World Robot Conference: The Clumsy Present and the Dashing Future

I've just returned from this year's World Robot Conference.

“There are so many people!” This was the first thing almost everyone said when they met at this year's World Robot Conference.

The World Robot Conference has always attracted large crowds, but this year, the feeling of being packed shoulder to shoulder was especially intense.

However, what was even more astonishing than the large number of people was the omnipresence of robots. Last year, it took three or five “bodyguards” to lead a robot around the venue on a leash. This year, especially on the first day of the exhibition, you might turn around and brush past a two - legged or even four - legged new species.

This year's robots are more eager than ever to interact with the world. They can fold clothes, pull up blankets, box with the audience, play football, and interact with the audience by passing objects... At the same time, this eagerness has also led to many embarrassing moments. You can almost find the blunders of all top companies on social media, whether it's a robot falling to the ground and twitching in a comical way or confidently trying to grab something in front of it but only grasping at air.

Compared with the carefully choreographed robot dances and the slow process of moving an object from one place to another in fixed interactions at previous robot exhibitions, you can clearly feel a new vitality emerging.

Compared with other industries, robot companies seem less concerned about showing their blunders. This kind of unscripted demonstration has a beauty that comes with economic growth: a confidence that says “I know I'll make mistakes, but I'm still the future.” It's chaotic but charming, a unique beauty of a high - growth industry.

This is also the deepest impression I got from the World Robot Conference: as long as there is a future and the robots are growing fast enough, people are willing to wait for a temporarily useless robot.

01 Accelerated Evolution: Why Is the Industry So Excited?

People are so tolerant because the robot industry, especially the humanoid robot sector, is still evolving its “two wings” at an extremely high speed.

One of the wings is the locomotion ability of bipedal humanoid robots.

This is the third World Robot Conference I've attended. At the 2023 World Robot Conference, Wang Xingxing, the CEO of Unitree Robotics, was still explaining to his robot industry peers why he started working on humanoid robot algorithms the previous year. When interviewed by GeekPark, he mentioned that “Boston Dynamics was far ahead in terms of software at that time.” At that time, only a few humanoid robots on the whole site could walk independently.

In just two years, the locomotion ability of humanoid robots has been greatly upgraded.

Unitree Robotics has, as always, demonstrated first - class motion control capabilities. Watching the combat demonstration on - site, the power of the robots was really astonishing.

This improvement in locomotion ability is not limited to Unitree; it's happening in multiple places.

Zhongqing Robotics excels in human - like movements. Its dancing movements are incredibly human - like.

The newly released full - size humanoid robot LimX Oli by Zhujidongli is 1.65 meters tall and has 31 active degrees of freedom. It also completed brisk walking, model - like strutting, and goose - stepping on - site, including high - dynamic dance and kung - fu performances.

Songyan Dynamics also presented a live demonstration of a robot doing a somersault.

On the other hand, the dexterous and generalizable operation ability of robots is also evolving rapidly.

In previous years, it was already a top - level demonstration if a robot could understand voice commands and slowly pick up a specific small ball and hand it to you. This year, the demonstrations of many companies have targeted the operation of flexible objects, which is a crucial step from “being able to grasp” to “being able to use.”

The difficulty of operating flexible objects lies in their unpredictability. A rigid bottle has constant physical properties, while a piece of clothing or a sachet is amorphous and soft, and each grasp may result in a different shape. To hold it firmly, a robot not only needs to “see” but also “understand,” and it has to predict its deformation within milliseconds and adjust the grip force immediately. The grasping of rigid objects only accounts for a small part of daily life, while flexible objects are the key to entering household and service scenarios.

This year, many companies have shown amazing results in this field:

For example, Xinghaitu demonstrated the ability to pull up a blanket. Although the video was sped up, the robot was able to autonomously reason and find the corner of the blanket to pull it open, and it succeeded in multiple random repetitions. It was a very impressive demonstration.

Zibianliang's sachet - making process also involves the operation of flexible objects throughout.

Looking at the progress of robots on a yearly basis, the improvement is undoubtedly significant.

More importantly, these capabilities are not monopolized by any single company. In this rapidly developing field, new players are constantly entering, bringing more possibilities. For example, Qinglang Intelligence, which is well - known for its service robots, also launched its own bipedal robot this year and plans to implement commercial applications in familiar scenarios such as hotels. Li Tong, the CEO of Qinglang, said in an interview with GeekPark that the technical barriers to bipedal robot research are no longer high, and “if you see an opportunity, you can do it.”

In this high - speed development field, many powerful players founded in 2024 or even 2025 have entered the market, each with their own unique skills, staking their claims in this competitive landscape.

Of course, the continuous flow of investors who smell business opportunities has also contributed to this evolution. They are voting with real money to ensure that this high - stakes bet on evolution has enough fuel to continue at the current dazzling speed.

02 The Somewhat Embarrassing Commercialization: Bipedal Robots Provide Emotional Value, while Operational Robots Look for Niche Scenarios

“Can this robot be commercially viable?”

This was the loudest question floating in the air at the exhibition. Everyone, from investors to entrepreneurs and curious audiences, was very concerned about this topic.

After visiting the entire World Robot Conference, I found a somewhat embarrassing situation: at present, humanoid robots are largely “useless.”

A senior investor estimated for me that the total annual shipment of humanoid/anthropomorphic robots in the entire industry this year might be around 30,000 units.

My on - site visits to various booths gave me an even more conservative impression: most companies' annual shipment targets are only in the hundreds. Those that can exceed 1,000 units are the top players in the industry.

