The Spring Festival Gala stage serves not only as a performance venue but also as a cultural window to observe China’s technological and industrial advancements. This year, 2026, represented by four companies—Unitree Technology, Songyan Power, Magic Atom, and Galaxy General—humanoid robots have transformed from mere auxiliary tools into narrative performers, showcasing a profound integration of cutting-edge technology and traditional festive spirit.
The development of the embodied intelligence industry has evolved to a point where the research and development of robots has transitioned from time-consuming “handcrafted workshops” to “super factories” that require just days, thanks to the availability of vast amounts of high-fidelity synthetic data.
Behind the impressive performances of humanoid robots lies the crucial role of data. While audiences cheer for their complex movements, few realize that these actions are driven by extensive data input. The “2025 Embodied Intelligence Development Report” highlights that a lack of data has become a significant barrier to advancements in embodied intelligence. The challenge lies in acquiring real data, which often comes with high costs and is not widely available in high quality and diversity.
Yang Haibo, co-founder and president of Light Wheel Intelligence, explained to reporters that the demand for data in humanoid robots is enormous. He stated, “The data requirement for humanoid robots is at least 1,000 times that of autonomous driving. While autonomous driving relies on visual algorithms to prevent collisions, humanoid robots need not only visual understanding but also the ability to comprehend physical interactions and movements in the world. For instance, they must differentiate between materials like metal and fabric and understand the varying magnetic and damping forces when opening a refrigerator door.”
Light Wheel has introduced “synthetic data” to effectively address industry challenges. Notably, training and evaluating robots in the real world is prohibitively expensive and inefficient, making it impractical. For example, training a robot to assemble batteries requires building specialized production lines and preparing vast amounts of components. Each collision or failure halts production and necessitates part replacements, leading to months of training cycles and high costs, which deter many companies.
In response to the limitations of real data, there is a growing consensus in the industry that training and evaluating robots through simulation is not optional but a prerequisite for the establishment of embodied intelligence learning. Based in Haidian, Beijing, Light Wheel Intelligence is pioneering this viable path. They have developed a comprehensive self-researched “solving-measuring-generating” triad simulation technology architecture, which comprises a self-developed physics solver, an automated measurement factory, and a scalable evaluation platform, forming a complete and controllable system.
Grounded in this engineering closed-loop, Light Wheel has created a “digital parallel world” that accurately simulates fundamental physical phenomena like gravity, collisions, and friction, and can also replicate complex interactions such as material deformation, liquid flow, and manipulation of flexible objects. In this “digital parallel world,” the typically lengthy training cycles are compressed to weeks or even shorter, costs are reduced by a factor of a hundred, and the skills acquired can be deployed across different entities.
Moreover, Light Wheel’s self-developed training core has been deeply optimized for domestic CPU/GPU/NPU compatibility, fully integrating into the domestic computing ecosystem. The robot team showcased during this year’s Spring Festival Gala owes its stage movements to the indispensable data support provided by Light Wheel Intelligence.
Chinese original simulation technology has gained global market recognition. “The quality of simulation technology and data is ultimately determined by the clients,” Yang Haibo emphasized during the interview. He mentioned that Light Wheel does not just provide data but also possesses algorithmic capabilities, collaborating with clients to iterate on data “formulas.” Currently, all top three global world model teams are cooperating with Light Wheel, and over 80% of the simulation assets and synthetic data in leading international embodied intelligence teams come from them. Major companies such as NVIDIA, Google, ZhiYuan Robotics, Galaxy General, ByteDance, and Alibaba are all clients or partners of Light Wheel, which has quickly gained global recognition thanks to this self-developed simulation technology.
Yang Haibo stated, “We insist on full-stack self-researched simulation because current leading international simulation software does not meet our needs. We must develop our own simulation to ensure our data generation and evaluation capabilities. Additionally, we recognize the critical role that simulation plays as a technological foundation for the development of humanoid robots.” Light Wheel’s leading simulation technology is turning into a standard-setting authority in the international industry.
Light Wheel has collaborated with NVIDIA to define the physical and engineering standards for simulation assets, ensuring that these assets serve practical purposes for training, evaluation, and strategy validation. Together with NVIDIA, they have jointly open-sourced the Isaac Lab-Arena embodied evaluation benchmark framework, systematically integrating tasks, environments, metrics, and acceptance criteria for the first time. Furthermore, Light Wheel is working closely with World Labs, founded by AI pioneer Fei-Fei Li, to build the first scalable evaluation system for embodied intelligence globally. Additionally, Alibaba’s Tongyi Qianwen has also adopted Light Wheel’s embodied evaluation platform to jointly promote the establishment of industrial-grade evaluation standards. Moreover, Light Wheel’s self-developed LeIsaac simulation workflow has been incorporated into the official documentation of Hugging Face, the largest AI open-source community, becoming a “standard configuration” for millions of developers worldwide.
The original simulation technology represented by Light Wheel is continuously evolving through practical iterations with top-tier clients, leading the way in defining global standards. Various agencies predict that by 2026, leading humanoid robot manufacturers will achieve deliveries in the thousands, with sales potentially exceeding 100,000 units. Crossing this threshold will again escalate the demand for data.
Moreover, as embodied intelligence becomes a crucial component of future industries, it has officially entered national strategic planning and is regarded as a new engine for economic growth. If 2025 is considered the “year of embodied intelligence,” then 2026 will undoubtedly be the “year of embodied data.” As a pioneer and core driving force in this field, Light Wheel Intelligence is leading the explosive growth of large-scale embodied intelligence data through its full-stack self-researched simulation capabilities.
Yang Haibo reiterated the importance of synthetic data for the embodied intelligence industry: “Light Wheel Intelligence aims to provide the industry with a physics AI infrastructure centered around simulation, allowing robots to transition from laboratories to commercialization, truly integrating into factories, hospitals, and various other scenarios. Only by controlling the entire chain of data generation and training can we maintain the initiative during fluctuations in computing power, scene changes, and hardware upgrades.” With the continuous accumulation and innovation of synthetic data, the capabilities of humanoid robots will continue to improve. The remarkable performance of robots on the Spring Festival Gala stage not only marks a significant milestone in the scale development and commercialization of the industry but also vividly demonstrates the role of the embodied intelligence industry in empowering human society.
Original article by NenPower, If reposted, please credit the source: https://nenpower.com/blog/the-rise-of-humanoid-robots-at-spring-festival-gala-how-guanglun-intelligent-leads-the-era-of-scalable-embodied-ai-data-through-self-developed-simulation-technology/
