From the record-breaking robot marathon, see China's embodied intelligence

When the Glory "Lightning" robot crossed the finish line in 50 minutes and 26 seconds, it not only outperformed the human men's half - marathon world record but also cut the robot's half - marathon time from a year ago by 110 minutes. On April 19, this "legendary moment" at the 2026 Beijing Yizhuang Humanoid Robot Half - Marathon set the entire embodied intelligence industry ablaze.

However, this racing feast that gathered 102 teams and over 300 robots was never just a simple "show of skills." The real value of this robot marathon lies in peeling back the true nature of China's embodied intelligence: hardware is iterating at breakneck speed, and capital enthusiasm remains high. But there is still a long gap to cross from "being able to run" to "being useful."

01 More Than a Race, It's an Industry "Medical Report"

For ordinary people, a robot running a marathon might just be a "spectacle of high - tech," but for industry insiders, it's the most rigorous "full - dimensional medical examination."

This event set three historical records: 102 teams participated, the completion rate exceeded 45%, and 18 teams completed the race with autonomous navigation. The most astonishing was the championship result of the Glory "Monkey King" team—50 minutes and 26 seconds. Compared with the 2 - hour, 40 - minute, and 42 - second result of the champion in the first event in 2025, in just one year, the robot's half - marathon time was directly shortened by nearly 110 minutes, equivalent to leaping from "toddling" to "striding briskly."

Behind this is the concentrated explosion of four core technologies: overall weight reduction, joint heat dissipation, battery life, and autonomous navigation, which finally allowed robots to shed the label of "remote - controlled puppets." In other words, the track of this marathon is essentially a touchstone for the embodied intelligence industry to move from the "laboratory" to the "real world."

02 Chinese Players in the Capital Carnival

The popularity of the robot marathon is just a microcosm of China's embodied intelligence track. The current industry is in a contradictory period of "coexistence of capital frenzy and technological bottlenecks."

First, look at the booming capital side: In 2025, the industry's total financing scale exceeded 51.1 billion yuan, 3.5 times that of 2024. In just over two months of 2026, the financing amount was close to 30 billion yuan, more than seven times that of the same period last year. A 2.5 - billion - yuan financing by Galaxy General directly refreshed the domestic single - round financing record in the embodied intelligence industry, with a post - investment valuation exceeding 20 billion yuan.

The capital madness has spawned a group of "billion - dollar unicorns"—12 humanoid robot companies such as Unitree Robotics, Zhipu AI, Galaxy General, Qianxun Intelligence, Zhipingfang, Xinghaitu, Independent Variable Robotics, Xingdong Jiyuan, and Pacini Perception have crossed the billion - yuan valuation threshold. Among them, 6 have completed full - stack layouts, and 5 have stood out in the independent R & D of the embodied "big and small brains."

But behind the excitement, the industry's "weaknesses" are also obvious. The most core bottleneck is the lack of data. Yao Maoqing, co - founder of Zhipu Robot, revealed that the training corpus of GPT - 5 is equivalent to about 10 billion hours, while the entire industry's embodied data is only about 500,000 hours, not even a fraction of what is needed to train a high - quality model. In addition, 60% of the participating robots still rely on remote control. Autonomous navigation robots frequently fall and lose control, and the closed - loop of "perception - prediction - control" has not been established. No matter how well - developed the hardware "muscles" are, the "brain" reaction speed can't keep up.

The more practical problem is that the commercialization path is still being explored. Zhou Zipeng, executive general manager of CICC Research Institute, said bluntly: "The problem in the track is no longer 'whether there is capital' but 'to whom the capital is concentrated.' Mass - production capacity and system reliability are the core criteria for screening players."

03 How Long Will It Take to Go from "Track Show - off" to "Scene Profit - Making"?

In 2026, the entire industry defines it as the "year of commercialization expansion" for embodied intelligence. But all players are aware that there is still a long way to go from "breaking records in the competition" to "making money in the market."

First, look at some optimistic data: IDC predicts that the global humanoid robot shipments will exceed 18,000 units in 2025 and will exceed 510,000 units by 2030, with a compound annual growth rate of nearly 95%. Morgan Stanley gives an even more aggressive prediction that the sales volume of Chinese humanoid robots will reach 28,000 units in 2026, a year - on - year increase of 133%.

But commercialization is never an "overnight" thing. The industry has reached a consensus: start with industry, then move to business, and finally to households. Taking it step by step is the way to go.

