It was nearing midnight on Lunar New Year’s eve when 36Kr met Wang Xingxing, founder of Unitree Robotics, at the company’s robot training base inside a film and television production complex in Beijing’s Daxing district. He had just returned from the live recording of the annual televised gala, which was still airing.
Despite the late hour, Wang appeared energized. As the conversation shifted to technical details, the typically reserved founder became more animated, using his arms to mimic combat sequences, his voice rising as he described the robots’ movements.
“After the performance ended, the stone in my heart finally dropped,” Wang told 36Kr. Although this was Unitree’s third appearance on the gala stage, the pressure had not eased.
“We couldn’t just dance again this year because we danced last year. That wouldn’t be much of a challenge.”
Judging from the final result, the effort appears to have paid off.
In this year’s martial arts program, Unitree deployed 24 G1 humanoid robots to spar with human performers, executing movements that included table vaults, backflips, and spinning kicks.
Wang said the team chose martial arts to better demonstrate the robots’ physical capabilities. To prepare, Unitree and the gala’s directing team reviewed martial arts techniques from around the world and selected dozens of representative moves for the choreography. Behind the fluid performance was a series of technical hurdles.
The difference from last year’s show was noticeable. Previously, transitions relied on slower walking repositioning. This year, the robots ran between formations, reaching speeds of up to four meters per second.
As the number of robots increased, so did the complexity of coordination. Unitree developed cluster control technology to manage group movement. Each robot uses onboard sensors to perceive its surroundings in real time. During high-intensity movements that can cause positional drift, the system enables the robots to recalibrate and return to formation without manual intervention.
Wang said the martial arts program was more than a stage performance. The technologies developed for the show, including cluster coordination and high-speed movement, are intended for real-world applications.
“These technologies are very practical and will help enable large-scale robotic cluster operations in the future,” Wang said.
Behind the spectacle is Unitree’s broader push into new application scenarios in 2026.
Following its appearance at the 2025 gala, Unitree has drawn attention as one of several robotics companies exploring commercial use cases for humanoid machines. In 2026, Wang said that in addition to consumer products, research and education, and commercial performances, the company plans to expand into industrial and commercial service settings.
He estimates that Unitree’s shipments in 2026 could reach 10,000–20,000 units.
“For humanoid robots, the biggest market growth comes from technological maturity,” Wang said. “When technology becomes increasingly valuable, the market gradually opens up.”
The following transcript has been edited and consolidated for brevity and clarity.
36Kr: What inspired the martial arts performance?
Wang Xingxing (WX): We’ve always wanted to do better than before. We couldn’t just dance again this year because we danced last year. That wouldn’t be much of a challenge.
We chose martial arts because it allowed us to challenge more difficult movements and further improve the robots’ athletic capabilities, whether in positioning or physicality.
36Kr: How are you feeling now that the program has concluded?
WX: Over the past one or two months, I personally felt a lot of pressure. We wanted to present something far better than last year on the gala stage, so we needed more standout content and a different format.
We worked closely with the directing team and compiled martial arts techniques from around the world. The robots learned to replicate them, and we selected dozens of the best movements for the show.
Preparation went beyond capturing motions and teaching them to the robots. To make everything coherent, synchronized with the music, and interactive with human performers, we made many fine adjustments. That took a great deal of time.
After this performance, I feel like a weight has been lifted.
36Kr: Many companies showcased during this year’s gala. What does that mean for the industry?
WX: As the industry heats up, it’s normal for more startups to enter the field.
But for Unitree, our biggest competitor is always ourselves. The greatest pressure was last year’s performance. We wanted to surpass it in terms of technology and final presentation.
We refined every movement carefully, aiming to reach a level that’s even more advanced than global benchmarks.
36Kr: What challenges did you encounter?
WX: There were many challenges. To achieve the program’s effect, we completed several firsts in humanoid robotics history.
For example, in last year’s performance, transitions relied on slower walking repositioning. That pacing didn’t look good on stage. This time, the robots were repositioned by running.
The high-speed movement capability was developed specifically for this gala. The robots can run along arbitrary trajectories at speeds of up to four meters per second. This technology is very practical and beneficial for large-scale robotic cluster operations in the future.
We also attempted more extreme movements. For example, there’s a move where the robot takes two steps on a wall before leaping into the air.
At first, the robot could only manage one step on the wall. We felt that wasn’t challenging enough, so we optimized it for a long time until it could step twice and jump higher for a stronger visual effect.
36Kr: How long did Unitree prepare for this performance?
WX: We began preparing in November. Selecting martial arts moves, arranging the storyline, and fine-tuning actions took the most time.
36Kr: The difficulty level seemed high, and there were more robots this year. What were the key breakthroughs?
WX: Last year we had 16 robots. This year, we had 24 G1 units, and the formations were more uniform.
