Embodied intelligence has been a buzzword for years, but who has actually stepped into the factories?
The fervor for embodied intelligence at WAIC in July has yet to fully subside, while the industrial showcase at the WRC (World Robot Conference) in August was quietly kicking off. As these two major events unfold in succession, we observe increasingly bustling booths and more spectacular tech demonstrations from manufacturers. However, looking closely, most exhibitors are showcasing dancing robots and coffee-making machines, while genuine embodied intelligence solutions operating stably on actual factory production lines remain scarce.
The significant gap between demonstration and actual deployment stems from poor "software-hardware synergy." Traditional industrial robots possess robust hardware capabilities but lack intelligence, while AI algorithms, despite their learning capacities, struggle to integrate seamlessly with the robot's physical body, creating a profound chasm between the two.
However, at WAIC, we noted that Efort and Qizhi Robotics presented a breakthrough approach: "building a solid foundation and achieving software-hardware synergy." Leveraging nearly two decades of accumulation in autonomous and controllable hardware, Efort pairs its expertise with Qizhi Robotics' general-purpose technology foundation to create a closed loop of data, perception, decision-making, and execution, enabling true software-hardware synergy. If embodied intelligence at WAIC largely remained at the stage of conceptual demonstrations and capability validation, the keyword for Efort & Qizhi at WRC 2026 is "practical deployment." With WRC 2026 officially opening on the 19th, the production line deployment solutions on display serve as the ultimate proof of whether the general-purpose technology foundation has truly been validated.
At this year's WRC, Efort and Qizhi Robotics co-exhibited, showcasing content that can be categorized into two types: newly launched products making their public debut, and industrial embodied intelligence workstations already deployed and validated at OEM (Original Equipment Manufacturer) factories. The highlights of this release include the next-generation composite robot and the new HALO Human Skill Collection Suit. Of even greater significance are the intelligent welding and intelligent painting workstations built upon Qizhi's general-purpose technology foundation, which will demonstrate how they address specific pain points in industrial scenarios.
1. New Product Launch: Composite Robot and HALO Human Skill Collection Suit
01. Next-Generation Composite Robot
Composite robots have long been confined to the simplistic integration logic of "mobile chassis + robotic arm." Under this framework, the chassis handles mobility, the arm handles grasping, and the vision module operates independently. The underlying systems of these three components remain disconnected, resulting in insufficient coordination accuracy and low environmental fault tolerance. Consequently, most can only perform simple material transfer tasks and fail to penetrate the core operational processes of the production line.
The newly launched next-generation composite robot from Qizhi Robotics truly achieves trinity collaborative control of "hands, eyes, and feet." At the underlying control system level, it enables deep integration between the mobile chassis, robotic arm, and visual perception. A single unit can independently execute the full closed-loop process of "precise positioning - scene recognition - path planning - fine operation." This capability allows it to handle more complex production line tasks such as high-density material sorting and precision assembly with fastening, directly expanding the efficiency boundaries of flexible manufacturing.
Unlike numerous exhibition pieces that remain at the laboratory prototype stage, this composite robot has already been deployed at an OEM factory and is currently undergoing testing and validation.
At the exhibition, this capability was demonstrated across two complete production line scenarios.
The first is the 4-in-1 intelligent flexible assembly workstation, targeting high-end sectors such as automotive manufacturing. It fully replicates the entire process of material loading by the composite robot, intelligent assembly and fastening by the dual-arm robot, and visual quality inspection by the collaborative robot. The greatest challenge of this solution lies in ensuring the stable and precise operation of three robots with different configurations under visual guidance, force-controlled grasping, and precision assembly. Leveraging Qizhi’s general-purpose technology foundation to unify interfaces and standardize the development environment, developers can decouple and integrate different hardware and process packages. This allows factories to rapidly build intelligent workstations tailored to their specific production lines, much like "playing with LEGO."
The second is the Lineside Warehouse Intelligent Sorting Workstation (SPS). According to Qizhi representatives, this workstation has also been deployed at an OEM factory and is currently undergoing intensive development, testing, and validation. The composite robot can move flexibly within narrow aisles, enabling an unmanned closed-loop delivery from incoming material sorting to finished product loading. Utilizing 3D intelligent recognition, it can precisely grasp hundreds of types of automotive parts, streamlining the entire process of automatic material loading, intelligent sorting, finished product loading, and automatic bin return, while also achieving automatic task dispatch and real-time data traceability.
