The most difficult competition for robots has just started.

Has the robot outrun Bolt? Then what?

They can run, jump, somersault, and dance alongside professional dancers. Today's humanoid robots are already very good at showing their most impressive strengths to human audiences.

But what enterprises truly need is never just a wonderful performance, but more importantly, a reliable "employee" that they can trust completely.

There is still a long distance between the two. And over the past weekend, Agibot just traveled from one end of this distance to the other.

On the night of August 22, it appeared at the National Speed Skating Oval in Beijing; the next day, it arrived at Chimelong Hengqin more than 2,000 kilometers away.

One is a competition venue, the other is a real workplace. When viewed together, their meanings are completely different.

Robots Begin to Compete in Two Core Capabilities

Let's first talk about the competition on the field.

The value of the 2nd World Humanoid Robot Games is very straightforward. Speed, strength, balance, perception, control — what a robot can actually achieve can be fully demonstrated once it is brought to the competition field.

The scale of this event is also expanding rapidly. The number of competition events has increased from 26 in the first session to 51, the number of participating teams has expanded from 280 to 666, the number of robots has increased from about 500 to 2056, and the competition schedule has been extended from 3 days to 5 days. Participants come from 16 countries across 6 continents.

These figures show that robots are moving out of laboratories at an increasingly faster pace and entering open competition. Agibot has also become one of the most outstanding teams in this competition. Throughout the Games, Agibot won 18 gold medals and 46 medals in total, ranking first on both the gold medal table and the total medal table.

However, the competitive performance on the field, and a single handling task in a factory or a service delivery in a hotel, test two completely different sets of capabilities.

This exactly forms the two competition dimensions of embodied intelligence today. One is the upper limit of technology, and the other is commercial implementation. This is also the most noteworthy change of this Games: Competitions are starting to step out of standard venues and enter real scenarios.

This edition of the event has added 21 new scenario-based events, covering four major fields: emergency rescue, hotel services, library management, and family life.

Among them, Agibot won 6 gold medals in scenario-based events. While this achievement is certainly important, what is more noteworthy is the scenarios where Agibot appeared: fire stations, libraries, these places do not have a standard environment specially prepared for robots. Robots need to complete tasks in the existing environment, and will also encounter unexpected situations.

The assessment of the competition has also extended from what actions the robot can perform to whether it can complete a whole task successfully.

On the night of August 22, 24 Agibot Expedition A3 robots danced on the same stage with 20 professional dancers, and then played table tennis against world champion Ding Ning.

The next day, it showed up at Chimelong.

From the National Speed Skating Oval to Chimelong, thousands of kilometers apart from north to south, the robots face two completely different challenges. The former scenario requires them to push their capabilities to the extreme, while the latter requires them to apply their capabilities to real work.

Being Proficient in Only One Capability Is Far From Enough

If we map today's embodied intelligence industry to a coordinate system, we will see two very distinct development paths.

One path emphasizes commercialization. Robots that deliver food in hotels, provide services in industrial parks, and perform simple handling tasks can already get a large number of orders. Their advantages are controllable cost, clear demand and large deployment scale, but their technical ceiling is relatively limited.

The other group of players focuses on technical capabilities. More and more robotics companies regard the Games as a technology showcase, constantly refreshing their performance in speed, strength, balance and dexterity. They even customize robots specifically for competitions, pushing a certain capability of a single machine to the extreme.

To be frank, both paths can tell their own success stories. But today, the market has begun to raise the third question:

Is there a robotics company that can not only rank among the industry leaders in technical capabilities, but also actually deliver robots to commercial sites?

What is truly rare is to take the lead in both capabilities at the same time.

On August 19, Unitree was listed on the Sci-Tech Innovation Board. With an issue price of 150.8 yuan, its share price once surged to 1100 yuan at opening, and its market value peaked at 444.9 billion yuan. It fell back to 845 yuan at close that day, with a market value still exceeding 340 billion yuan. On the second trading day, the share price dropped by 18.7%, and the market value fell to 277.9 billion yuan.

In just two trading days, the market has fully demonstrated both sentiment and rationality.

This actually means that the capital market's evaluation of embodied intelligence has begun to become more specific. Having orders and large deployment scale can prove commercial value; being able to run, jump and complete difficult actions can prove technical strength.

But if a company can only have one of the two capabilities, the ceiling of its development story is very clear.

According to data from Counterpoint Research, global shipments of humanoid robots exceeded 22,000 units in the first half of 2026, a year-on-year increase of nearly 300%. The Ministry of Industry and Information Technology of China estimates that the annual output of complete machines in China is expected to exceed 100,000 units.

The number of robots is growing rapidly, but the growth of shipments is only the first step.

After a robot officially enters the working state, the problem will immediately fall to the operation account. Running time, task efficiency, fault recovery, maintenance cost, all these factors will finally be calculated together with the price that customers are willing to pay.

The cost of advanced complete robot models ranges from 20,000 to 30,000 US dollars

Entry-level models cost 8,000 to 10,000 US dollars

The operating cost is about 2 US dollars per hour

These figures have already made it possible for robots to enter commercial scenarios. But after they are actually put into operation, cost, efficiency, reliability, battery life and maintenance will all affect the final economic efficiency at the same time.

Technological breakthroughs can be achieved at press conferences, but commercial value can only be verified through repeated operations in the real world.

These problems rarely appear on the stage of press conferences, but they will determine whether customers will place orders in the end.

Another Competition Is Taking Place at Chimelong

If the capital market has begun to question whether robots can create value, Chimelong provides a site where this can be directly verified.