When robots perform martial arts with intense movements, each one may drift slightly. So we developed positioning technology to bring them back into fixed positions, ensuring visual uniformity and precision.
36Kr: There was a moment when a robot fell and stood back up. Was that accidental?
WX: It was part of the storyline. The program is performative, so it needs some humor.
Drunken boxing has that feeling of almost falling. If the robot falls and then stands up on its own, it looks impressive and fits the artistic concept.
36Kr: With so many robotics companies appearing at the gala this year, what’s the value of participating?
WX: The gala is an excellent opportunity to showcase. It’s a national stage to demonstrate products and technology, allowing people across the country to see how far robotics has developed.
36Kr: In 2026, what are Unitree’s priorities in models and application scenarios?
WX: On the modeling side, we’re exploring multiple approaches. There isn’t a unified path for embodied intelligence yet. For example, there are world models, vision-language-action (VLA) models, and VLA models combined with reinforcement learning.
We are exploring various possibilities for the robot “brain,” including collaborating with third-party companies. In artificial intelligence development, there is sometimes an element of luck. No company globally has clearly broken through the ceiling of embodied intelligence yet, so we remain open in this area.
In terms of commercialization, beyond consumer products, research, and education, we will expand into industrial and commercial service applications.
Since 2024, we have promoted humanoid robots for production, assembly, and material handling in automotive factories. However, mass deployment has not yet been achieved. In factories, it ultimately comes down to efficiency compared with humans, and globally robots still cannot outperform people in many cases.
36Kr: Is the consumer market Unitree’s next focus?
WX: We’ve always offered B2C products. Our quadruped and humanoid robots are available on JD.com and Taobao. The R1 humanoid robot starts at RMB 29,900 (USD 4,186).
36Kr: Beyond stage performances, what new real-life applications might emerge this year?
WX: Currently, if we program a robot specifically for a given environment, it can operate well. But every household environment is different. We can’t write a separate program for each home. The embodied intelligence brain is not yet general-purpose enough, so large-scale household adoption remains early.
However, in guided tours or relatively fixed industrial environments, we could see solid deployment results this year or next.
This wave of humanoid robotics has developed for only about three years. If we push too aggressively at an early stage, it could be counterproductive for the industry.
36Kr: How can robots move from the stage into more practical use cases?
WX: We have always valued making robots work in real life. But this is still an early stage of exploration. For now, we are focused on improving overall mobility.
We believe that a robot capable of working effectively must first have strong movement capability. Mobility is a necessary prerequisite for embodied intelligence to translate into real-world applications.
After improving mobility, we can elevate intelligence so robots can perform tasks.
36Kr: Some forecasts suggest global humanoid robot shipments could grow from several thousand to tens of thousands in 2026. Do you agree?
WX: In 2026, global humanoid robot shipments should reach at least tens of thousands of units. We ourselves may reach around 10,000–20,000 units.
For humanoid robots, the biggest market growth comes from technological maturity. If technological progress continues this year, market growth will be relatively predictable. If there is no meaningful technological advancement, shipment volumes may not be as optimistic.
36Kr: Robotics companies are emerging rapidly. Is this driven by technological maturity or short-term capital enthusiasm?
WX: Compared with truly mature, large-scale applications, the current wave is still relatively measured.
Every technological wave has peaks and troughs. We may be at a peak now, and things could cool down. People have expectations for technology, and capital does play a role, but overall I believe the industry remains healthy.
36Kr: How can the industry avoid low-level duplication and destructive competition?
WX: A lot of capital has entered the industry, so homogenized competition is serious. I advocate for rational restraint. We should avoid malicious or ineffective competition. For example, if everyone sells robots at a loss, it could damage the entire industry.
36Kr: Beyond preparing for the show, has Unitree made preparations in production capacity, delivery, and sales?
WX: Last year, as attention toward the company increased, orders rose sharply, and we faced significant production pressure in the first half of the year. This time, we prepared more inventory in advance. We are also improving our sales and after-sales service systems.
36Kr: What is Unitree’s current production capacity?
WX: At this stage, our annual production capacity for humanoid robots is at least in the tens of thousands. For quadruped models, it is also in the tens of thousands.
36Kr: Many robotics companies struggle at the delivery stage. What are the pitfalls?
WX: Delivery is critical because only when robots reach customers can a commercial loop be completed.
After-sales service is extremely important. Customers inevitably encounter issues during use. Often the problem isn’t the product itself, but that robots are still new and customers aren’t fully familiar with them.
We aim to help customers resolve usage issues and improve product usability so that robots can eventually be operated in a simple, user-friendly way.
KrASIA features translated and adapted content that was originally published by 36Kr. This article was written by Qiu Xiaofen for 36Kr.