Currently, embodied AI has become the core track of global AI industry competition, with the focus of competition among countries shifting from technological R&D to the game of standards and rules. The US, Japan, and South Korea have laid out standards for the entire chain of underlying technologies, safety and compliance, and mass production applications in advance, seizing key seats in international standardization organizations and continuously building rule barriers. China's humanoid robot and intelligent equipment industries have significant scale advantages, but the construction of the standard system has prominent shortcomings such as lagging supply, insufficient cross-field collaboration, weak international discourse power, and weak support for testing and certification, making it difficult to match the needs of rapid industrial iteration and global expansion. To seize the strategic window period when international standards are not yet fully finalized, this article systematically reviews the characteristics of global standardization layout, analyzes the existing domestic bottlenecks, and proposes four types of practical measures: system improvement, collaborative construction, international positioning, and basic empowerment, to help enhance the global rule competitiveness of China's embodied AI.
Global Development Trends of Embodied AI Standards
The global competition in embodied AI has shifted from technological innovation and iteration to the game for the dominance of standard rules. The international standard system is accelerating its solidification. Relying on industrial and technological advantages, the US, Japan, and South Korea are deploying in a three-dimensional manner, forming strong rule barriers and narrowing the window for track competition.
The global competition pattern is iteratively upgrading. Global embodied AI has entered the critical phase of "standards first, rules set the game." Standard discourse power has become the core tool for reshaping the value distribution of the global industrial chain, and the window for industry positioning is narrowing. First, developed countries take the lead in defining technological paths. The US dominates the hardware interface standards based on chip architectures. Most global enterprises need to adapt to these standards to ensure compatibility, indirectly controlling the direction of hardware technology evolution. The EU adheres to "legislation first," establishing data privacy and dynamic authorization principles through the Artificial Intelligence Act, setting strict compliance thresholds for embodied AI data collection and processing, and attempting to build market access barriers with high standards. Second, data and simulation standards become a new focus. Addressing the bottlenecks of high costs for real-machine data and weak generalization capabilities, the world is promoting virtual-real data fusion standards. For example, the DOME platform attempts to unify data interfaces and format specifications to break data silos. Establishing simulation benchmarks and data open platforms for key scenarios such as intelligent manufacturing and high-risk operations has become a key measure to build a common language for the industry. Third, safety governance and responsibility division standards urgently need to be unified. As embodied AI transforms from a tool to a quasi-social subject, accident liability determination and ethical guidelines have become the uncharted waters for standard formulation. The international community is accelerating the construction of hierarchical protection mechanisms, exploring AI liability insurance models, and regulating the safety bottom lines for group collaboration and autonomous navigation.
The US dominates the underlying core rules. Relying on the technological monopoly advantages of leading enterprises, it controls the underlying foundation and general technical standards of embodied AI, seizes core seats in international standardization organizations, and builds systematic rule barriers. First, relying on leading enterprises to control the underlying technology ecosystem. Leveraging advantages in cloud computing, large models, and chip architectures, it builds a closed-loop system of technology platform and industrial ecosystem. For example, by defining hardware interface standards based on chip architectures, NVIDIA requires most global robot enterprises to adapt to these standards to ensure compatibility. This enables the US to control key links such as app stores, cloud services, and AI interfaces, thereby manipulating the distribution of global digital economy revenues. Second, seizing international standard seats to build systematic rule barriers. The US deeply participates in the formulation of technical standards by organizations such as the International Organization for Standardization (ISO) and the International Electrotechnical Commission (IEC), and has long dominated the World Trade Organization's e-commerce negotiations, vigorously promoting rules favorable to its expansion, such as the free flow of cross-border data and the prohibition of data localization. Third, using administrative means to suppress global competitors. The US combines technical standard rules with legal regulations, restricting other countries' access to key technologies such as advanced chips and lithography equipment through means like the entity list, export controls, and security reviews. It packages its own technologies, standards, and systems as open and free, while describing other countries' data localization and digital sovereignty measures as trade barriers.