These two scenarios are not isolated customized workstations; rather, they have evolved into standardized, upgradable, and scalable "lights-out" smart factory templates. Once validated on a single line, the solution can be replicated across additional production lines at minimal cost, with low marginal costs for subsequent expansion. This represents the most compelling advantage of this exhibit.
02. HALO Human Skill Collection Suit
If the composite robot serves as the execution terminal for embodied intelligence on the production line, the concurrently launched HALO Human Skill Collection Suit acts as the core data entry point for embodied intelligence on the supply side.
Although the concept of embodied intelligence has been around for years, data remains the core bottleneck for large-scale deployment. For robots to replicate human operational skills, they require massive amounts of high-quality, multi-modal training data. However, traditional data collection methods are costly and inefficient. Furthermore, critical data such as force control, motion timing, and posture adjustments during human operations are difficult to fully capture using teach pendants or single-vision systems. This directly forces robot skill development to rely on scene-by-scene debugging, resulting in weak generalization capabilities and persistently high deployment costs.
The HALO suit is designed precisely to address this common industry pain point. Serving as the front-end collection core of the HumanGPT human skill foundation model ecosystem, this wearable device can accurately capture full-modality data during human operations. It covers multi-dimensional information such as limb movement trajectories, joint torques, operation timing, and force feedback, providing the most authentic raw materials for foundation model training.
The HALO Human Skill Collection Suit showcased at WRC 2026 introduces a next-generation display version, with upgrades focusing on the wearing experience and overall presentation: the new all-in-one garment integrates full-body sensing units directly into the fabric, ensuring they are perfectly positioned and ready for data collection the moment it is worn. It allows for quick donning and doffing, with each wear providing a stable fit and consistent sensor placement. The garment is washable, and the sensing units are easy to maintain. With no exposed, tangled wiring across the body, the design offers a cleaner, more futuristic aesthetic.
Regarding the most critical sensing and data acquisition capabilities, the next-generation display version retains the mature configuration of the first generation: an 8K panoramic camera, first-person RGBD, head-up display, haptic hand sensors, 27-node inertial motion capture, surface electromyography (sEMG), and swappable batteries, fully preserving its full-modality direct human data collection capabilities.
More importantly, the true value of the HALO Human Skill Collection Suit can only be fully realized within a complete technological closed loop: the front end collects authentic human operational data at low cost via the suit; the middle stage utilizes the Dayan Data Platform for data cleaning, governance, and the construction of high-quality datasets; this data is then fed into the HumanGPT foundation model for skill distillation; and finally, these skills are adapted to robots of various configurations via the RobotGPTs transfer models. This forms a comprehensive pipeline of "human skill collection - model training - robot skill deployment."
2. Industrial Embodied Workstations: Multi-Scenario Deployment Validation
The value of the general-purpose technology foundation must ultimately be demonstrated in specific industrial scenarios. The intelligent welding and intelligent painting workstations showcased by Efort at this WRC are precisely the representative achievements of this foundation transitioning from technical concepts to actual production line deployment.
01. Intelligent Welding Workstation
In industries such as shipbuilding and steel structures, workpieces are characterized by "high-mix, low-volume" production. Every time a workpiece changes, traditional welding robots require re-teaching and re-commissioning. For orders with small batch sizes but a wide variety of products, the commissioning time can even exceed the actual production time.
Efort's intelligent welding workstation does not merely focus on welding precision; instead, it leverages the general-purpose technology foundation to reconstruct the entire commissioning process. The workstation first uses 3D vision to automatically identify the position and angle of the weld seams, while the foundation model generates the optimal welding path online. Operators only need to confirm and make minor adjustments via the graphical interface of the Modou IDE, eliminating the need for point-by-point teaching. This process, orchestrated on the general-purpose technology foundation, reduces the commissioning cycle for a single workpiece from 72 hours to just 4 hours.
02. Intelligent Painting Workstation
The pain points in the painting industry are similar to those in welding. Flexible production with high-mix, low-volume requirements demands rapid line changeovers, whereas traditional painting lines suffer from high barriers to programming and commissioning, along with long cycle times. Efort's intelligent painting workstation is also built upon the general-purpose technology foundation, adopting a "low-code + WYSIWYG (What You See Is What You Get)" approach. Utilizing 3D cameras, point cloud modeling is completed in milliseconds, and the painting foundation model automatically generates trajectories based on the model, followed by simulation fine-tuning via the Modou IDE. Throughout this process, operators do not need to write code; they only need to confirm and adjust on the visual interface, significantly lowering the barrier to use. For small and medium-sized enterprises (SMEs), this means no longer having to "reinvent the wheel" for every new model, drastically reducing flexible production costs.
