
HEFEI, Anhui, September 2, 2026 — The morning of August 26 brought a significant new chapter for embodied intelligence in Anhui, as the “Robots Enter Lei Street” launch ceremony for Zero Power Robotics’ robot shop scene was held in Lei Street. At the same event, Zero Power Robotics announced that four new robot shops had begun synchronized operation across four major brand districts: Lei Street East District, Lei Street South District, Hechai 1972, and Gong Street. The coordinated launch marks the scaled implementation of Anhui’s first embodied intelligence robot unmanned retail project in Baohe District, signaling a transition from isolated demonstrations toward continuous real-world commercial operation for embodied intelligence.
The launch ceremony drew attention from provincial, municipal, and district officials, as well as senior industry representatives. Among the guests were Kong Zhi, Party branch secretary and general manager of Hefei Guoxian Holding Co., Ltd.; Min Yuheng, founder and chief executive officer of Zero Power Robotics; and Ma Xiaolong, co-founder of Zero Power Robotics. Their remarks centered on the development of the embodied intelligence industry, the opening of innovative application scenes, and the large-scale commercial deployment of robots. The event was not merely a ceremonial opening of four retail points. It represented a broader argument about how embodied intelligence can be developed, validated, and scaled through daily commercial activity.
1. A Scaled Milestone for Embodied Intelligence in Anhui
The four robot shops in Hefei constitute a milestone because they move embodied intelligence out of controlled laboratory conditions and into repeated, variable, public-facing operation. For embodied intelligence, the challenge is not only whether a robot can perform a task once under favorable conditions. The deeper challenge is whether an embodied intelligence system can perform reliably across changing objects, changing environments, changing customer behavior, and changing task states. The new robot shops are designed to confront that challenge directly.
As Ma Xiaolong, co-founder of Zero Power Robotics, explained during the ceremony, the simultaneous operation of four stores in Hefei is not simply an increase in the number of robot retail locations. It is another scaled practice carried out by Zero Power Robotics in a real commercial scene. Lei Street, as a representative urban commercial and consumer scene in Hefei, offers continuously changing foot traffic, merchandise, and consumer demand. Those changing conditions provide a more realistic and diverse operating environment for robots, and they also establish an important foundation for the continuous validation and iteration of robot capabilities in real scenes. For embodied intelligence, this is precisely the kind of environment that cannot be fully replicated by a fixed test bench.
The phrase embodied intelligence describes a class of systems that must perceive, reason, decide, and act within the physical world. Unlike purely digital artificial intelligence, embodied intelligence cannot be separated from the body, the environment, and the consequences of physical action. A robot that recommends a product is engaging in one kind of intelligence. A robot that identifies a product, reaches for it, grasps it, moves it, and delivers it while reacting to real-world variation is engaging in embodied intelligence. The four robot shops place embodied intelligence at the center of ordinary commercial exchange.
2. Four Robot Shops Across Four Brand Districts
The newly launched robot shops are located in four distinct brand districts, each with its own pattern of activity and flow. The distribution matters because embodied intelligence systems are tested not only by the tasks they perform but also by the environments in which they perform them. A retail environment with changing goods, lighting, placement, crowding, and user behavior creates a continuous stream of physical interaction problems.
| Item | Details |
|---|---|
| Project | Anhui’s first scaled embodied intelligence robot unmanned retail project |
| District | Baohe District, Hefei, Anhui |
| Launch event | “Robots Enter Lei Street” — Zero Power Robotics robot shop scene launch ceremony |
| Event date | August 26 |
| New robot shops | Four |
| Locations | Lei Street East District, Lei Street South District, Hechai 1972, Gong Street |
| Key attendees | Provincial, municipal, and district officials; Kong Zhi of Hefei Guoxian Holding Co., Ltd.; Min Yuheng of Zero Power Robotics; Ma Xiaolong of Zero Power Robotics |
| Core theme | Embodied intelligence industry development, innovation scene opening, and scaled commercial robot application |
| Project logic | Robot shops as real-world commercial scenes and real-world data fields for embodied intelligence |
The table above summarizes the factual foundation of the deployment. The larger significance lies in what the deployment enables. Each robot shop is not only a point of sale. It is also a point of data generation, model validation, and operational learning for embodied intelligence. When four shops operate across different brand districts, the embodied intelligence system encounters a broader range of physical situations than a single location can provide.
