Reporter: Luo Yiqi
Editor: Luo Yifan
Video: Tang Nas
During the 2025 Snapdragon Summit in China, Wang Xingxing, founder, CEO, and CTO of Yushu Technology, discussed the advanced roadmap and industrialization considerations for the development of humanoid robots during a guest dialogue session.
He believes that innovation in communication connectivity is crucial for the development of humanoid robots. The mobility and spatial constraints of the robots themselves impose unique and more stringent requirements on chip computing power and power consumption control. Whether it is general AI models, chips, communication protocols, or communication architectures, the future of embodied intelligence requires more open collaboration and joint innovation within the industry to accelerate its development to the next stage.
Regarding specific plans, he mentioned that since the first half of this year, robots have been learning various human dance and martial arts movements, achieving a level of smoothness and effectiveness far superior to previous attempts. However, at this stage, being able to do this is still not enough.
“In the next stage, in addition to performing actions in a fixed sequence, we hope to be able to execute any action in real-time. That is, if we give any command, the robot can start working immediately,” he pointed out, indicating that this would bring the robots close to a state of being able to ‘work’.
He admitted that the company team is still working hard to advance this stage, but they are already very close to the expected goal. “If all goes well, by the end of this year, or at the latest by the first half of next year, we should be able to enable a humanoid robot to generate various actions freely and in real-time.”
This is still not enough. In Wang Xingxing’s plan, by the end of next year, he hopes to enable humanoid robots to operate naturally in unfamiliar environments by conveying any command to the robot. For example, the robot should be able to autonomously find a bottle of water in a new environment and hand it to a guest. “If all goes well, this could be achieved next year, or at the latest the year after.”
When discussing one of the significant challenges currently facing the robotics industry, Wang Xingxing believes it is related to cabling.
Taking the relatively mature field of industrial robots as an example, based on his rough understanding, he noted that currently, 60% to 70% of the factors affecting the failure rate and lifespan of industrial robots are related to cable failures.
“It may be hard to imagine, but sometimes our team spends a lot of time and effort to reduce the weight of a single wire,” he explained. In a sense, reducing the number of cables is particularly important for improving the performance and reliability of a robot.
Wang Xingxing envisions a future where humanoid robots are connected with just one cable between the main control brain and the limbs. Achieving this goal still requires a lot of work to be explored.
Moreover, developing corresponding large model capabilities is also key to enabling robots to achieve better general capabilities.
“I believe that at this point in time, everyone can adopt a more open attitude. After all, the models that everyone develops cannot be directly used, so it is better to be more open,” Wang Xingxing pointed out. This is similar to the early development of OpenAI, where the large model capabilities and commercial value had not yet been realized, and they chose to open-source ChatGPT versions 1 and 2 to attract the industry to jointly promote the development of large model industries.
This is also the reason why Yushu Technology advocates for the promotion of open-source model development. On September 15, the company announced the open-source of UnifoLM-WMA-0, which is said to be Yushu Technology’s open-world model-action architecture across multiple types of robotic bodies, specifically designed for general robot learning. Its core component is a world model that can understand the physical laws of interaction between robots and their environments.
Wang Xingxing emphasized that the company has completely open-sourced the model’s dataset, training source code, deployment source code, and more. “I believe that at this point in time, all companies or scholars in the hardware industry can adopt an open attitude and open-source as much as possible to accelerate the entire industry to the next level. Once we enter the next stage, the market will grow significantly.”
SFC
Produced by 21 Finance Client, 21st Century Business Herald
Editor: Li Yutong, Intern Lin Zhiyi
Recommended Reads from 21
Futu and Tiger Brokers Tighten Account Opening Channels for Mainland Residents, Customer Service Responds
Zhang Xuefeng Responds to Multiple Accounts Being Banned from Following
2% Large Amount Time Deposits Sold Out Instantly
