When browsing WRC, the most attractive feature of ‘Fixed Focus One’ is still the humanoid robot. In the front of the exhibition hall, fighting, dancing, and backflips are staged in a rotating manner, with three layers inside and three layers outside the audience.
The robot fighting displayed in WRC C Pavilion
Going further inside, the robots on the exhibition stand are doing inconspicuous but more like “working people” tasks: the conveyor belt keeps spinning, the robotic arm lands time and time again, some sorting packages, some moving material boxes, and some robots bowing their heads, trying to fold clothes and put the miscellaneous items on the table back in place.
This year’s highlights are not all in the busiest booths. After brewing tea, can I hand over the cup to someone? Can we still find the grabbing position after changing a batch of materials? The box is crooked, will it adjust itself? If someone takes the cup halfway through the task, can we continue working? These things are commonplace for humans, but increasingly difficult for robots.
The whole machine is no longer the only protagonist. Agile hands, tactile sensors, joints, motors, reducers, visual modules, data acquisition devices, and models have all come to the forefront.
On the forum, entrepreneurs and researchers discussed the brain, ontology, world models, developers, open source, and data supply chains, as well as family caregiving, security governance, and responsibility boundaries. The topics are different, but they all end up in practical use: can the robot continue to work steadily after leaving the booth.
The purchaser’s concern is more direct: whether it can be integrated into their own business process, who to contact when there is a malfunction, and whether this account can be calculated in the end.
After visiting the conference, looking back, there are three parallel lines in WRC this year. One way to ‘nourish the body’: components, supply chain, and manufacturing capabilities to make robots more stable, lighter, and cheaper; A ‘brain boosting’ approach: models and data that enable it to adapt to changes, complete a series of actions, and continue to work even if it makes mistakes; Another approach that is closer to the business itself is that robots are beginning to enter factories, warehouses, and stores, and procurement, delivery, maintenance, and long-term costs are also being put on the table.
This is a conference approaching the market from the runway. Robots have indeed started looking for jobs. As for whether it can stay, we will only know after it has been running in factories, warehouses, and homes for a while.
01. It’s not the whole machine that caught the spotlight this year, it’s the people who sell shovels
The whole machine is still the most eye-catching part in the exhibition hall. But this year, issues such as lifespan, compatibility, mass production, and cost, which were not easily seen in the past, have also been brought to the forefront.
Taking dexterous hands as an example, this year’s comparison is no longer just about how many degrees of freedom they have or whether their movements resemble those of humans, but also about load, weight, tactile accuracy, lifespan, and cost. For customers, completing one grab is just the first level, and more importantly, whether they can continuously handle different materials and adjust when encountering slippage, misalignment, and soft objects.
For example, when handing a cup to someone, if the force is too strong, it may be crushed, and if the force is too weak, it may fall off; To install the parts into the fixture, it is not only necessary to find the correct position, but also to determine whether they are stuck and whether they have been placed firmly.
The content of joint, actuator, tactile sensor and other component companies has also changed accordingly. In the past, they mainly introduced parameters and performance, but now they are also beginning to explain what problems these components can solve when installed on robots.
Moving forward, some companies are beginning to showcase their manufacturing capabilities. Lingxin Clever Hand showcased a production line for assembling dexterous hands. Compared to simply displaying grasping actions, this production line brings the problem to the production end: whether dexterous hands can produce stably and in bulk. For whole machine manufacturers, moving from prototypes to procurement not only depends on whether the product can be used, but also on yield, cycle time, cost, and whether the supplier has the ability to deliver continuously.
In addition to hardware, data collection devices are starting to be separately mentioned. Remote control, motion capture, demonstration, and simulation are all solving the same fundamental problem: what kind of data robots need, and how can enterprises continue to obtain this data. Human movements, object positions, and materials can all be converted into training data through these methods.
02. The scene shifts from performance to work, and the factory is closest to large-scale orders
This year, WRC has seen an increasing number of robots performing tasks. Sorting, handling, palletizing, loading and unloading, testing, assembly, logistics, retail, catering, and inspection can all find corresponding demonstrations.
From the projects presented by WRC and feedback from enterprises, it can be seen that the commercialization stages in different scenarios are far apart: factories are closest to large-scale orders; Logistics follows closely behind; The willingness to pay in extreme environments is clear, but more inclined towards vertical projects; Catering retail is easy to implement, but it needs to be proven to be more cost-effective than hiring people; Family is the farthest.
The tasks in the factory are also graded. Robots often do fixed position, repetitive actions, and clear processes first. Handling, palletizing, sorting, loading and unloading, and scanning are relatively easy to calculate input-output.
