Wang Xingxing, founder of Unitree Technology: This year, robots will run faster than Usain Bolt.

On March 17th, at the 2026 Yabuli Forum, Wang Xingxing, the founder of Unitree Robotics, delivered a speech. He predicted that although robots can't outrun humans in the 100 - meter race at present, in a few months, by the middle of this year, humanoid robots globally, especially those in China, should be able to run faster than humans, even faster than Usain Bolt. The sprint speed in the 100 - meter race should be within 10 seconds.

Wang Xingxing pointed out that for embodied intelligence to reach a revolutionary moment similar to that of ChatGPT, several key technological challenges still need to be overcome. Among them, the lack of generalization ability is the most core bottleneck: robots can achieve a nearly 100% task success rate in pre - trained scenarios, but their performance will decline significantly when the environment changes, making it difficult to adapt to complex and ever - changing real - world scenarios.

In response to this problem, Wang Xingxing proposed three breakthrough directions: First, improve the model's expressive ability, enhance the richness of robot movement instructions and actions, and lay a solid foundation for generalization ability; Second, improve data utilization. Training data in the robot field is extremely scarce, so higher value needs to be mined from limited data; Third, strengthen the scale effect of reinforcement learning to further unleash the potential of algorithms. He clearly stated that he is optimistic about the world model based on video generation in the long run.

The core logic of this path is similar to human behavior: First, let AI imagine and generate high - quality videos of robots performing tasks in its "brain", then precisely align the generated videos with the robot's actions and convert them into actual execution instructions, enabling the robot to complete complex tasks. In his speech, Wang Xingxing specifically mentioned ByteDance's Seedance 2.0 video generation technology, believing that high - fidelity and highly controllable video generation is an important foundation for this path. At the same time, he admitted that this direction currently faces a common global problem: it is difficult to precisely align and unify the video - generated content with the robot's actual actions. This is the key bottleneck restricting the transformation of the robot's "brain imagination" into "real - world actions". Once this problem is solved, the general model of robots will achieve a fundamental breakthrough, and the "ChatGPT moment" of embodied intelligence will truly arrive.

A month ago, on February 17th, Wang Xingxing said that the global shipment volume of humanoid robots this year would reach at least tens of thousands of units, and Unitree Robotics aims to ship around 10,000 - 20,000 units. However, he also pointed out that due to technological bottlenecks such as the embodied intelligence brain, humanoid robots are still in the early stage of application. According to him, in 2026, Unitree Robotics is developing several new products, including more practical service - type robots and some exploratory products that challenge the technological frontier.

In an interview program launched by CCTV Finance on February 13th, Wang Xingxing expressed a similar view that the future popularity of embodied intelligence may far exceed that of the mobile Internet. "Currently, in the robot industry or the embodied intelligence industry, the biggest driving or uncertain factor is basically dominated by AI, but the growth curve may become steeper. If there are real large - scale applications of embodied intelligence AI models and breakthroughs in robot technology in the next few years, the popularity at that time may be at least 100 times, or even 1000 times higher than now. And I think this popularity will far exceed that of the mobile Internet. So, I think the current stage is definitely not the hottest, and it may still be in the climbing or plateau stage of the climbing process."

On February 23rd, Unitree Robotics officially released a video showing its G1 robot at the Temple of Heaven in Beijing, carrying out an activity called "Martial BOT Prays at the Temple of Heaven". The robot wore a red top, lined up on its own, and re - enacted the martial arts moves from the "Martial BOT" program at the 2026 CCTV Spring Festival Gala. During the long holiday, Unitree Robotics also released videos of the robots forming characters to pay New Year greetings and behind - the - scenes footage. Wang Xingxing demonstrated the robots' cluster collaboration and rapid scheduling capabilities on site.

This is also the third time that Unitree Robotics has appeared on the Spring Festival Gala. Previously, in the 2021 Spring Festival Gala of the Year of the Ox, Unitree's robotic ox "Benben" made its debut; last year, Unitree's humanoid robots participated in the "Yang BOT" program directed by Zhang Yimou and successfully achieved a "phenomenal" breakthrough.

According to a clarification statement about the 2025 sales data released by Unitree Robotics on January 22nd, the actual shipment volume of the company's humanoid robots in 2025 exceeded 5,500 units (referring to the actual number of units shipped to end - customers, not the order quantity, which was higher), and the number of mass - produced units offline in 2025 exceeded 6,500 units. The above numbers are all for Unitree's pure humanoid robots, excluding other robot products such as two - armed wheeled robots.

Public information shows that Unitree Robotics' current product portfolio covers three major series: quadruped robots, humanoid robots, and dexterous robotic arms. Among them, the consumer - grade humanoid robot G1 was launched on JD.com as early as last February, with a starting price of 99,000 yuan; the relatively lightweight R1 (about 1.2 meters tall) is priced from 29,900 yuan. Another model, the H1, which appeared on the 2025 Spring Festival Gala stage, is priced as high as 650,000 yuan.

This article is from the WeChat official account "Jiemian News", author: Peng Peng, published by 36Kr with authorization.