Cultural tourism scenarios naturally feature high concurrency of tourists, long-time operation, multilingual interaction, dynamic environment and clear service standards. Chimelong has been deeply engaged in the cultural tourism industry for more than 30 years, and has very mature experience in tourist service, park operation and on-site management. When robots enter such a scenario, it means that they need to face a highly mature service system.

If the Games answer what robots can do, Chimelong answers whether robots can turn these capabilities into continuous work outputs.

According to the agreement between the two parties, the cooperation between Agibot and Chimelong will cover four global projects: the world's first immersive robot theme park, the world's pioneering large-scale robot circus show, the world's most distinctive cyber-themed parade, and the world's first robot service hotel. The two sides also plan to jointly build a joint laboratory for embodied intelligence in the cultural tourism industry, to carry out technical research and scenario verification in multiple parks.

There is no standard laboratory environment in these scenarios at all.

Tourists, languages, tasks and the environment are changing at any time. What robots face is no longer a set of pre-programmed actions. They need to understand humans, understand the environment, complete movement, interaction and task execution, and maintain stability during long-term operation.

More importantly, after robots truly enter the parks, real scenarios will continuously generate feedback. Which actions are effective, which interactions fail, which tasks require human intervention, all will become input for the next round of product iteration. This is another kind of test that commercial scenarios pose to technical capabilities. The longer the project runs, the more experience and data it will accumulate.

From Robot Ontology to Delivery, Every Step Is a Threshold

Entering the site is just the beginning. The real challenge is to turn a single successful demonstration into continuous reliable delivery. When robots are in the real environment, what they face is no longer just the performance of the robot ontology itself.

There is a detail that is easily overlooked in the two events of the Games and Chimelong: scale.

At the Games, Agibot's three series of robots and dexterous hands all won gold medals, appearing in the opening ceremony performance and multiple competition events. Then in Zhuhai, Agibot gathered hundreds of robots to put into multiple scenarios of Chimelong.

The two events are more than 2,000 kilometers apart, and were held almost simultaneously.

This is no longer a capability demonstration of a single robot. In essence, it is a stress test for Agibot's delivery capabilities.

Mass production solves the problem of how many robots can be manufactured, while delivery solves the problem of whether so many robots can be put into use at the same time.

Every step from manufacturing, scheduling to on-site support must keep up. The courage to push such a large number of robots to two high-intensity sites at the same time is itself a proof of the company's mass production capacity and delivery system.

This also means that the final competition of embodied intelligence will not only focus on the performance of a single robot. The extreme performance on the competition field and the continuous delivery in real scenarios seem to be two different dimensions, but inside an enterprise, what they ultimately compete for is the system capability. The robot body is responsible for entering the physical environment, the model is responsible for understanding tasks, software and data support continuous operation, and dedicated personnel are required to complete delivery and maintenance when problems occur.

The "Three Intelligences in One" that Agibot has always emphasized, including motion intelligence, interaction intelligence and operation intelligence, has found specific implementation landing points this time. From this Games to Chimelong, Agibot has appeared in two completely different actual combat sites at the same time.

The Games push the technical capability to a higher upper limit, while Chimelong puts these capabilities into the commercial environment.

Being able to win awards on the competition field and achieve continuous delivery in real scenarios at the same time is the real threshold for today's robotics companies.

Agibot is doing both of these two things right now.

The Real Competition Has Just Begun

For Agibot today, the really interesting part of these two events is that they put a problem that used to be easily discussed separately on the same company: robots must not only have outstanding technologies, but also have viable and scalable businesses.

Deng Taihua, Founder, Chairman and CEO of Agibot, once judged that 2026 is the key transition year for embodied intelligence to move from the development stage to the deployment stage. He described this path with the "XYZ Curve".

The X curve represents the development and trial stage when robots can move like humans. The Y curve represents the deployment and growth stage when robots can work like humans. The Z curve represents the deployment and popularization stage, where driven by large-scale real scenario data and combined with continuous algorithm innovation, embodied intelligence will achieve a qualitative change from quantitative accumulation and move towards wider application.

From the perspective of the whole industry, this XYZ curve exactly corresponds to the process of robots evolving from "being able to do things" to "being usable", and then to large-scale popularization.

Mass production solves the problem that robots can be manufactured, while large-scale deployment faces another challenge: when more and more robots enter different scenarios, whether their performance and stability can still be maintained.

Whether a robot can be continuously used by customers depends on hardware only as the foundation. How much work it can finally complete also depends on the model's understanding of the environment and tasks, as well as the stability of the system during long-term operation.

For enterprises, purchasing robots inevitably requires calculating operation accounts, such as how many tasks can be completed in a month, how much cheaper it is than human labor, what the failure rate is, and how much the maintenance cost is. These problems rarely appear in technology press conferences, but they will determine how far robots can go in the end.

The two scenarios of Agibot this time just bring together the changes that are taking place in the industry. It also accelerates the competition of embodied intelligence from the product competition between robotics companies to the competition of systematic engineering capabilities.

In the past, embodied intelligence players could get public attention by relying on their technologies, or obtain orders by relying on commercialization. Going forward, the market will pay more and more attention to one thing:

Can two satisfactory report cards be obtained at the same time?

This transition is happening faster than we expected, and Agibot is just standing at the node where this transition takes place.

Words Beyond the Layout:

A robot outrunning Usain Bolt can only prove that it can run fast.

What truly changes the industry is to not only push the technology to a certain height, but also bring these capabilities into the production environment, obtain orders and customers, and truly develop a continuously replicable business model.

When a company begins to enter this positive cycle, embodied intelligence can truly enter its next stage of development.

This article is from the WeChat Official Account "Beyond the Layout", written by Huahua, authorized for release by 36Kr.