At Jianghuai Laboratory, a Qijiang 2M humanoid robot stands in position, identifies the opening of a pot, adjusts its posture, raises its arm, and throws an arrow. The movement is a reenactment of a competition routine, but it also illustrates a broader shift in robotics: the humanoid robot is no longer only a laboratory demonstration or a stage performance. It is becoming a system that must perceive a changing environment, make decisions in real time, and execute physical actions with increasing reliability.
The humanoid robot program at Jianghuai Laboratory earned a gold medal in the pot-throwing legged project and a silver medal in the smart charging scenario at the Second World Humanoid Robot Games. The two robot athletes have now returned to the laboratory, where research teams are continuing work on the technologies that made their performances possible. On September 2, a visit to Jianghuai Laboratory provided a closer look at the two award-winning teams and at the laboratory’s next steps in embodied intelligence.
The laboratory is now working with Heli Group, Juyi Technology, and other partners on joint technical projects. These projects are intended to accelerate the application of embodied intelligence in smart logistics, intelligent assembly for new energy vehicles, and other real-world scenarios. For the humanoid robot, the competition is therefore not an endpoint. It is a testbed that exposes both progress and remaining gaps, from perception under difficult lighting to millimeter-level manipulation in charging operations.
The results also show why the humanoid robot has become a focus for industrial research. A humanoid robot must combine mobility, sensing, planning, and manipulation in one platform. It must operate in spaces designed for people, use tools and interfaces built for people, and adapt to conditions that are not fully controlled. The laboratory’s two competition programs reflect different parts of this challenge: one emphasizes dynamic throwing and whole-body coordination, while the other emphasizes precision, infrastructure compatibility, and repeated operation in practical settings.
1. A Humanoid Robot Repeats the Winning Throw
Inside the laboratory, the Qijiang 2M humanoid robot once again picks up an arrow and reproduces the sequence that won the pot-throwing legged project. The competition required more than a single gesture. The humanoid robot had to stand still, identify the pot opening, adjust its body posture, swing its arm, and release the arrow. Each step had to be completed within a limited competition window.
The team that developed the entry was the Qijiang Hundred-Step-Piercing-Willow Team. Its captain, Zhou Zhengtong, explained that the competition lasted only three minutes. Within that time, the humanoid robot had to complete the entire action sequence and ultimately place an arrow into the pot. For a humanoid robot, a complete pot-throwing attempt is not simple. It involves taking an arrow, walking, positioning the body, releasing the arrow, changing hands, and throwing. The real difficulty, however, is whether the humanoid robot can make real-time judgments based on its own position relative to the pot opening.
- Perception: The humanoid robot must use its camera to sense the position of the pot opening relative to its own body.
- Posture adjustment: The humanoid robot must adjust its entire body posture rather than only moving a single arm.
- Real-time decision-making: The humanoid robot must decide how to throw based on the current relationship between its body and the target.
- Whole-body motion planning: The humanoid robot must coordinate its torso, arm, and release timing so that the sequence remains连贯 and effective.
- Brain-cerebellum coordination: The humanoid robot must combine higher-level control with lower-level motion execution.
The competition environment added further difficulty. Large spotlights and colored lighting on site could interfere with the humanoid robot’s visual perception. To help the humanoid robot “see accurately,” the team collected a large amount of data in different environments, including the experimental site, the competition venue in Beijing, and outdoor locations. The team also deliberately trained the humanoid robot on floors of different colors, including black and blue surfaces, so that visual perception would be more robust when conditions changed.
As Zhou Zhengtong put it, only accurate sight can lead to accurate throwing. Pot throwing is therefore not a isolated action. It is the result of coordinated control across perception, decision-making, and execution. The humanoid robot must complete brain-cerebellum coordinated control and whole-body motion planning, and then execute a series of actions in a continuous and stable manner. The gold medal reflects the integration of these capabilities in a humanoid robot platform.