Japan and South Korea delve into industrial application standards. Japan focuses on segmented fields such as safety compliance and human-machine interaction to create a refined standard system; South Korea closely follows mass production scenarios, promoting the deep binding of standards and industrial commercialization. Japan delves into safety and human-machine interaction, building trust with soft rules. The core of Japan's standardization strategy is to respond to the social reality of its deep aging and labor shortage, so the focus of standard formulation always falls on trust and safety in human-machine coexistence. As the earliest country to face a super-aged society, Japan's standards for elderly care and nursing robots are globally leading. Its robots not only emphasize functions but also pursue delicate and humanized experiences such as emotional companionship and etiquette interaction. For example, SoftBank's Pepper robot can recognize emotions and interact, backed by a whole set of standard considerations regarding human-machine interaction and affective computing. South Korea closely follows mass commercialization, driving industrial development with a certification system. South Korea's standardization is a targeted industrial sprint, aiming to pave the way for large-scale commercialization through standards. The South Korean government has set a clear goal: to have humanoid robots in 10 industries enter commercial use by 2028. To this end, standards must come first, and a humanoid robot safety certification center is being built vigorously, aiming to cover the full life cycle verification from AI model reliability to communication security vulnerabilities, nipping risks in the bud.
China's Advantages and Disadvantages
Favorable Conditions for China to Promote the Standardization of Embodied AI
1. Prominent Advantages in Industrial Foundation and Scenario Applications
First, the industrial foundation for humanoid robots and intelligent equipment is relatively complete. China has formed a relatively complete industrial chain covering core components, complete machine manufacturing, system integration, and scenario applications. It has strong industrial supporting capabilities in key links such as servo systems, reducers, sensors, and power systems, which can provide a good industrial foundation for the standard verification and scaled application of embodied AI. In particular, relying on advantages such as complete manufacturing categories and complete industrial chain supporting, new technologies and products can quickly enter actual scenarios such as production manufacturing, warehousing and logistics, and commercial services, forming a rapid cycle of R&D, verification, iteration, and application. Second, rich application scenarios provide a large-scale practical foundation for standard formation. China has a large manufacturing scale and many types of industrial scenarios. At the same time, it has extensive embodied AI application needs in fields such as elderly care services, public services, commercial retail, and special operations, which can provide a large number of real scenarios and test data for standards such as safety, human-machine interaction, task execution, and performance evaluation. Compared with some countries that mainly rely on a single industry or specific scenario to promote standard formulation, China has the unique advantage of extracting common requirements from multiple industries, regions, and tasks, and has the conditions to take the lead in forming a standard system adapted to scaled applications.
2. Prominent Advantages in Technological Innovation and Engineering Transformation
First, the integration speed of AI and robotics technologies is relatively fast. In recent years, China has continued to make breakthroughs in fields such as large models, multimodal perception, motion control, environment understanding, and dexterous manipulation. Embodied AI enterprises can quickly combine AI algorithms with robot bodies and continuously optimize product performance through actual deployment, providing good conditions for the transformation of technical indicators into standard clauses. Second, the advantages in engineering and cost control are obvious. Relying on a mature manufacturing system and supply chain, China's robot products have strong advantages in component selection, complete machine integration, scale manufacturing, and cost optimization. A batch of technical solutions has shifted from laboratory verification to pilot testing and batch applications. Standard formulation can fully absorb industrial engineering experience, combining technological advancement, production feasibility, and application economy, to enhance the industrial applicability and international competitiveness of China's standard solutions.
3. Prominent Advantages in Data Resources and Training Verification
First, large-scale applications can continuously accumulate multimodal data. The development of embodied AI highly depends on multimodal data such as vision, voice, touch, motion trajectories, and environmental interaction. China's scenarios such as industrial manufacturing, logistics and warehousing, and commercial services are large in scale, providing conditions to form rich real-world data resources, which offer practical support for the formulation of standards for data collection, annotation, governance, training, and evaluation. Second, the construction of training grounds and public data platforms is accelerating. Many places in China are accelerating the layout of embodied AI training grounds, data collection platforms, and pilot verification bases. Through the collaboration of real data and simulation data, they provide infrastructure support for model training and complete machine testing. As standards for datasets, training grounds, and evaluation are gradually established, China is expected to rely on its massive scenario and data advantages to form data specifications and evaluation systems with international influence.
