-point technology. It is a compound moat formed together by real-scenario entry, data production capability, test & evaluation systems, engineering delivery capability, and world-model capability—and, even more, it completes full-chain layout in advance before policy arrives.

Tsinghua, Stanford, and veterans from the robotics industry—strong synergy

Qingyan Precision’s founder and CEO Dong Han, a PhD, studied at Tsinghua University under Professor Li Keqiang, an academician of the Chinese Academy of Engineering. He formally founded Qingyan Precision in June 2018 under the incubation of Tsinghua University.

Over its eight-year history, Qingyan Precision has brought AI inspection, simulation, and test/verification products into the core supplier chains of virtually all domestic vehicle manufacturers and power battery enterprises, shipping more than ten thousand units, deploying in 30+ countries, and covering core tracks including new energy vehicle OEMs, power batteries, energy storage, core components, mines, and electricity, among others.

(From left to right

Qingyan Precision’s embodied intelligence segment—Precision Vision CEO Cao Qitong, with an academic background in engineering from Stanford University. She previously conducted interdisciplinary research on life sciences and AI at Stanford Computer Science Research Institute, and related results were published as first author in a Nature sub-journal. At Qingyan Precision, Cao Qitong mainly coordinates the company’s technology transfer and iteration roadmap as well as the realization of business scenarios, highlighting the company’s core advantage in conquering the “last mile” of industrial embodied intelligence deployment.

Her core research areas involve the state-evolution laws of high-dimensional, multi-modal, and dynamic data reasoning systems. When transferred to industrial scenarios, the fundamental problem is similar: what a robot sees is not just one workpiece, but a dynamic physical system composed together by vision, force perception, touch, process parameters, and environmental variables. This closely matches the industrial physical world model built by Qingyan Precision.

Qingyan Precision’s embodied intelligence chief engineer and Precision Vision CTO Zhao Ran previously served as the负责人 for embodied Infra at two hundred-billion-yuan-level leading embodied companies, Qianxun Intelligent and ZhiPingang Technology. Zhao Ran’s addition provides solid assurance for Qingyan Precision to build embodied infrastructure and support engineering-scale deployment. As a team member of robot field authority Ding Han, Zhao Ran has focused on robotics for more than ten years, combining solid academic grounding with practical experience in industrial deployment.

He previously led the team to build, from 0 to 1, platforms for teleoperation, data acquisition, the underlying data closed-loop, and simulation. Over ten-plus years of robotics technical accumulation, he can more systematically connect key links such as the robot body, data, simulation, and models, forming the core capabilities needed for building embodied intelligence infrastructure. His platform-based, engineering-based experience and the team’s deep R&D accumulation form synergy, further driving a deep integration between “top-domain” academic genes and “grounded” industrial engineering capabilities.

Since then, the team now combines world-class forward-looking vision, industrial engineering-grade foundation, and hundred-billion-yuan commercial validation in one—already standing at the forefront of China’s embodied intelligence industrialization, becoming the industry’s widely recognized “benchmark for technology” and a “leader in deployment.”

Industrial physical AI engineering base

On top of these foundations, Qingyan Precision has successfully completed a strategic upgrade and capability spillover—rising from a new energy vehicle inspection company to a physical AI engineering base. It is set to serve as the physical AI base for embodied intelligence to land in the industrial field.

In line with the “Special Action for Humanoid Robots and Embodied Intelligence Practical Training in Real Scenarios,” Qingyan Precision’s industrial sites accumulated over many years are already in place. In different industrial domains, the company has deployed more than 2,000 industrial perception nodes on real workstations. From PACK inspection of new energy power batteries to full vehicle final assembly, from surface factories to underground mines, key workstations are transformed into data fields and training fields for embodied intelligence. These scenarios have data, have workstations, and involve real work—making them the most able to validate value.

An embodied model is the “brain,” and Qingyan Precision provides the training bases and “textbooks” that allow the brain to learn “body coordination” and validate its capability. It doesn’t build robots (the body), but it creates robots’ ability to work in industrial sites.

In addition, the “Special Action for Humanoid Robots and Embodied Intelligence Practical Training in Real Scenarios” mentions that it will insist on application-driven development—training in real scenarios to continuously optimize embodied intelligence model algorithms and accumulate high-quality real-machine data.

And now, Qingyan Precision is, in effect, a provider of physical AI data bases.

Qingyan Precision independently developed the TsingLoop multi-modal data engineering pipeline—it takes raw signals distributed across multiple systems and, through unified time-space-semantic alignment, converts them into standardized, reusable data asset packages. Data collected in one operation, after being processed by the pipeline, upgrades raw data into industrial “data assets.” Historical data can automatically merge with newly added data and be iterated continuously, forming a continuously growing data flywheel.

Additionally, based on the TsingLoop multi-modal data engineering pipeline, Qingyan Precision is building a Robot-in-the-Loop testing system for industrial scenarios.

This system can be understood as the industrial version of an “acquire-simulate-verify-evaluate-iterate” closed loop: robots or workers execute tasks at real workstations, while TsingLoop simultaneously captures multi-modal data such as vision, force perception, touch, trajectories, process parameters, equipment status, and execution results. Then, the system reconstructs a digital-twin scenario based on real data, replays historical operating conditions in a simulation environment, reproduces anomalous samples, and performs low-cost, high-frequency hypothesis reasoning for different action strategies.

But simulation is not the endpoint. Industrial robots ultimately must enter real workshops. Therefore, Qingyan Precision will further introduce Robot-in-the-Loop testing: forming a closed-loop interconnection among real robot bodies, controllers, end effectors, sensors, and simulated scenarios, so that action strategies, force-control boundaries, safety envelopes, and abnormal takeover mechanisms can be verified in advance without directly occupying the customer’s production lines.

After deployment on-site, the evaluation module will continuously output standardized evaluation reports, including task success rate, cycle time, anomaly rate, collision risk, energy consumption, and stable operation duration, among other metrics. These evaluation results are not only grounds for acceptance; they also feed back into the TsingLoop data pipeline, driving continued model optimization and ongoing strategy updates.

Systematically, it answers three more critical questions: can tasks be completed stably under real operating conditions? can it pass customer acceptance? can it be reused on the next production line? With this, a data base is achieved.

So far, Qingyan Precision has described its end-state vision: “one base, one brain, and hundreds of vertical scenario application.” With a data engineering system as the base and an industrial cognition world model as the brain, it will deposit reusable physical intelligence across more than one hundred industrial tasks with clearly defined boundaries, such as in power electricity, construction machinery, new energy manufacturing, mines, and more.

At a key moment when physical AI moves from concept to industrial deployment, industrial capital has been betting on Qingyan Precision. What they value is its irreplaceable scenario-deployment capability.

While the industry is still debating algorithm routes, Qingyan Precision, rooted in industrial sites and quietly forging an engineering base for physical AI, has already quietly become the most core “equipment shoveler” in the era of embodied intelligence.

In the second half, the importance of this is self-evident.