The National Data Administration has signaled that it will promote the development of data standards for embodied intelligence at an appropriate time, following a symposium hosted by Director Liu Liehong on data-enabled embodied intelligence development. The meeting brought together representatives from research institutes, technology companies, robotics ventures, and a humanoid robot data training center to discuss how data can better support embodied intelligence as the field grows rapidly and becomes increasingly data-driven.
According to information disclosed on September 13, the symposium focused on the theme of data empowering embodied intelligence development. Participants were invited to offer opinions and suggestions on the opportunities and challenges facing embodied intelligence, particularly as demand continues to rise for high-quality, diverse, and large-scale data. The National Data Administration said it will remain problem-oriented, follow industry trends, strengthen scientific layout and development guidance, promote embodied intelligence data standards when appropriate, guide local data systems to carry out related work in an orderly manner, support healthy industry development, and encourage enterprises to increase data investment so that data can better empower embodied intelligence.
-
Symposium Places Embodied Intelligence Data at the Center of Policy Attention
The symposium hosted by Liu Liehong placed embodied intelligence data at the center of policy attention. The National Data Administration’s decision to convene representatives from multiple parts of the embodied intelligence ecosystem indicates that data is no longer viewed as a secondary technical input but as a foundational resource for the next stage of embodied intelligence development. The meeting’s theme, data empowering embodied intelligence development, reflects a broad recognition that embodied intelligence systems depend on diverse and continuous streams of information drawn from physical environments, human demonstrations, sensor systems, simulation platforms, and real-world operations.
Embodied intelligence is a field in which artificial intelligence systems are expected to perceive, reason, act, and adapt within physical or simulated environments. Unlike purely digital artificial intelligence applications, embodied intelligence must contend with the complexity of the physical world. That complexity creates an unusually demanding data problem. Embodied intelligence systems need data that captures not only objects and scenes but also motion, force, touch, spatial relationships, timing, cause and effect, and the consequences of actions. As a result, the demand for high-quality, diverse, and large-scale data for embodied intelligence continues to grow.
The National Data Administration’s involvement suggests that embodied intelligence data policy may increasingly intersect with national data governance, data infrastructure, standards, and enterprise investment. The symposium did not simply address a narrow technical question. It addressed how data can be organized, shared, governed, and used to support an emerging industry. That framing is significant for embodied intelligence because the field spans robotics, artificial intelligence, manufacturing, logistics, healthcare, services, and other domains. Each domain may generate different forms of embodied intelligence data, and each may require different standards for collection, annotation, storage, transfer, and evaluation.
By inviting opinions and suggestions from research institutes, technology companies, robotics developers, and a humanoid robot data training center, the National Data Administration gathered perspectives from both the supply side and the demand side of the embodied intelligence data ecosystem. This collaborative approach may help ensure that future standards for embodied intelligence are practical, interoperable, and aligned with real industry needs. It also signals that the administration intends to proceed carefully, using industry feedback and scientific analysis before advancing formal standards for embodied intelligence data.
-
Participants Represent a Broad Embodied Intelligence Data Ecosystem
The symposium included representatives from the Institute of Automation, Chinese Academy of Sciences; Beijing Institute for General Artificial Intelligence; Lightwheel; JD Group; Shendu Jizhi; Shijingshan Humanoid Robot Data Training Center; Galaxea; Zibianliang; and Beijing Humanoid Robot Innovation Center. The range of participants reflects the multi-layered nature of embodied intelligence. The field is not built by a single type of organization. It requires foundational research, general artificial intelligence capabilities, simulation and synthetic data, commercial deployment experience, humanoid robot training infrastructure, robotics innovation, and new venture activity.
The presence of research institutes indicates that embodied intelligence remains closely tied to long-term scientific inquiry. The presence of technology companies and enterprise groups indicates that embodied intelligence is also moving toward commercial applications. The participation of a humanoid robot data training center highlights the importance of specialized facilities for generating and refining data used by embodied intelligence systems. The involvement of robotics ventures and an innovation center further shows that embodied intelligence development is being pursued through a combination of startup energy, industrial resources, and institutional support.
For data policy, this diversity matters. If standards for embodied intelligence are designed only around one type of data, one type of robot, or one type of application, they may fail to support the wider ecosystem. If standards are too abstract, they may not be usable by engineers building embodied intelligence systems. The symposium therefore created an opportunity to examine embodied intelligence data needs across multiple organizational perspectives. The National Data Administration can use such input to consider how standards might support interoperability without suppressing experimentation.
The participants were asked to publish opinions and suggestions around data enabling embodied intelligence development. Although the meeting’s detailed discussions were not disclosed, the list of participants itself provides a map of the emerging embodied intelligence data landscape. It includes organizations focused on automation, general artificial intelligence, data generation, enterprise platforms, humanoid robot training, robotics systems, and innovation coordination. This map can help policymakers understand that embodied intelligence data is not a single commodity but a broad category that includes simulation data, real-world interaction data, multimodal sensor data, human demonstration data, evaluation data, and operational data.