3. Why Embodied Intelligence Needs Real Commercial Scenes
For embodied intelligence, commercial scenes have a meaning beyond product application. They allow robots to interact and evolve in a continuously changing real world. Min Yuheng, founder and chief executive officer of Zero Power Robotics, framed the launch in exactly these terms. He stated that for embodied intelligence, the significance of a commercial scene is not only to realize product application. More importantly, it allows robots to “interact and evolve” in the continuously changing real world. The “Robots Enter Lei Street” initiative, he said, is a connection among the innovation chain, the industry chain, the application chain, and urban life.
That connection is essential for embodied intelligence because real commercial scenes impose constraints that laboratory settings often remove. In a laboratory, objects may be placed in known positions. Lighting may be controlled. The number of people may be limited. The task may be repeated under similar conditions. In a busy brand district, none of these simplifications can be assumed. Products vary in material, size, weight, and packaging. Their positions shift. Light changes. Customers move. Shelves become occluded. Every pick-and-place action changes the physical environment for the next action. These are not edge cases for embodied intelligence; they are the normal conditions of physical operation.
Zero Power Robotics has therefore chosen to place robots directly into real commercial environments. The goal is to let robots continuously face different products, different positions, different environments, and constantly changing task states. In repeated real grasping and interaction, the system accumulates data. This is the logic behind the robot shop layout. It is also a statement about how embodied intelligence should be advanced: not only through simulation, not only through controlled trials, but through sustained exposure to the physical world.
4. From Robot Shops to Real-World Data Fields for Embodied Intelligence
For embodied intelligence at the current stage, the truly scarce capability is not the successful completion of a single action. The scarce capability is the ability to continuously obtain high-quality, diverse real-world physical interaction data. A robot that can pick up one known object is useful for a demonstration. A robot that can keep learning from many unknown objects and environments is useful for a scalable industry. The robot shop is designed to serve the second purpose.
A product’s material, size, weight, and packaging method differ from one item to another. Its placement, lighting environment, and occlusion relationships change constantly. Even the environmental change caused by each retrieval and placement of a product can create a new physical interaction problem. These variations are difficult to exhaust through pre-set programs. A system built only on predetermined rules will eventually meet a situation that the rules did not anticipate. Embodied intelligence must therefore learn from real interaction rather than rely solely on enumeration.
This is why Zero Power Robotics places robots directly into real commercial environments. The robot must continuously face different commodities, different positions, different environments, and continuously changing task states. Through repeated real grasping and interaction, it accumulates data. The robot shop becomes a long-running entry point for real-world data. As goods, environments, and user behavior continue to change, the physical world continues to present new problems. Every time the robot completes recognition, judgment, grasping, and delivery, it corresponds to an interaction with the real physical world. That interaction provides a new opportunity for learning and validation of model capability.
5. The Data Problem at the Heart of Embodied Intelligence
Scene selection determines data quality. Data quality determines the upper limit of the model. The upper limit of the model ultimately determines commercial scale. This sequence is central to the argument for embodied intelligence. A robot shop that operates every day and continuously generates physical interaction is itself a long-term real-world data entrance. The value is not only in the transactions it completes. The value is in the physical variability it experiences and records.
Embodied intelligence cannot be trained only on clean, static, or simplified data. It must encounter the messiness of physical reality: products that are slightly different from their packaging images, packages that deform, objects that are partially hidden, surfaces that reflect light differently, and human movements that alter the scene. Each of these factors can affect perception, planning, and control. A system that succeeds only when the scene is perfectly arranged is not yet a robust embodied intelligence system.