Testing, assembly, and flexible loading and unloading are more complex, and robots need to identify material differences, adjust actions, and restart after failed grasping.
Even more difficult is to connect multiple processes and let the robot complete the entire process from material picking, processing to placement on its own.
Logistics is located between factories and open environments. There is already a clear workflow for warehousing and logistics, but the goods, routes, and orders in the warehouse will continue to change.
At the exhibition, a type of enterprise incorporated robots into the actual operation process, emphasizing the stable completion of tasks; Another type of enterprise showcases full body teleoperation, mapping human turning, bending, and arm movements onto robots for data collection or remote operation. The former approach is already attempting to directly hand over tasks to robots, while teleoperation retains the method of manual intervention for complex tasks and can also collect training data at the same time.
The demand for extreme environments is more clear.
Underwater, mining sites, underground spaces, chemical pipe galleries, substations, and fire protection sites all have some dangerous, harsh, or unsuitable work for people to enter for a long time. Robots performing inspections, detection, transportation, and risk mitigation here, taking on risks on behalf of humans, are inherently valuable.
The underwater inspection robot displayed in WRC C Pavilion
This type of scenario is also more specialized. The terrain and tasks of underwater, mining, and firefighting sites vary greatly, resulting in different forms of robots such as wheeled, tracked, wheeled, quadruped, unmanned aerial vehicles, and underwater robots on site. Protection, communication, battery life, remote control, and fault handling often need to be re adapted according to the scene.
The fire extinguishing machine dog displayed in WRC C Hall
This type of project may generate orders earlier, but it is not easy to scale up like factories and logistics.
Another characteristic of emergency rescue is that it is often not a single robot working alone. The WRC Fusion Robot Challenge simulates accident rescue, where aircraft and unmanned vehicles must complete reconnaissance, information sharing, material transportation, and return. At the same time, the conference also released the application requirements of embodied intelligence in emergency scenarios. From competitions to actual needs, emergency rescue has become a type of robot application repeatedly mentioned at this year’s WRC.
WRC Robotics Competition
Catering, retail, and guided tours are the most popular types of scenes on site.
This year, WRC also set up a robot consumption street for the first time. Scenes such as coffee, ice cream, cocktails, and gift retail have been centralized together, and robots are starting to pick up cups, operate equipment, and make drinks according to orders. These actions are closer to real service than dancing and backflips.
The catering robot displayed in WRC C Hall
But before purchasing, merchants will definitely calculate efficiency and cost, how many orders the robot can take up in a day, how much space the equipment occupies, and whether this solution is cost-effective compared to hiring directly.
Families are the most imaginative and difficult to standardize.
On the booth, you can already see sequential tasks such as collecting debris, sorting garbage, folding clothes, watering flowers, and arranging flowers. But when you really enter the house, the cup will change position, the shape of the clothes will be different, and the robot will face more changes than the exhibition booth.
Compared to regular family services, the needs for elderly care and companionship are more clearly defined. The issue of universal access to people’s livelihoods in WRC will be discussed separately for elderly care, rehabilitation, and assistive robots. Robots can be used for action assistance, rehabilitation training, reminders, and environmental monitoring. The demand side also includes family members, nursing homes, and caregivers.
The elderly care scene places more emphasis on reliability. Whether robots can replace some nursing work and respond promptly in case of abnormalities will directly affect whether institutions and families are willing to use them for a long time.
03. They are all replenishing their brains, but the route has not converged yet
After robots start taking on more complex tasks, the “brain” also has more problems to solve.
At this year’s WRC, VLA、 World model, end-to-end Agent、 The technical keywords such as neuromorphic architecture are very frequent, and the problems they solve are not exactly the same. Some are solving the problem of how robots generate actions after seeing tasks, some hope that it will first judge how the surrounding environment will change, and some are concerned about whether the same set of abilities can still be used after changing their “body” or “workstation”.
Founder and CTO of Galaxy General Motors, Wang He, mentioned that robots cannot only complete trained actions, but also need to apply their learned abilities to new tasks. Galactic Universal attempts to integrate VLA and world models into the same system using the World Action Model (WAM). Robots not only need to decide how to move next, but also estimate what changes will occur in their surroundings if this action continues.
Besides how to make decisions, another frequently discussed issue is whether a set of abilities can be replaced with a ‘body’ and continue to be used.
Ant Lingbo uses the same universal brain in scenarios such as pharmacy sorting, industrial loading and unloading, and warehouse sorting in WRC. The on-site staff of the enterprise introduced that this set of brain has been adapted to multiple robot brands and different forms of ontology during the training stage.