Wang Xingxing, the CEO of Unitree, has always said that the ultimate goal of robot research is to make them work. However, when asked about the value of current bipedal robots other than emotional value, what else is there?

Wang Xingxing's answer was surprisingly honest. He said that the reason for developing bipedal robots is that the technology is relatively simpler, following a “Why not?” logic. Whichever technology matures first will be prioritized.

There is no clear benchmark for emotional value. In the field of bipedal robots, the market competition seems to be intensifying, making the acceptance of “uselessness with value” higher.

Unitree R1 WRC performance | Image source: Unitree Technology

If bipedal robots currently mainly provide emotional value, then operational robots with two hands face another more difficult problem - “the difficulty of being useful.”

Everyone understands the grand scenarios such as factories, smart retail, and hotel catering, but the reality is that the technology is still far from meeting the requirements.

On the one hand, the “brains” of robots are not smart enough. Most current model trainings can only enable robots to perform a few isolated actions. It is still not easy to make a robot successfully complete multiple tasks in a real - world scenario.

On the other hand, dexterous operation remains a major bottleneck, which requires improvements in multiple aspects such as data and hardware technology.

The currently popular VLA route in the industry heavily relies on data. In fact, high - quality operation data also depends on high - precision hardware for collection. And the iteration of hardware itself also requires a certain amount of time and practical operation for feedback and improvement.

This year, some companies have realized this core pain point and started to try to solve it.

For example, Lingcifang, founded in 2025, is said to have shipped 200 units this year. One of the reasons it was able to achieve such rapid shipments after its establishment in 2025 is that one of its core technical capabilities is high - quality data collection.

I also experienced the data collection tool of Xingchen Intelligence at its booth. The staff at the booth introduced that they use a rope - driven transmission and an extremely smooth hardware structure to collect “cleaner” data, which can then be used to train models with equally smooth movements.

While the grand scenarios are still difficult to implement immediately, some companies have adopted a clever strategy. They perform seemingly cool demonstrations such as cleaning the toilet and tidying up the house, but what they are actually selling is not household service robots but development platforms for researchers. These high - difficulty demonstrations are like carefully set “baits” to attract the most potential customers to pay for the exploration of “the difficulty of being useful.”

There are still many companies that choose to face the real environment head - on, collect data on - site, and try to find the entry point for commercialization.

This scenario may need to be somewhat demanding. It requires a certain degree of generalizable operation ability that industrial robotic arms couldn't cover before. At the same time, it should avoid frequent contact with people, high - speed requirements, and the possibility of being easily replaced by cheap labor.

Although the conditions are demanding, such scenarios do exist. For example, some special loading and unloading processes in factories or certain sorting tasks in the service industry. Each company is looking for its own niche area, believing that it can gradually increase the number of test machines from one or two to ten or dozens and gradually penetrate a specific scenario.

So, why does the industry seem not to care about this embarrassing situation of commercialization, whether it's “uselessness with value” or “the difficulty of being useful”?

The answer is simple and pure - although the robots are still far from being truly “useful,” the shipment volume of each company is actually increasing.

As long as the curve is still rising and the technology is still advancing, there is hope.

03 Why Wait? A Long - Distance Race with Incremental Progress

At this year's World Robot Conference, I observed an interesting paradox.

Almost every CEO of a robot company was extremely optimistic when describing the grand blueprint for the next two or three years, but then they would sincerely add, “Please be patient with us. Growth won't be that fast.”

This slightly contradictory attitude has also been accurately reflected in the public's perception.

In the online public opinion field, people's views on humanoid robots are almost polarized. Some worship them as the prelude to the next era, while others sneer at them, claiming that they are just expensive remote - controlled toys.

There is also no consensus within the industry. Some firmly believe that VLA (Visual - Language - Action model) is the “singularity” of technology, and the road to general intelligence is already paved, with only time and computing power being the remaining issues. Others think that the technology is still developing explosively in a chaotic state, and the industry is still waiting for a genius - level breakthrough that can define a new paradigm.

So, where does the unshakable confidence in the industry come from in the midst of all this contradiction, disagreement, and misunderstanding?

There may be only one answer, which is that compared with other cutting - edge technologies, the robot industry has an unparalleled privilege: “Even a small step forward is a victory.”

The fate of autonomous driving is “L5 or nothing.” Every step of commercialization before reaching the perfect end is fraught with risks. In contrast, every millimeter of progress in a robot's model ability may open up a new, extremely niche scenario - even if this scenario only allows a company to sell a few hundred more robots and obtain the resources to survive and fight the next, more difficult battle.

Perhaps capital has also seen the essence of this incremental game and is willing to keep investing. They are willing to endure a longer cycle to bet on a future that can be “redeemed in installments.”

But patience has its limits. How long can they keep betting?

At a forum hosted by NVIDIA, Wang Xingxing of Unitree and Wang He of Yinhe Tongyong had a face - to - face discussion and made predictions about the future of the industry.

Wang Xingxing of Unitree predicted that the industry would at least double in size every year, while Wang He predicted a ten - fold increase every year.

This huge gap between doubling and ten - fold growth is the most honest measure of the industry's future uncertainty.

It means that no one really knows how far the finish line is or the right path to get there.

The only thing they are certain of is this:

In a high - growth industry, speed is the only pass.

This article is from the WeChat official account “GeekPark” (ID: geekpark), author: Li Yuan, published by 36Kr with permission.