The significance of this humanoid robot demonstration lies in the combination of dynamic movement and target awareness. A humanoid robot that can perceive a target, adjust its posture, and execute a throw in a short window is demonstrating a control architecture that can be adapted to other tasks. The same architecture may support a humanoid robot in logistics handling, assembly assistance, or service operations, where the robot must respond to objects and positions that are not fixed in advance.
2. The Humanoid Robot Moves From Centimeters to Millimeters
On the other side of the laboratory, the Qingyu Team’s intelligent charging robot demonstrated a different set of capabilities. The intelligent charging robot participated in the smart charging scenario and earned a silver medal. Jianghuai Laboratory Executive Director Li Cheng noted that the intelligent charging robot had already been piloted in real scenarios involving Hefei buses and Tianjin buses in the previous year. This background is important because it shows that the humanoid robot ecosystem is not limited to one competition task; related robotic systems are already being tested in operational environments.
The charging sequence appears straightforward when performed by a person. The robot identifies a vehicle, walks autonomously, takes the charging gun, opens the charging port cover, and inserts the gun. For a robot, however, each step requires perception, localization, manipulation, and safety checks. The most demanding part is the precision gap between navigation and insertion. A robot chassis usually navigates with centimeter-level accuracy, but a State-standard charging port requires the insertion error to remain within one millimeter.
- Vehicle recognition: The intelligent charging robot must identify the target vehicle and confirm the charging interface.
- Autonomous walking: The robot must move to the correct position in a bus depot or parking area.
- Gun handling: The robot must take the charging gun and manage its orientation and weight.
- Cover opening: The robot must open the charging port cover without damaging the vehicle.
- Precise insertion: The robot must align the gun with the port and insert it within a one-millimeter tolerance.
- Compensation control: The robot must use visual sensors and force-control algorithms to compensate for positioning errors.
Qingyu Team Captain Ping Yuqing explained that it is difficult to complete such fine operation by relying only on the mechanical arm. The team therefore uses visual sensors and force-control algorithms to continuously compensate for the robot’s position and action. In effect, the system must correct the error created by centimeter-level navigation until the final insertion reaches millimeter-level accuracy. This is a clear example of how a humanoid robot or a related embodied intelligence system can bridge the gap between broad mobility and fine manipulation.
The charging robot also addresses an infrastructure problem. Existing bus depots and large parking lots may already have many charging piles. If all of them were removed and rebuilt, the cost would be high. The team’s approach is to let the robot adapt to different types of charging guns, including fast-charging and slow-charging models, by changing fixtures and related components. In this way, the robot can make greater use of existing charging facilities rather than requiring a complete replacement of infrastructure.
According to actual requirements the team has encountered, one robot may in the future serve multiple vehicles. The team is also exploring whether the robot can take on inspection tasks during idle periods at night. These possibilities show how a humanoid robot or an intelligent charging robot can become part of a larger operational system, rather than a single-function device. The humanoid robot’s value in such scenarios depends not only on its body, but also on its ability to integrate perception, manipulation, scheduling, and safety management.
| Program | Robot or Team | Event | Result | Technical Focus | Application Relevance |
|---|---|---|---|---|---|
| Pot-throwing legged project | Qijiang 2M humanoid robot; Qijiang Hundred-Step-Piercing-Willow Team | Second World Humanoid Robot Games | Gold medal | Camera-based pot-opening perception, real-time position judgment, whole-body posture adjustment, brain-cerebellum coordination, whole-body motion planning | Dynamic manipulation, perception-decision-execution integration, humanoid robot control under variable lighting |
| Smart charging scenario | Intelligent charging robot; Qingyu Team | Second World Humanoid Robot Games | Silver medal | Vehicle recognition, autonomous walking, gun handling, cover opening, insertion, visual sensing, force control, centimeter-to-millimeter compensation | New energy vehicle charging, compatibility with existing charging piles, multi-vehicle service, nighttime inspection |
The table above shows two complementary directions for embodied intelligence. The humanoid robot in the pot-throwing project demonstrates dynamic action and target awareness. The intelligent charging robot demonstrates precision and infrastructure compatibility. Together, they indicate that the humanoid robot is not a single application. It is a general platform concept that can be adapted through perception modules, control algorithms, end effectors, and task planning.
