Zhongguancun Forum | On the Eve of Large-Scale Deployment of Embodied Intelligence: Industry Standard Development as the Key to Breaking into the Trillion-Yuan Track

Reporter Li Jing from Zhongjing reports from Beijing

From the laboratory to real-world scenarios such as industrial logistics and public services, moving from a hundred billion in output value to a trillion market, embodied intelligence is ushering in a critical window for commercialization. After the industry completes technological foundation and scenario verification in 2025, 2026 is already regarded as the inaugural year for the scaling application of embodied intelligence.

However, the embodied intelligence industry is facing issues such as the lag in standardized construction of data interoperability, evaluation systems, and security regulations, which have become one of the bottlenecks restricting the industry from moving from “breakthroughs in isolated points” to “widespread adoption.” Reporters from China Business News noted that at the recent 2026 Zhongguancun Forum annual meeting, several leading entrepreneurs and experts in the embodied intelligence sector discussed the inadequacies of the industry’s standardization system in various forums.

Challenges of Standards

“To be frank, everyone today is talking about models and algorithms, but the embodied intelligence industry currently doesn’t even have a large-scale, recognized benchmark to evaluate model capabilities; everyone is still testing on some very basic simulation datasets,” pointed out Tang Wenbin, founder of Original Force. “Currently, the performance of each company’s model may excel on basic simulation datasets, but once it enters the complex and variable real physical world, its generalization ability and reliability lack a unified, objective measurement scale.”

At present, the data formats collected by each company’s robot “brain” vary, and the evaluation systems operate independently, leading to low efficiency in technological iteration and industrial collaboration. This is akin to a market without recognized measurement standards, making transactions difficult to conduct efficiently. Therefore, establishing an authoritative and open evaluation system for the embodied intelligence industry has become one of the most urgent standard needs.

Behind the evaluation standards is a deeper issue of data. Sun Mingjun, director of the Zhongguancun Intelligent Use of Artificial Intelligence Research Institute, pointed out to reporters: “The data collected by each robot cannot interconnect or recognize each other. This is actually a particularly important issue hindering the current development of embodied intelligence.”

Wang He, founder of Galaxy General, likened the data required for embodied intelligence to a “pyramid”: the base consists of vast amounts of internet video data, the middle layer is human behavior demonstration data, and the top layer is the expensive yet indispensable real robot operation and feedback data. However, currently, the data collection protocols and annotation systems of various embodied intelligence companies are incompatible, forming serious data silos.

“Data is definitely one of the bottlenecks now,” Tang Wenbin noted, “But today, collecting data is essentially a matter of money and time; more critically, it’s about how to enable robots to be used in real scenarios in bulk and form a closed-loop of data feedback.”

Zhang Peng, co-founder of Zhihui Square, also emphasized that data from real scenarios is the most valuable, but how to share it with clients while ensuring safety and how to use synthetic data and other technologies to reduce costs requires urgent industry consensus.

As technology attempts to enter reality, more complex issues of safety and responsibility standards follow closely. Xi Yue, co-founder of Star Motion Era, pointed out that behavioral safety standards are “the most urgent and fundamental.” After all, robots are entities acting in the physical world, and once they enter industrial, logistics, or even future home scenarios, their behavioral boundaries, error tolerance, and accident liability definitions need to be clarified.

“On the one hand, we cannot restrict the development of the entire industry; on the other hand, we need to grow and gradually improve standards as we make mistakes in a controllable manner,” Xi Yue believes that this requires the establishment of a three-dimensional framework combining common safety standards and industry-specific safety standards.

Moreover, the most important landing scenarios for embodied intelligence are in the industrial sector, and the interface standards for collaboration with industry also urgently need to be established.

Gao Yang, co-founder of Qianxun Intelligent, compared the future humanoid robot to a complex whole like a laptop or car, involving numerous parts and subsystem suppliers. However, currently, the interfaces between remote control systems, data collection equipment, and the robot body are varied, leading to extremely high adaptation costs.

“For example, what kind of instructions should the remote control interface send to the robot? This is still a very painful and non-standard process,” Gao Yang called for the need to co-build standardized interfaces to clarify industrial division of labor and accelerate the development of the entire ecosystem.

Path to Resolution

For the challenges of standard absence facing the embodied intelligence industry, individuals from both the industry and academia view “co-building” and “collaboration” as the key to breaking the deadlock.

Sun Mingjun introduced that the inability to interconnect data is a key difficulty hindering the current development of the embodied intelligence industry. To this end, the Ministry of Industry and Information Technology has established a standardization technical committee for humanoid robots and embodied intelligence, and the “Humanoid Robot and Embodied Intelligence Standard System (2026 Edition)” has recently been released, with our country promoting the construction of an industry standard system from the top down.

In the industrial sector, different demands exist: robot body manufacturers with a certain market scale may tend to establish a standard system centered on themselves, while companies focused on visual language models (VLM), motion control “brains,” or specific components are more eager for open and interconnected universal standards.

Sun Mingjun pointed out to reporters: “Looking at the development history of other fields, interconnectivity, maximum consistency, and open source are undoubtedly the direction that this field must take.”

Zhang Peng proposed a three-layer framework for standard construction: the first layer is data and evaluation standards; the second layer is an evaluation system for robot intelligence levels and operational capabilities, which can refer to the L1-L5 grading of autonomous driving to establish industry consensus; the third layer is forward-looking discussions on legal regulations, clarifying robots’ behavioral guidelines and accident liability definitions.

Moreover, the exploration of embodied intelligence in specific scenarios is itself the best testing and promotion of industry standardization. The industry is optimistic about the key development scenarios in 2026, including industrial manufacturing, logistics warehousing, and public services (such as retail space capsules), which all exhibit characteristics of being “semi-structured” and “error-tolerant.”

Wang He shared that the hundreds of “space capsule” robots deployed by Galaxy General in dozens of cities, while demonstrating autonomous picking capabilities and educating the market, have accumulated over 80,000 hours of real-world autonomous picking data. This data closed loop formed in real scenarios, including environmental interactions and cases of success and failure, will play an important role in future standard-setting for operations, safety, data formats, and more.

Wang He believes that through data feedback from real scenarios and model iteration, the embodied intelligence industry is expected to achieve a critical leap from the “GPT-2 era” to the “GPT-3 era” in 2026.

From hundreds of billions to trillions, the blueprint for the trillion-dollar track of embodied intelligence is becoming clear, while the construction of standards serves as the “pass” for this leap. As multiple industry insiders have said, the construction of the standard system for embodied intelligence is not the work of a single entity, but requires collaboration among governments, enterprises, and research institutions, forming consensus in practice, gradually perfecting the system through co-construction, so as to allow technological innovation and scene landing to form a positive cycle, promoting embodied intelligence to truly integrate into various industries and become a new engine for the development of the digital economy.

(Editor: Zhang Jingchao Reviewed: Li Zhenghao Proofread: Yan Jingning)