4. Prominent Advantages in Policy Guidance and Organizational Mobilization
First, policy guidance and top-level coordination continue to strengthen. Embodied AI has become an important direction for China to cultivate future industries and develop new quality productive forces. The state and local governments continue to strengthen policy deployment in fields such as humanoid robots, AI, and intelligent manufacturing, gradually embedding the standard system construction into the whole process of technological research, industrial cultivation, and scenario application, providing strong policy support for the research, verification, and promotion of embodied AI standards. The Standard System for Humanoid Robots and Embodied AI (2026 Edition) has been compiled by the Standardization Technical Committee for Humanoid Robots and Embodied AI of the Ministry of Industry and Information Technology, organizing more than 120 units. The National Robot Standardization Technical Committee also has the TC591/SC5 Sub-technical Committee for Humanoid Robots. Second, the advantages in organizing and coordinating standardization resources are obvious. Relying on professional institutions such as the National Information Technology Standardization Technical Committee, the National Robot Standardization Technical Committee, and the National Data Standardization Technical Committee, China can effectively gather innovation entities in the industrial chain such as research institutes, universities, leading enterprises, and testing and certification institutions to carry out collaborative research on key common technologies and urgently needed industrial standards. In particular, during the window period when international standards are not yet fully finalized, China has achieved breakthroughs in international standard projects in fields such as humanoid robot datasets, laying a good foundation for further transforming industrial scale, application scenarios, and technological innovation advantages into standard rule advantages. In 2026, the Humanoid Robot Dataset led by China has become the first international standard project in the field of embodied AI, and promoted the establishment of ISO/TC 299/WG16, with a Chinese expert serving as the convener.
China's Shortcomings
1. Lagging and Imbalanced Supply of the Standard System
First, the top-level architecture was formed late, and the supply of mature standards is insufficient. It was not until February 2026 that China formed the first top-level standard design covering the entire industrial chain and full life cycle. There is still a time gap between the rapid development of the industry and the supply of mature standards. Currently, a large number of key standards are still in the research and development stage. For example, the national standard plan for Humanoid Robot Dataset Part 1: General Rules was issued on August 6, 2025, with a project cycle of 18 months, and is currently in the formulation process. Standards for embodied AI training grounds, evaluation, trustworthiness, operating systems, and dexterous manipulation are also mostly in the drafting or approval stage, making it difficult to support rapid product iteration in a timely manner. Second, the coverage structure of standards is unbalanced. By 2026, specialized standard systems for humanoid robots and embodied AI, specialized sub-technical committees, and a batch of embodied AI projects have emerged, forming a specialized layout. However, mature published standards are still relatively limited, lacking mature and widely accepted unified indicators for multimodal data formats, model and body interfaces, virtual-real transfer, continuous learning, dynamic human-machine safety, and long-sequence task acceptance. The inconsistency of interface standards further exacerbates the difficulty of cross-platform adaptation, increases repeated R&D costs, and restricts specialized division of labor and scaled mass production.
2. Cross-field Standard Coordination Mechanisms Still Need to be Strengthened
First, the standard management system is relatively scattered. The humanoid robot dataset standards are managed by the National Robot Standardization Technical Committee TC591, the embodied AI grading, system specifications, and training ground guidelines are managed by the National Information Technology Standardization Technical Committee TC28, and the relevant specifications for high-quality datasets are responsible by the National Data Standardization Technical Committee TC609. Although the parallel layout of multiple technical committees conforms to the interdisciplinary characteristics, it also increases the difficulty of terminology unification, interface connection, and project coordination. Second, the closed loop of research, production, standardization, and application has not yet been opened. The drafting entities of existing standard projects are mainly concentrated in research institutions, platform enterprises, and leading complete machine manufacturers. The participation of small and medium-sized component enterprises, testing institutions, and terminal scenario users is still insufficient, which is prone to the problem of emphasizing technical descriptions while neglecting engineering acceptance. Different enterprises build their own datasets, communication protocols, and evaluation benchmarks, leading to problems such as difficult data sharing, difficult mutual recognition of results, and repeated adaptation of software and hardware. The pre-guiding role of standards in product R&D and scenario landing is insufficient.