3. Competition Results Show the Humanoid Robot’s Integration Challenge
The one-gold and one-silver result matched the province’s previous medal record at the games. Li Cheng views the achievement as evidence of the technical foundation that Jianghuai Laboratory has built in embodied intelligence over time. The laboratory is not only developing complete robot systems. It is also working on core components and control technologies that support the humanoid robot and related embodied intelligence platforms.
- Perception modules: The laboratory develops sensing modules that allow a humanoid robot to understand its environment.
- Joint modules: The laboratory develops joint modules that support motion and manipulation.
- Brain-cerebellum control systems: The laboratory works on control architectures that combine high-level decision-making with low-level execution.
- Robot models: The laboratory continues to develop models that support robot behavior and learning.
- Algorithms and control: The laboratory conducts research on algorithms and control technologies for embodied intelligence.
These elements are closely connected. A humanoid robot cannot perform a throwing task without perception, planning, and execution working together. It cannot achieve millimeter-level insertion without sensors, force control, and error compensation. It cannot move from a competition venue to a real workplace without robust models, adaptable algorithms, and reliable control systems. The humanoid robot is therefore both a product concept and an integration challenge.
The competition results also highlight the importance of environmental robustness. In the pot-throwing project, lighting and floor color affected visual perception. The team responded by collecting data across different environments and training the humanoid robot on multiple surfaces. In the charging scenario, the challenge came from the precision gap between navigation and insertion. The team responded with visual sensors and force-control algorithms. In both cases, the humanoid robot had to compensate for uncertainty rather than rely on a perfectly controlled environment.
| Technology Area | Elements Mentioned in the Humanoid Robot Program | Why It Matters for Real-World Deployment |
|---|---|---|
| Perception | Camera-based target sensing, visual sensors, data collection under different lighting and floor colors | A humanoid robot must recognize objects, openings, vehicles, and interfaces under variable conditions |
| Decision and planning | Real-time position judgment, whole-body motion planning, long-horizon task planning | A humanoid robot must decide how to move and act when the target or environment changes |
| Execution and control | Brain-cerebellum coordination, force control, error compensation, precise insertion | A humanoid robot must convert plans into stable physical actions with limited error |
| Core components | Perception modules, joint modules, brain-cerebellum control systems | Reliable components determine whether a humanoid robot can operate repeatedly in real settings |
| Application scenarios | Smart logistics, new energy vehicle smart assembly, medical robotics, charging operations | A humanoid robot must adapt to different tasks, tools, and safety requirements |
For the humanoid robot, the competition is a controlled test of integration. The real world is less controlled. Lighting changes, floors vary, objects move, and human activities introduce uncertainty. The laboratory’s work therefore extends beyond a single demonstration. It aims to build a humanoid robot capability base that can be transferred across tasks and industries.
4. Humanoid Robot Research Expands Into Logistics and Smart Assembly
Jianghuai Laboratory is pursuing joint technical projects in several directions. It is working with Heli Group on intelligent logistics robot development. It is working with Juyi Technology on robotics for intelligent assembly in new energy vehicles. It has also established an applied robotics industry college with Anhui Medical University to promote demonstration applications of embodied intelligence in the medical field.