If every time a robot or workstation is replaced, data needs to be collected, trained, and debugged again, it will be difficult to quickly replicate a project.
When the robot arrives at the customer site, it will constantly encounter situations that have not been seen during training.
Xiao Zhongyang, founder and CEO of Corona Open, mentioned that it is difficult to rely solely on laboratory data to cover materials, friction, sensor noise, and various long tail situations in real environments. He also sees product delivery as a part of learning: when robots arrive on site, where they make mistakes and need to be manually taken over, these data can be retrieved to train the model.
The DaXiao robot also demonstrated a similar path at WRC. On one end are world models and environmental data collection tools, and on the other end are instant retail, hotel services, and outdoor open scenarios. Robots work in these scenarios and generate data that is then fed back into the model for subsequent iterations.
There is also a relatively independent technology branch on WRC: brain computer interface.
The BCI Brain Controlled Robot Competition showcased projects such as brain controlled wheelchairs, brain controlled robotic arms, and brain controlled humanoid robots. It solves another type of problem: how to transmit human intentions more directly to machines, which is also a human-machine collaboration method in rehabilitation assistance and other scenarios.
On site WRC BCI Brain Controlled Robot Competition
For more robots that hope to work autonomously, the current problem is still very specific: once items are changed or the environment changes, their performance is easily affected.
Wang Xingxing, founder and chairman of Yushu Technology, mentioned in his WRC speech that in a fixed scenario, with sufficient data collection and training, the success rate of some tasks can already be achieved very high; But if the items being operated are changed or the environment changes slightly, the success rate may significantly decrease. Sometimes, the problem is only a few centimeters away. The robot appears to have almost grasped something, but if the last physical error is not corrected, the entire task will fail.
Wang Xingxing mentioned that in about 80% of unfamiliar scenarios, if robots can complete about 80% of tasks through language instructions, it can be seen as an important critical point for general ability. He estimated that at this point, it would take 2-3 years if it was fast, and 5-10 years if it was slow.
At least at this year’s WRC, it is not clear which solution has solved these problems well enough.
04. The buyer enters the market, and the channel also arrives
This year’s WRC has attracted a group of major buyers.
The conference has set up a procurement day for the first time, with demand sides from energy, manufacturing, logistics, inspection, and emergency fields entering the venue in a centralized manner. Some have production lines, warehouses, power stations, and mines in their hands, while others have access to retail, logistics, and other channels, all looking for scenarios where robots can enter.
At the opening ceremony, Xin Guobin, Deputy Minister of the Ministry of Industry and Information Technology, proposed to “take intelligence as the main line and practicality as the driving force” to coordinate technological breakthroughs, industrial cultivation, scene expansion, and security governance.
Central enterprises are an important group of demanders. The Central Enterprise Robot Innovation Alliance has also released ten innovative achievements and ten high-value application scenarios, covering energy, manufacturing, logistics, emergency and other fields.
But there are still many links in between from scene integration to procurement. After requirements communication, on-site testing, small-scale trials, and solution adjustments are implemented, deployment can be expanded.
The M7 showcased on site by Xingdong Era is a case study of entering the stage of normalized operation. This robot does not have complete lower limbs, and adopts a column structure below the waist to collect tasks at fixed workstations, mainly for logistics operations such as sorting and scanning. The on-site staff introduced that the relevant plan has already been operated in more than ten logistics centers of SF Express and China Post nationwide, and it still needs to be seen whether it can be further expanded in the future.
If we have to create a new plan every time we visit a client’s site, the business will still be closer to a project-based approach. The same set of products can be used in similar scenarios, and customers are willing to continue buying after the first batch of trials, which gives the opportunity for scale to increase.
In addition to customers who purchase directly, JD brings channel entry.
It connects robot brands and also has retail, logistics, and consumer scenarios. For robotics companies, in the past, it was more about finding customers on a project by project basis, but now channels are also emerging to help showcase, compare, and sell products.
At the channel end, how to set the price, whether the supply can keep up, whether the installation and use are troublesome, and whether the same robot can perform consistently when delivered to different customers will all affect sales.
05. After selling, there are still operations, responsibilities, and costs involved
After the robots are sold, there is still a batch of problems that will not be exposed until they are actually put into use.
This year, WRC also provided a window to observe the position of China’s embodied intelligence industry.
The number of industrial robots added in China in a year is much higher than that in the United States; In the field of humanoid robots, Chinese manufacturers will account for over 97% of global shipments in the first half of 2026. A complete supply chain of components, a large number of manufacturing scenarios, as well as cost reduction and mass production speed, enable Chinese companies to make robots faster and deliver them to the site.
相关文章