The robot shops create a disciplined way to confront this variability. They are commercial spaces, so they have real customers and real goods. They are also data spaces, so they produce continuous physical interaction records. The combination is important. A warehouse test may produce data, but it may not reflect consumer behavior. A retail demonstration may attract attention, but it may not operate long enough to reveal long-tail variation. A continuously operating robot shop can do both: it serves customers and it generates the raw material for embodied intelligence improvement.
6. Commercial Scenarios as Living Laboratories for Embodied Intelligence
Commercial scenarios are living laboratories for embodied intelligence because they combine repeatability with variation. A shop repeats similar tasks many times per day: recognize, reach, grasp, move, deliver. Yet each repetition occurs under slightly different physical conditions. That combination is valuable. Repetition without variation produces narrow competence. Variation without repetition produces noisy, disconnected experience. Embodied intelligence needs repeated tasks across varied conditions.
The four robot shops in Hefei expand the range of that variation. Different brand districts may have different layouts, different customer flows, different product assortments, and different environmental conditions. For embodied intelligence, that diversity is not a distraction. It is a training resource. It helps the model move from memorizing specific objects and positions toward generalizing across classes of physical problems.
In this sense, the robot shop is not merely a place where embodied intelligence is displayed. It is a place where embodied intelligence is tested, stressed, and improved. The public sees a robot shop. The development process sees a real-world data field. The industry sees a possible pathway toward scalable embodied intelligence. These perspectives are connected by the same physical operation.
7. From Picking a Single Item to Understanding the Physical World
In the entire process of interaction between embodied intelligence and the physical world, the most basic and most critical entry point is the robot’s grasping capability. Grasping is where perception meets action. It requires the robot to identify an object, estimate its pose, plan a reach, control contact, and adapt to unexpected movement or deformation. A failure in any part of this chain can prevent a successful pick. For embodied intelligence, grasping is not a minor accessory. It is a foundational skill that enables higher-level physical tasks.
In real life scenes, many objects are not fully standardized. Zero Power Robotics aims to solve not the problem of writing one grasping program for each commodity. Instead, it aims to enable robots to gradually acquire autonomous operation capability when facing unknown objects and unknown environments. This is a different objective. A dedicated program can make a robot pick a known item under known conditions. A general capability can allow a robot to approach a new item under new conditions and still find a way to act.
Around this goal, Zero Power Robotics uses real-world physical interaction data as a foundation to continuously train the ZerithV series of embodied native action models. The aim is to move the robot from completing a deterministic Picking action toward a generalizable PhysicalAction. This progression is central to embodied intelligence. Deterministic Picking proves that a robot can execute a known task. Generalizable PhysicalAction suggests that a robot can adapt its physical behavior across tasks, objects, and environments.
8. The Picking Challenge in Real-World Embodied Intelligence
The Picking challenge is difficult because physical reality does not present objects as clean data labels. A bottle may be full or partially empty. A package may be upright or lying down. A product may be positioned near the edge of a shelf or behind another item. Lighting may create glare or shadow. A customer may move an object after the robot has already planned its action. These conditions require more than object recognition. They require embodied intelligence to maintain a continuous relationship between perception and action.
For embodied intelligence, Picking is also a data-rich activity. Every successful grasp provides positive examples. Every failed grasp provides information about mismatch between the model’s expectation and the physical world. Every recovery action provides a sequence of corrective decisions. Over time, this data can help the model learn not only what a successful grasp looks like but also how to adjust when the first attempt does not succeed. That ability to adjust is a hallmark of embodied intelligence in real operation.
Zero Power Robotics is not treating Picking as a one-time technical demonstration. The robot shops allow Picking to occur repeatedly under commercial conditions. The system must handle different products, different positions, different environments, and different task states. The result is a stream of experience that can support model improvement. The more the robot operates, the more physical problems it encounters, and the more opportunities it has to refine its embodied intelligence.