It is understood that Efort's intelligent painting solution has achieved batch commercial deployment at the Ganzhou Furniture Shared Factory. The furniture industry is a quintessential high-mix, low-volume scenario, featuring workpieces with irregular shapes and diverse materials, which demand high painting uniformity. The ability to successfully deploy and commercialize at scale in such a challenging environment demonstrates that this solution is highly mature and has been validated by the market.
3. Behind the Production Line: An Autonomous and Controllable General-Purpose Technology Foundation
The successful deployment of these production lines is inseparable from the general-purpose technology foundation that has been refined over nearly two decades.
Efort is the first industrial robot company listed on the STAR Market (688165.SH) and a leading enterprise in the national robot industry chain. Its core modules are 100% autonomous and controllable, encompassing core self-developed components such as the E-BOX controller, integrated drive and control cabinets, joint modules, and cycloidal reducers. This represents Efort's "wheel of autonomy," addressing the fundamental question of availability. Meanwhile, according to MIR Research, Efort has also taken the lead in undertaking a National Science and Technology Major Project, receiving the largest single-project funding in this funding cycle.
At this WRC, the Efort & Qizhi booth also showcased the EXR explosion-proof robot.
If intelligent welding and painting address efficiency and cost issues, the EXR explosion-proof robot tackles safety and survival challenges. In high-risk sectors such as defense, civil explosives, and chemicals, manual operations mean direct exposure to explosion risks. Meanwhile, traditional imported explosion-proof robots are not only prohibitively expensive but also suffer from long delivery lead times and slow maintenance response.
In the special operations sector, Efort's EXR series of factory-built explosion-proof robots demonstrates the substitution capabilities of domestic high-end robots, making it a key focus in the special robot category at this exhibition. Unlike the industry's prevalent retrofitting solutions, this series eliminates ignition sources at the design stage. It has obtained the highest dual-level explosion-proof certifications for both gas and dust environments and is fully equipped with battery-free mechanical multi-turn absolute encoders.
In terms of deployment progress, this series of products has achieved batch applications across multiple high-risk industries, including defense, civil explosives, chemicals, pharmaceuticals, and food & beverage, resolving the compliance and reliability challenges associated with automation upgrades in hazardous environments. This explosion-proof robot technology is an outcome of the National Science and Technology Major Project. Efort is also the robotics company that has undertaken the most research tasks in this funding cycle, and the technological maturity of its products now meets the application standards for the domestic substitution of high-end explosion-proof robots.
However, hardware alone is not enough. For industrial embodied intelligence to be successfully deployed, a "wheel of intelligence" is also required to address the question of execution quality—this is precisely the general-purpose technology foundation that Qizhi is building.
In 2024, building on Efort’s industrial expertise and partnering with the National Advanced Manufacturing Industry Investment Fund and Wuhu Science and Technology Innovation Group, the team founded Qizhi (Wuhu) Intelligent Robot Co., Ltd. Focusing on the R&D and industrialization of the general-purpose technology foundation for intelligent robots, the company pursues a path of full-stack in-house development and autonomous controllability, while supporting the ecosystem adaptation of domestic chips and domestic foundation models.
This general-purpose technology foundation forms a closed loop comprising four modules:
HumanGPT is the human skill foundation model, encompassing three major components: the HALO Human Skill Full-Modality Data Collection Suit (direct human data collection), the HumanGPT Human Skill Pre-trained Foundation Model (skill distillation), and the RobotGPTs Human-Robot Skill Transfer Models (skill transfer);
The Dayan Data Platform handles full-pipeline data processing, from collection and governance to training;
The Modou IDE is a low-code integrated development environment, enabling engineers to perform simulation, programming, and deployment via a graphical user interface;
The Openmind OS is an AI-native control system that adapts to hardware at the lower level and supports applications at the upper level, seamlessly integrating industrial-grade real-time control with AI-driven intelligent evolution into a single platform.