3. Weak Discourse Power in International Rules
First, the ability to set core agendas is still insufficient. The secretariat of the core international AI standardization organization ISO/IEC JTC 1/SC 42 is undertaken by the US ANSI, and the secretariat of the core committee for robot standards, ISO/TC 299, is the Swedish SIS. A standard system covering robot safety, performance, interfaces, and interoperability has been formed. It was not until April 2026 that China took the lead in establishing the Humanoid Robot Dataset Working Group in ISO/TC 299 for the first time, achieving a zero breakthrough in the construction of international robot standard working groups. Second, the output of international standards does not match the industrial scale. The international standard for humanoid robot datasets led by China was approved for project establishment with an 88% approval vote, indicating that some advantageous technologies already have the ability of international output. However, it still belongs to a single-point breakthrough. In fields such as robot safety, model interfaces, testing and evaluation, data governance, and scenario certification, a systematic and sustainably exportable Chinese solution has not yet been formed, and the market scale advantage has not been fully transformed into international proposal power and rule dominance.
4. Insufficient Support Capacity for Standard Transformation
First, the foundation for testing, evaluation, and experimental verification is weak. Standards such as embodied AI evaluation, trustworthiness requirements, data generation platforms, and training ground construction have not been fully implemented. The supply of national-level public testing platforms, unified benchmark datasets, and authoritative third-party certification services is insufficient. It is difficult to horizontally compare the performance of enterprise products, and application units also lack unified bases for procurement, acceptance, and liability determination. Second, the channel for international benchmarking and mutual recognition is unsmooth. For embodied AI products to enter markets in different countries, they need to meet local requirements for robot safety, functional safety, electromagnetic compatibility, cybersecurity, and data compliance. For example, the US market has formed robot testing and certification systems such as UL 3300 and ISO 13482. Domestic enterprises often need to retest and rectify. Currently, China still mainly focuses on the product going global of complete machines and components, and a complete chain of coordinated output of standards, testing, certification, and solutions has not yet been formed.
Next Development Trends
The construction of global embodied AI standards is moving from fragmented special formulation to a new stage of systematic, internationalized, and scenario-based collaborative promotion. Rule competition is shifting from the competition of single technical standards to the comprehensive rule game of the entire industrial chain and full life cycle. Relying on comprehensive advantages in industry, scenarios, and data, China is expected to achieve a rule overtaking on the curve in segmented tracks. At the same time, it needs to face the long-term competitive pressure from the underlying standards and safety compliance systems of the US, Europe, and Japan. Standard research, international games, and industrial landing will present four clear development trends.
The standard system evolves from fragmented gap-filling to layered full-chain coverage, with basic general, safety and ethics, and scenario application standards accelerating simultaneously
Currently, most global embodied AI standards focus on single performance and single safety indicators, making it difficult to adapt to the whole-process control needs of the complete machine's full life cycle R&D, production, use, and recycling. In the future, standardization institutions in various countries will simultaneously accelerate the supply of three types of standards. First, underlying general standards accelerate unification. General specifications such as hardware interfaces, multimodal data formats, virtual-real simulation interaction, and operating system compatibility will become the focus of layout in various countries, solving the problem of cross-brand and cross-platform adaptation of components and reducing repeated R&D costs across the industry. The US will continue to consolidate its dominant position in the underlying standards of chips and computing power interfaces. China, relying on massive industrial scenarios, will focus on creating characteristic basic standards such as datasets, simulation training grounds, and evaluation benchmarks. Second, safety and ethical standards become rigid demands. As humanoid robots enter homes, factories, and elderly care scenarios, relevant specifications for dynamic human-machine safety, accident liability division, AI ethical constraints, and data privacy protection will gradually shift from recommended standards to mandatory rules for market access. The EU will continue to rely on the Artificial Intelligence Act to export a global stringent compliance template. Third, specialized standards for segmented scenarios are issued in batches. Differentiated tracks such as intelligent manufacturing, elderly care companionship, special operations, and commercial services will form exclusive performance and task acceptance standards. Japan and South Korea will continue to delve into segmented standards for service robots. Relying on a complete manufacturing system, China's industrial embodied equipment scenario standards will form unique advantages. Overall, future standard construction is no longer a single indicator supplement, but a systematic layout of the entire industrial chain and full life cycle. The completeness of the standard system will become a core indicator to measure the industrial rule competitiveness of various countries.