The industrial scenario breaks through first: Structured scenarios such as logistics sorting, automobile assembly workshops, and industrial inspections have become the "first stop" for robot implementation. In these scenarios, the tasks are single, and the environment is controllable. Robots only need to perform simple operations to achieve cost reduction and efficiency improvement, which is the core driving force for enterprises to pay for them.

The business and special scenarios are breaking the deadlock: The Warrior Challenge added to this marathon allows robots to go up mountains, cross rivers, and climb slopes, simulating real - world scenarios such as engineering construction and emergency rescue. Qinglang Intelligence has put humanoid robots into chain coffee shops, and AutoNavi has played a "trump card"—the quadruped robot "Tutu," which serves visually impaired people as a guide dog, achieving full autonomous navigation in an open environment.

The household scenario is still an unattainable "ultimate goal."

Hu Biao, an associate professor at China Agricultural University, hit the nail on the head: "Currently, it's still humans taking care of robots. Behind one robot, there may be 3 to 5 engineers for maintenance. Since we haven't fully grasped the factory scenario, it's too early to talk about household implementation."

As for the much - anticipated "AGI moment of embodied intelligence," the industry generally holds an "optimistic but cautious" attitude. Mao Jiming, a partner at Jijia Vision, believes: "The path is becoming clearer. It is expected to reach the 'GPT - 3 moment' by the end of 2026, but there is still a long way to go to truly achieve general intelligence."

04 Beyond the Race, Implementation Is the Real Competitiveness

There are two highlights worth focusing on in this robot marathon—the "speed breakthrough" of Glory and the "scenario breakthrough" of AutoNavi, which correspond to the two development paths of the current embodied intelligence industry.

Glory's championship is undoubtedly a concentrated manifestation of its years of in - depth exploration of "core technologies." The "Lightning" robot is 169 cm tall, with an appearance modeled after top human athletes. It is equipped with a self - developed high - dynamic motion system, and its structural design borrows from mobile phone simulation technology. The self - developed liquid - cooling system solves the battery life pain point. The 50 - minute and 26 - second result not only surpasses similar robots but also outperforms elite human athletes, highlighting the differentiated advantages of consumer electronics manufacturers entering the field.

For example, the heat - dissipation technology and simulation technology that have attracted much external attention recently. Glory has directly transferred its strong heat - dissipation capabilities from mobile phones to robots. The self - developed liquid - cooling heat - dissipation system installed in the "Lightning" is the core for it to run at high intensity for 50 minutes without a decrease in speed. In long - distance running, the heat - dissipation ability directly determines whether the robot will "overheat" and stop working.

At the same time, simulation technology provides key data support for robot material and hardware reliability verification, becoming the core for quickly developing a marathon humanoid robot.

According to People's Daily Online, in 2025, Glory's R & D investment accounted for 11.5% of its annual revenue. The Alpha Strategy released last year plans to invest over $10 billion in building an AI ecosystem in the next five years.

So the threshold for the humanoid robot industry is very high, and not just any mobile phone manufacturer can easily cross over.

But we must be clear - headed that race results do not equal market value. As the on - site audience saw, some robots stopped and started, some rushed straight into the isolation barriers, and some were even "carried away" midway. These "flip - over" scenes in the race precisely show that no matter how fast the robot is, if it cannot adapt to the complex changes in the real scenario, it will ultimately be just a "toy in the laboratory."

In contrast, the appearance of AutoNavi's "Tutu" is even more eye - catching. This quadruped robot did not participate in the race but accompanied a visually impaired teenager to complete an obstacle course. Without relying on a preset route or manual remote control, it can independently judge road conditions, avoid risks, and even understand user intentions to complete the guide - dog task in an open environment.

The AI technology self - media Houchangcun said: Although this robot marathon has made great progress, beyond the race, implementation is the real battlefield. This "marathon" of embodied intelligence has just begun.

05 Conclusion:

A robot marathon has shown us the "rapid development" of China's embodied intelligence: the hardware iteration speed is leading globally, capital enthusiasm remains high, and scenario implementation has begun to break through. But we should also see that the industry is still in the "early stage," and data bottlenecks, lagging models, and implementation difficulties are still gaps that all players need to cross.

The ultimate goal of embodied intelligence has never been to make robots win marathons but to make robots become human helpers and enter factories, households, and every scenario where they are needed. In this industrial marathon, the race is just the starting point, and only hard work can lead to long - term success.

This article is from the WeChat official account "Houchangcun," written by Houchangcun and published by 36Kr with authorization.