9. Training the ZerithV Embodied Native Action Models
The ZerithV series of embodied native action models represents the technical core of this approach. The models are trained on real-world physical interaction data. The purpose is to help the robot move from deterministic Picking to generalizable PhysicalAction. In practical terms, this means the robot is not expected to rely on a separate handcrafted solution for every product or every shelf arrangement. It is expected to develop a reusable capability for physical action.
This is a challenging objective for embodied intelligence because physical action is sensitive to small changes. A slight shift in object pose can change the required grasp. A change in surface friction can affect lifting. A change in weight distribution can affect stability. A change in occlusion can affect recognition. The model must therefore learn representations that connect visual and spatial information with action outcomes. Real-world data is essential because these relationships are difficult to specify fully in advance.
The robot shops provide a continuous source of such data. They also provide a continuous test of whether the model’s learned capabilities transfer across contexts. A model that works in one shop but fails in another is not yet a scalable embodied intelligence model. A model that can handle multiple shops, multiple product sets, and multiple environmental conditions is closer to the kind of generalizable capability that commercial deployment requires.
10. Physical Grasping API-ization as a Foundation for Embodied Intelligence
Zero Power Robotics proposes the API-ization of physical grasping as a foundational step for embodied intelligence. The idea is that when a robot can consolidate its perception, judgment, and operation capabilities for different objects into a stable and general model capability, grasping will no longer require complex development for every scene. Instead, it can be called directly by different robot bodies and commercial scenes, much like a software interface is called in the digital world.
This concept matters for embodied intelligence because it addresses a central scaling problem. If every new retail location requires a new set of custom grasping programs, deployment remains expensive and slow. If physical grasping can be offered as a reusable capability, then new scenes can be brought online with lower marginal cost. The robot shop is therefore not only a commercial application scene. It is also a real-world practice field for validating the API-ization of physical grasping.
| Concept | Role in Embodied Intelligence |
|---|---|
| Embodied intelligence | Systems that perceive, reason, decide, and act in the physical world rather than only in digital environments |
| Real-world physical interaction data | Data generated when a robot touches, moves, grasps, and delivers objects under changing physical conditions |
| Picking | A deterministic grasping action that completes a known task under specific conditions |
| PhysicalAction | A generalizable physical action capability that can adapt across objects, positions, and environments |
| ZerithV | A series of embodied native action models trained on real-world physical interaction data |
| Physical Grasping API-ization | The effort to make physical grasping a stable, reusable capability callable by different robot bodies and scenes |
| N-Power Plan | A plan to let the real world serve as infrastructure for continuous model evolution in embodied intelligence |
The table above places the key concepts in relation to one another. Embodied intelligence requires real-world physical interaction data. That data supports models such as ZerithV. Those models aim to move from Picking to PhysicalAction. The API-ization of physical grasping is the attempt to turn that capability into a reusable foundation. The N-Power Plan is the broader strategy for allowing the real world to serve as infrastructure for continuous model evolution.
11. The N-Power Plan and the Infrastructure of Embodied Intelligence
The N-Power Plan is built on a simple but demanding premise: let the real world become infrastructure for the continuous evolution of models. For embodied intelligence, this means the real world is not only the place where a finished model is deployed. It is also the place where the model continues to learn, validate, and improve. A single robot that can grasp a product proves a technical capability once. The real question for embodied intelligence is whether the same capability can operate stably across stores, products, and environments and be replicated continuously at sufficiently low deployment and operational cost.
The four new robot shops in Hefei are important because they increase the number of operating points. As the number of points increases, the variety of products, environmental changes, and real interactions faced by Zero Power Robotics also increases. More scenes bring more high-quality real data. More real data drives continuous improvement in model capability. Improved model capability further reduces the marginal cost of deploying new scenes and supports more points entering real operation. This creates a continuous cycle: scene, data, model, stronger generalization ability, and more scenes.