The relationship between these four modules can be understood as follows: HumanGPT serves as the skill foundation model, the Dayan Data Platform handles full-pipeline data processing, the Modou IDE acts as the integrated development environment (development phase), and the Openmind OS functions as the control system platform (runtime phase). Together, they address a single core question: how to free robots from reliance on line-by-line pre-programming and instead achieve skill generalization and transfer through a data closed loop. In industry terms, this is akin to building the "Android system for the robotics world," enabling one development effort to support operation across multiple hardware configurations and brands.
If a foundation only serves a company's own robots, no matter how powerful it is, it remains merely an internal tool. What sets Qizhi apart is its positioning from the very beginning as a general-purpose technology foundation open to the entire industry.
According to public information, Qizhi has collaborated with 8 985 Project universities, including Harbin Institute of Technology and Tsinghua University, as well as 6 leading industrial robot manufacturers (accounting for over 70% of domestic sales) to jointly tackle key technological challenges. Currently, 4 industrial robot brands and 3 humanoid robot brands have been integrated into the system. For the general-purpose technology foundation to truly become public infrastructure for the entire industry, technical capabilities and ecosystem partnerships alone are insufficient; institutional guarantees are also required.
According to an exclusive interview published by the 21st Century Business Herald in July 2026, Qizhi will be spun off from Efort to operate independently in the future, with the National Advanced Manufacturing Industry Investment Fund having invested in two consecutive rounds. You Wei stated that this "open-source + independent operation" model aims to eliminate peers' concerns about the risk of being subject to technological blockades, avoid "reinventing the wheel," and ensure that Qizhi's general-purpose technology foundation remains open to the entire industry from its inception.
This is uncommon in the domestic robotics industry—most companies treat core software as a competitive moat, whereas Qizhi has chosen to open its general-purpose technology foundation to the entire industry.
Viewed on an international scale, this combination of an "open foundation plus industrial accumulation" possesses unique characteristics. The AI solutions of international giants are often locked into their proprietary hardware ecosystems, lacking openness; meanwhile, domestic pure-AI teams lack on-site industrial process know-how and hardware supply chains. Efort's nearly two decades of industrial accumulation, combined with Qizhi's general-purpose technology foundation, perfectly sits at the intersection of these two domains. It does not strive to outperform others in isolated algorithms; rather, in the dimension of "industrial-grade stability combined with cross-configuration openness," it approaches international advanced standards while forging its own distinct path.
4. WAIC Review
When we examine the production-line deployments on display at the WRC booths and look back at WAIC one month earlier, we see that the two exhibitions perfectly illustrate two facets of the general-purpose technology foundation. With the theme "Intelligent Partners Co-creating the Future," Qizhi's booth at the Shanghai World Expo Exhibition and Convention Center used three major exhibits to systematically answer a single question: what exactly can the general-purpose technology foundation do?
In the "Nine Stars Aligned" human-robot co-creation demonstration, multiple robots of different brands and configurations collaboratively completed the same task driven by HumanGPT, intuitively showcasing skill generalization and cross-configuration transfer capabilities. The multi-robot collaboration under extreme lunar surface conditions transported the scenario to a simulated lunar environment, where the Dayan Data Platform, Openmind OS, and HumanGPT worked in tandem to validate the adaptability of the general-purpose technology foundation in extreme, unknown environments. Meanwhile, the intelligent painting solution demonstrated the entire low-code development process, from workpiece scanning to path generation, through the combination of 3D cameras, the Dayan Data Platform, the painting foundation model, and the Modou IDE.
The common thread among these three exhibits points to the "capability boundaries" of the foundation—it can adapt to diverse needs across different configurations, scenarios, and industries, proving the technical viability of the general-purpose technology foundation.
By the time of WRC, the focus of the exhibition shifted from "what it can do" to "what it has already achieved": the 4-in-1 embodied intelligence workstation and the SPS sorting system are real production lines already deployed at OEM factories, while the intelligent welding workstation, intelligent painting workstation, and EXR explosion-proof robot all have concrete industry deployment cases. Viewed together, the two exhibitions perfectly illustrate the complete path of embodied intelligence from technical validation to industrial deployment.
Conclusion
WRC 2026 has officially opened. While the robots on display will undoubtedly attract massive attention, what truly deserves focus are the solutions already running stably on actual production lines. From WAIC to WRC, what we witness is not merely the product iteration of a single company, but a crucial turning point for China's robotics industry: when "autonomous and controllable" is no longer just a slogan, and when "intelligence" is no longer confined to booth demonstrations, the deployment of embodied intelligence truly transitions from conceptual discussions to the stage of performance validation.