International standard competition shifts from single-point proposal breakthroughs to systematic rule output, with underlying core tracks becoming the main battlefield for the game
At this stage, China has only achieved a single-point breakthrough in international standards in the field of humanoid robot datasets. The US, Europe, and Japan have long formed a systematic international standard matrix in the fields of safety, hardware, and testing. The gap still objectively exists, and the subsequent international game pattern will change significantly. On the one hand, the competition for core seats in international standardization organizations tends to be white-hot. The convener and editor seats of the working groups in the two core institutions, ISO/TC 299 and ISO/IEC JTC 1/SC 42, will become resources that all countries must compete for. The US will continue to consolidate its agenda-setting power relying on the secretariat advantage. China will continue to increase the cultivation of international standardization talents and continuously expand new working groups in advantageous tracks such as datasets and multimodal interaction. On the other hand, the standard output model upgrades from a single text proposal to a complete set of solutions output including standards, testing benchmarks, and datasets. Simply submitting text standards has limited persuasiveness. Supporting public testing platforms, open-source benchmark datasets, and complete verification cases will become key chips to support the passage of international standards. In addition, the differentiation of rule camps is gradually emerging. The US and Europe are building a standard system biased towards technological control and the free flow of cross-border data. China, based on data security and independent and controllable industry, is forming a differentiated rule route. Japan and South Korea are in the middle to adapt to the two camps. Benchmarking and mutual recognition between different systems will become a long-term issue. It is difficult for global standards to be completely unified in a short time, and the pattern of rule segmentation in segmented tracks may exist for a long time.
Standard research is deeply bound with industrial mass production and testing verification, and the closed-loop model of industry leads, standards synchronize, testing provides the bottom line is fully popularized
In the past, the industry generally had the lag problem of industry advances first, standards follow later. Products have been mass-produced for years without unified evaluation and acceptance standards. This development model will be completely reversed in the future. First, standard research is embedded in the technological R&D cycle in advance. Policies in various countries will promote major embodied AI research projects to be simultaneously equipped with standard compilation tasks. Enterprises will participate in indicator argumentation simultaneously during the new product R&D stage, avoiding closed self-research and interface fragmentation from the source. Second, national-level public testing and verification platforms become standard for standard landing. Unified evaluation benchmarks, open-source standard datasets, and virtual-real combined simulation training grounds will accelerate layout, solving the pain points that current enterprise evaluation results cannot be mutually recognized and product performance is difficult to compare horizontally. South Korea has taken the lead in building a full life cycle safety certification center for humanoid robots. China will also accelerate the layout of national-level testing bases covering multiple scenarios to support the landing and application of standards with unified hardware, algorithm, and safety evaluation systems. Third, standards and certification systems are deeply integrated. Standard clauses are directly transformed into market access testing items. The cross-border mutual recognition of testing and certification results becomes a key supporting factor for industrial going global. Enterprises no longer need to undergo multiple rounds of repeated rectification and testing when going global. The integrated service chain of standards, testing, and certification will become the core supporting policy for various countries to support the globalization of local enterprises.
The two-way transformation of domestic standard internationalization and international standard localization accelerates, and standards become the core competitiveness for the globalization of the industry
Along with the batch going global of China's humanoid robots and industrial embodied equipment, the shortcomings of disconnection between domestic and foreign standards and high compliance barriers are becoming increasingly prominent. Two-way benchmarking and transformation will be the key direction of future standardization work. Domestically, it will continuously and dynamically track mainstream international standards such as UL, ISO, and IEC, carry out adaptive revisions combining China's local needs for data security and human-machine protection, balance international integration and the bottom line of industrial safety, and avoid the risk of industrial chain adaptation caused by copying foreign standards. Externally, relying on China's unique large-scale manufacturing and elderly care service scenario practice experience, mature engineering indicators and massive real-scenario data specifications will be transformed into international general clauses, turning scenario scale advantages into rule advantages. At the same time, the mutual recognition cooperation of domestic and foreign standards continues to advance. China will carry out mutual recognition cooperation on safety, electromagnetic compatibility, and functional performance testing results with multiple countries to break down overseas market access barriers. In the long run, the going global model that solely relies on low-price product competition will gradually fail. Enterprises with independent standard systems and complete testing and certification service capabilities will gain greater initiative in the global market. Standardization capability will become a new core dimension to measure the international competitiveness of enterprises.
Authors | Embodied AI Research Group, Research Institute of Science, Technology and Standards, CCID (China Center for Information Industry Development): Liang Kun, Zhang Peng, Fu Qingfen
Editor | Wu Lilin
Art Editor | Ma Liya Supervisor | Lian Xiaodong