This is the scaled pathway for embodied intelligence that Zero Power Robotics is exploring. It is not a pathway that waits for a perfect model to appear and then searches for commercial scenes. It is a pathway in which the model and the scenes grow together in the real commercial world. The robot moves from being able to grasp toward being able to generalize. Eventually, Picking in the physical world can become a foundational capability that can be called at scale. That is the next destination of the N-Power Plan.
12. The Closed Loop: Scenes, Data, Models, Generalization, and More Scenes
The closed loop described by Zero Power Robotics is central to understanding why the Hefei deployment matters for embodied intelligence. A single robot shop can generate data. A network of robot shops can generate diverse data. Diversity is important because embodied intelligence must generalize across conditions. If all data comes from one environment, the model may overfit to that environment. If data comes from multiple brand districts, multiple product sets, and multiple customer patterns, the model has a better chance of learning transferable physical capabilities.
The loop works as follows. More scenes create more interactions. More interactions create more high-quality real-world data. More data supports model training and validation. Better models improve generalization. Better generalization reduces the cost and risk of entering new scenes. Lower cost and risk make it possible to add more scenes. The cycle then repeats. This is not a one-time linear process. It is an operating system for embodied intelligence development.
For embodied intelligence, the closed loop also provides a way to manage the long tail. The long tail of physical interaction includes rare objects, unusual placements, unexpected lighting, and uncommon customer actions. These cases may not appear often in a single shop, but across multiple shops and longer operating periods, they become more likely to appear. Capturing and learning from them is necessary for robust embodied intelligence. The robot shop network is designed to encounter and absorb that long tail over time.
13. Scaling Embodied Intelligence Without Waiting for a Perfect Model
A key insight of the N-Power Plan is that scaling embodied intelligence does not require waiting for a perfect model. Instead, it requires a system that can improve while operating. The robot shops allow operation and learning to occur together. The system does not need to solve every possible physical interaction before deployment. It needs to be able to encounter new interactions, collect data, and improve over time.
This approach reflects the reality of embodied intelligence. The physical world is too varied to be fully specified in advance. A model that is perfect in simulation may still encounter unexpected conditions in a real shop. A model that is imperfect but continuously improving in real operation may ultimately be more valuable. The robot shops provide the operational context for that improvement. They also provide commercial value while the improvement occurs.
For embodied intelligence, this is a pragmatic path. It connects technical development with real deployment. It connects data collection with customer service. It connects model iteration with commercial scale. The four robot shops in Hefei are therefore not only a retail experiment. They are an example of how embodied intelligence can be developed through use rather than only through preparation.
14. Leadership Perspectives on Embodied Intelligence and Urban Life
The launch ceremony included remarks from leaders and company executives who framed the project within broader questions of embodied intelligence, innovation, and urban life. Min Yuheng described the “Robots Enter Lei Street” initiative as a connection among the innovation chain, the industry chain, the application chain, and urban life. He also stated that Zero Power Robotics will continue to root itself in Anhui, advance research and development of embodied intelligence technology and commercial implementation, and let robots begin from real scenes and real tasks. The goal is to continuously improve the ability to serve industry, cities, and daily life.
Ma Xiaolong emphasized that the four-store launch is not a simple addition of robot retail points. It is another scaled practice in a real commercial scene. Lei Street, as a representative urban commercial and consumer scene in Hefei, provides continuously changing foot traffic, goods, and consumer demand. Those conditions give robots a more realistic and diverse operating environment. They also provide an important foundation for continuous validation and iteration of robot capabilities in real scenes. For embodied intelligence, this is the kind of foundation that supports long-term progress.
The presence of provincial, municipal, and district officials, along with representatives from Hefei Guoxian Holding Co., Ltd., reflects the broader interest in embodied intelligence as an industry direction. The event brought together participants around topics such as embodied intelligence industry development, innovation scene opening, and scaled commercial application of robots. These topics are connected. Innovation scenes provide the conditions for testing. Commercial application provides the conditions for scale. Industry development provides the ecosystem for sustained progress.