These collaborations show how the humanoid robot and embodied intelligence agenda can connect to existing industrial strengths. Smart logistics requires robots to move, identify, grasp, and transport objects in dynamic environments. Intelligent assembly for new energy vehicles requires robots to handle components, follow precise sequences, and work safely alongside people and production equipment. Medical robotics requires careful interaction, reliability, and task-specific safety. In each case, the humanoid robot concept can contribute perception, planning, and manipulation capabilities, while the application domain imposes its own requirements.
| Partner | Focus | Application Domain | Relevance to Humanoid Robot Development |
|---|---|---|---|
| Heli Group | Intelligent logistics robot development | Smart logistics | Movement, identification, grasping, and transport in dynamic industrial environments |
| Juyi Technology | Robotics for new energy vehicle intelligent assembly | New energy vehicle smart assembly | Precision manipulation, sequence control, and safe operation in production settings |
| Anhui Medical University | Applied robotics industry college and demonstration applications | Medical robotics | Reliability, careful interaction, and task-specific safety for embodied intelligence |
The humanoid robot is not isolated from these sectors. A humanoid robot platform can benefit from the perception modules, joint modules, and control systems developed for logistics, assembly, and medical applications. Conversely, those applications can provide real-world data and operational requirements that make the humanoid robot more robust. The laboratory’s collaboration model therefore supports a two-way transfer of technology between the humanoid robot program and industry-specific robot systems.
Li Cheng has described embodied intelligence as a future industry. That description is important because it sets expectations. The humanoid robot is not yet a universal worker, and the laboratory does not present the competition results as proof of complete commercial readiness. Instead, the results show that specific capabilities, such as dynamic throwing, target perception, precision insertion, and error compensation, can be demonstrated in a humanoid robot or related embodied intelligence system. The next step is to test those capabilities in more scenarios and under more demanding conditions.
5. From Future Industry to Real-World Testing
The competition has ended, but technical research and industrial application continue. Li Cheng said that the laboratory will continue to work on core technologies, including embodied perception, world models, long-horizon task planning, and precise compliant operation. These areas are directly relevant to the humanoid robot. A humanoid robot that can perceive its environment, model the world, plan over extended tasks, and perform precise but compliant operations will be better prepared for real workplaces.
- Embodied perception: The humanoid robot must interpret sensor data in the context of its body and task.
- World models: The humanoid robot must maintain an internal representation of objects, spaces, and changes in the environment.
- Long-horizon task planning: The humanoid robot must plan sequences that extend beyond a single action.
- Precise compliant operation: The humanoid robot must manipulate objects with accuracy while adapting to contact and force.
- Real-scenario testing: The humanoid robot must be evaluated in actual industrial and service environments.
- Joint innovation: The humanoid robot must be developed with backbone enterprises that understand application requirements.
The laboratory also plans to push existing technologies into real scenarios and strengthen joint innovation with key enterprises in the province. The goal is to let robots be tested and to play a role in more industrial settings. For the humanoid robot, this means moving from a competition routine to repeatable tasks. It means moving from a single successful throw to reliable operation over many attempts. It means moving from a charging demonstration to a system that can work with existing infrastructure and serve multiple vehicles.
The humanoid robot’s development path therefore has two parallel tracks. One track improves core capabilities: perception, planning, control, and manipulation. The other track validates those capabilities in real environments: bus depots, parking areas, logistics operations, assembly lines, and medical demonstration settings. The two tracks must inform each other. A humanoid robot that performs well only in a laboratory will not be sufficient. A humanoid robot that is deployed without robust perception and control will not be safe or useful. The laboratory’s approach, as described by its teams, is to continue research while expanding real-world testing.
6. What the Humanoid Robot Team Demonstrated
The pot-throwing humanoid robot demonstrated a complete perception-decision-execution loop. The humanoid robot had to see the target, decide how to position its body, and execute a throw within a limited time. The team’s preparation across different environments and floor colors shows that visual robustness is a practical requirement. The humanoid robot could not rely on a single lighting condition or a single background. It had to generalize from data collected in the laboratory, the competition venue, and outdoor settings.
The charging robot demonstrated a different kind of capability. It had to identify a vehicle, move autonomously, handle a charging gun, open a cover, and insert the gun with millimeter-level accuracy. The gap between centimeter-level navigation and one-millimeter insertion is a clear example of why embodied intelligence requires more than mobility. A humanoid robot or intelligent charging robot must use sensing and force control to compensate for positioning error. It must also consider the existing infrastructure, because replacing all charging piles would be costly. The ability to change fixtures and adapt to fast and slow charging guns is therefore as important as the insertion action itself.
