The Ministry of Industry and Information Technology (CWW) recently approved the release of the “YD/T 6770-2026 AI Key Basic Technology Tailored Intelligence Benchmark Testing Method”, which is the first industry standard in the field of embodied intelligence. It will be officially implemented on June 1st, marking a new stage of embodied intelligence evaluation with “standards to follow”.
This standard establishes a unified benchmark testing framework for the field of embodied intelligence, standardizing the environment settings, task library construction, testing process, and indicator calculation methods for conducting embodied intelligence benchmark testing in both simulated and real environments. At present, the standard has been simultaneously promoted for international standard approval.
Experts from the China Academy of Information and Communications Technology have stated that a reliable evaluation system serves as a bridge and benchmark connecting technology research with large-scale industrial applications. This standard will provide a unified technical specification framework and establish a foundation of mutual trust for the industry. At the level of technological research and development, providing a unified measurement basis for the assessment of embodied intelligence capabilities, guiding the direction of technological iteration, helping the industry identify and converge to high-value technology routes early, and reducing the ineffective investment of research and development resources; At the application implementation level, provide standard and standardized support for product selection and application verification for industry users, avoiding the phenomenon of “bad money driving out good money” in the market; At the level of industrial development, we will accelerate the transition of embodied intelligence from the laboratory to real-life scenarios, and promote the engineering and industrial application of embodied intelligence technology achievements.
Figure 1 “YD/T 6770-2026 AI Key Basic Technology Embodied Intelligence Benchmark Testing Method”
The overview of “YD/T 6770-2026 AI Key Basic Technology Embodied Intelligence Benchmark Testing Method” is shown in Figure 1, which describes the benchmark testing framework, indicators, and methods for embodied intelligence systems in simulated and real environments, including benchmark environment settings, task library construction, testing process, and indicator calculation methods. It is applicable to benchmark the “perception decision execution” full chain capability of a single embodied intelligence system.
This standard establishes a unified embodied intelligence benchmark testing framework (as shown in Figure 2), which includes four parts: environment settings, testing task library, testing process, and indicator calculation. In terms of testing environment, the construction requirements for simulation testing environment and real testing environment have been stipulated to ensure that the environment settings can reflect the complexity of the real environment. In terms of testing tasks, a method for constructing a testing task library and a generalization evaluation benchmark have been proposed to support the evaluation of the performance and generalization of the tested object in different task complexities and ability dimensions. In terms of testing process, testing is mainly conducted on two types of testing objects: models and deployment models of the entire system. Testing methods include static simulation testing, dynamic simulation testing, real environment testing, and combination testing. In terms of testing indicators, it mainly includes five indicators: task execution efficiency, task success rate, human intervention rate, scene disturbance attenuation rate, and average task energy consumption.
Figure 2 Framework diagram of embodied intelligence benchmark testing
The formulation of this industry standard is of great significance for promoting technological progress, industrial application, and standardization construction in the embodied intelligence industry. On the R&D side, it provides a unified measurement basis for the assessment of embodied intelligence capabilities, guides the direction of technological iteration, helps the industry identify and converge to high-value technology routes early, and reduces the ineffective investment of R&D resources. On the application side, provide standard and standardized support for product selection and application verification for industry users, avoiding the phenomenon of “bad money driving out good money” in the market. On the industrial side, accelerate the transition of embodied intelligence from the laboratory to real-life scenarios, and promote the engineering and industrial application of embodied intelligence technology achievements.