15. What the Hefei Deployment Means for the Future of Embodied Intelligence
The Hefei deployment means that embodied intelligence in Anhui has moved from a conceptual or demonstration phase toward scaled commercial operation. The project is described as Anhui’s first scaled embodied intelligence robot unmanned retail project in Baohe District. That description matters because scale changes the nature of the problem. A single robot shop tests whether a robot can work. Multiple robot shops test whether the system can operate, be maintained, and improve across locations.
For embodied intelligence, the four shops create a network effect at the data level. Each shop contributes its own physical interactions. Together, they provide a broader distribution of objects, positions, environments, and user behaviors. This broader distribution is essential for training models that can generalize. It is also essential for validating whether the API-ization of physical grasping can work across different robot bodies and commercial scenes.
The robot shop is therefore both a commercial application and a validation field. It is a place where embodied intelligence serves customers. It is also a place where embodied intelligence learns from the physical world. The dual role is important. If the project were only commercial, it might optimize for short-term transactions. If it were only technical, it might not encounter the full variability of real operation. By combining both, the robot shop creates a practical engine for embodied intelligence progress.
16. From One Grasp to One Shop, From One Shop to a City
The vision described by Zero Power Robotics moves from one grasp to one shop, from one shop to a city, and from every interaction in the real world toward a scaled and callable PhysicalAPI. This progression summarizes the logic of embodied intelligence in commercial operation. A single grasp demonstrates capability. A shop demonstrates repeatability. A city-wide network demonstrates scalability. A callable physical API demonstrates a new layer of infrastructure for embodied intelligence.
The path is not linear in practice. It requires continuous data collection, model training, validation, deployment, and operational learning. It requires scenes that are diverse enough to produce meaningful data and stable enough to support daily operation. It requires a model that can improve without disrupting service. It requires a commercial environment that can provide real tasks and real feedback. The robot shops in Hefei are designed to bring these requirements together.
For embodied intelligence, the next stage is not only about making robots more capable in isolated tasks. It is about making physical capability reusable, transferable, and scalable. The API-ization of physical grasping is one expression of that goal. The N-Power Plan is a broader expression. The four robot shops are an early operating example. They show how embodied intelligence can be placed in the middle of daily commercial life and allowed to learn from the physical world while serving it.
17. Conclusion: Embodied Intelligence Enters Daily Commercial Operation
The launch of four Zero Power Robotics robot shops across Lei Street East District, Lei Street South District, Hechai 1972, and Gong Street marks a significant moment for embodied intelligence in Anhui. The project is described as Anhui’s first scaled embodied intelligence robot unmanned retail project in Baohe District. It brings together real commercial scenes, real products, real customers, and real physical interaction. For embodied intelligence, this combination is essential.
The robot shops are not only retail points. They are real-world data fields for embodied intelligence. They generate the physical interaction data needed to train and validate models such as the ZerithV series of embodied native action models. They support the movement from deterministic Picking to generalizable PhysicalAction. They test the idea of physical grasping API-ization. They contribute to the N-Power Plan, in which the real world becomes infrastructure for continuous model evolution.
As more points operate, more products, environments, and interactions enter the system. More high-quality real data supports stronger model capability. Stronger model capability reduces the marginal cost of new deployments. New deployments generate more scenes and more data. This closed loop is the scalable pathway for embodied intelligence that Zero Power Robotics is exploring. The robot shop is where that pathway becomes visible, measurable, and commercially real.
From a single grasp to a shop, from a shop to a city, and from every real-world interaction toward a scalable PhysicalAPI, the work in Hefei represents a practical step for embodied intelligence. It shows that embodied intelligence can be developed not only by waiting for a perfect model, but by operating in the real world, learning from real tasks, and growing with real commercial scenes. The four robot shops in Baohe District are a beginning of that next stage for embodied intelligence in Anhui and beyond.