- Robust perception: The humanoid robot must see targets and interfaces under variable lighting and backgrounds.
- Real-time decision-making: The humanoid robot must adjust its plan as its position relative to the target changes.
- Whole-body coordination: The humanoid robot must coordinate posture, arm movement, and release or insertion timing.
- Precision compensation: The humanoid robot must correct navigation error to meet millimeter-level requirements.
- Infrastructure compatibility: The humanoid robot and related systems must work with existing facilities where possible.
- Multi-task potential: The humanoid robot platform may support charging, inspection, logistics, assembly, and medical applications.
These capabilities are not independent. A humanoid robot that can perceive accurately but cannot control its body will fail. A humanoid robot that can plan a task but cannot compensate for contact forces will fail in precise insertion. A humanoid robot that can perform one demonstration but cannot adapt to a new environment will remain limited. The laboratory’s competition results are therefore best understood as evidence of integration: perception, decision-making, execution, and control working together in a humanoid robot or embodied intelligence system.
7. The Road Ahead for the Humanoid Robot
The humanoid robot team’s return to the laboratory marks the beginning of a new phase. The gold medal in the pot-throwing legged project and the silver medal in the smart charging scenario have brought attention to the laboratory’s work, but the laboratory’s focus is now on the next set of technical and application challenges. The humanoid robot must be tested in more real scenarios, and the laboratory must continue to develop the core technologies that support reliable operation.
Jianghuai Laboratory’s work covers perception modules, joint modules, brain-cerebellum control systems, robot models, algorithms, and control technologies. It also includes collaboration with Heli Group, Juyi Technology, and Anhui Medical University. This combination of core research and application partnerships is important for the humanoid robot. Core research provides the capabilities; application partnerships provide the requirements, data, and validation environments. The humanoid robot can then be improved through repeated testing rather than remaining a fixed demonstration.
The laboratory’s stated next steps include embodied perception, world models, long-horizon task planning, and precise compliant operation. These are demanding areas. Embodied perception requires a humanoid robot to interpret sensor data in relation to its body and task. World models require a humanoid robot to represent objects, spaces, and changes over time. Long-horizon task planning requires a humanoid robot to organize multiple actions toward a goal. Precise compliant operation requires a humanoid robot to control contact and force while maintaining accuracy. Each area contributes to the humanoid robot’s ability to work in real environments.
The laboratory also intends to strengthen joint innovation with key enterprises and push existing technologies into real scenarios. For the humanoid robot, this means continued testing in smart logistics, new energy vehicle smart assembly, medical robotics, and charging operations. It means learning from the charging robot’s experience with existing infrastructure. It means applying the pot-throwing humanoid robot’s perception and whole-body control lessons to other dynamic tasks. It means turning competition performance into repeatable, safe, and useful operation.
The humanoid robot remains a future industry, but the direction is clear. The humanoid robot must see accurately, decide quickly, move precisely, and adapt safely. The humanoid robot must work with people, tools, and infrastructure that already exist. The humanoid robot must prove itself not only in a three-minute competition window but also in the longer cycles of industrial operation. The return of the two robot athletes to Jianghuai Laboratory is therefore more than a homecoming. It is a transition from winning routines to the harder work of real-world validation, where the humanoid robot’s value will be measured by its ability to perform useful tasks reliably, repeatedly, and safely.
The laboratory’s teams have shown that a humanoid robot can combine perception, decision-making, and execution in a dynamic throwing task. They have shown that an intelligent charging robot can bridge centimeter-level navigation and millimeter-level insertion. They have shown that embodied intelligence research can connect to logistics, new energy vehicle assembly, and medical applications. The next stage will determine how quickly those demonstrations become dependable systems. For the humanoid robot, the arena has provided a visible test. The factory, the bus depot, the parking area, and the hospital will provide the tests that matter next.