- Research institutes can contribute to foundational theories, benchmarks, and long-term embodied intelligence data methodologies.
- Technology companies can contribute to scalable data pipelines, annotation systems, and software infrastructure for embodied intelligence.
- Robotics ventures can contribute to real-world deployment feedback and hardware-specific embodied intelligence data requirements.
- Humanoid robot data training centers can contribute to specialized data collection, training, and validation for embodied intelligence.
- Innovation centers can contribute to coordination, pilot projects, and ecosystem building for embodied intelligence.
-
Data Demand Grows as Embodied Intelligence Becomes More Data-Driven
The National Data Administration noted that the embodied intelligence industry is developing vigorously and shows clear data-driven characteristics. Demand for high-quality, diverse, and large-scale data continues to grow. This observation captures a central dynamic in embodied intelligence: progress depends not only on algorithms and hardware but also on the availability of data that can teach systems how to operate in complex environments. As embodied intelligence systems become more capable, they may require more varied data, more precise labels, more realistic simulations, and more robust evaluation datasets.
High-quality data for embodied intelligence is difficult to produce at scale. Physical interaction data often requires robots, sensors, controlled environments, human supervision, and safety procedures. Real-world data may be expensive to collect and difficult to reproduce. Simulation data can be scaled more easily, but it may not fully capture the unpredictability of the physical world. Human demonstration data can provide valuable examples, but it must be aligned with machine-actionable formats. Multimodal data can enrich embodied intelligence systems, but it also increases the complexity of storage, synchronization, annotation, and governance.
Diversity is equally important. An embodied intelligence system trained only in a narrow environment may fail when introduced to new objects, surfaces, lighting conditions, human behaviors, or task requirements. Diverse data can help embodied intelligence systems generalize across conditions. Large-scale data can help them learn robust representations and policies. However, scale alone is not enough. Poorly organized or inconsistent data can create noise, bias, and safety risks. For embodied intelligence, the combination of quality, diversity, and scale is therefore a policy-relevant challenge, not merely an engineering preference.
Standards can help address this challenge by creating common expectations for how embodied intelligence data is described, structured, validated, and shared. Without standards, different organizations may develop incompatible datasets, metadata schemas, evaluation methods, and security practices. That fragmentation can slow innovation, duplicate effort, and limit the reuse of valuable embodied intelligence data. With appropriate standards, organizations can collaborate more effectively, compare systems more reliably, and build on shared foundations. The National Data Administration’s willingness to promote embodied intelligence data standards at an appropriate time reflects an understanding of this need.
-
National Data Administration Outlines a Measured Standards Path for Embodied Intelligence
The National Data Administration said it will adhere to a problem-oriented approach, grasp industry development trends, strengthen scientific layout and development guidance, promote embodied intelligence data standards in due course, guide local data systems to carry out related work in an orderly manner, promote healthy industry development, actively support enterprises in increasing data investment, and better empower embodied intelligence development. Each element of this statement has significance for the future of embodied intelligence data governance.
A problem-oriented approach suggests that standards work will focus on concrete obstacles rather than abstract ambitions. For embodied intelligence, such obstacles may include inconsistent data formats, unclear quality metrics, limited interoperability, weak metadata practices, difficulty in comparing models, and uncertainty around data rights and security. By starting from problems, policymakers can avoid creating standards that are disconnected from practical embodied intelligence development. They can also prioritize areas where coordination would deliver the greatest benefit.
Grasping industry development trends is important because embodied intelligence is evolving quickly. New robot platforms, sensor suites, simulation engines, foundation models, and training methods appear frequently. Standards that are too rigid may become obsolete. Standards that are too vague may fail to guide implementation. A trend-aware approach can help policymakers design flexible frameworks that support current embodied intelligence needs while remaining open to future changes. It can also help align data standards with advances in artificial intelligence, robotics, computing infrastructure, and data security.
Strengthening scientific layout and development guidance implies a coordinated approach. Embodied intelligence data is not limited to one sector or one region. It involves research institutions, enterprises, data centers, training facilities, cloud platforms, and local governments. Scientific layout can help connect these pieces. Development guidance can help stakeholders understand priorities, avoid duplication, and invest in shared infrastructure. For embodied intelligence, such guidance may include recommendations on data collection, annotation, quality assurance, sharing mechanisms, and evaluation practices.
The phrase “at an appropriate time” indicates a measured approach. The National Data Administration is not suggesting that embodied intelligence data standards should be rushed. Instead, it appears to favor preparation, consultation, and alignment with industry maturity. This is important because premature standards can constrain innovation, while delayed standards can allow fragmentation to become entrenched. The administration’s stated intention is to promote embodied intelligence data standards when conditions are suitable, which suggests a balance between urgency and prudence.
Guidance for local data systems is another key element. Local data systems may include regional data exchanges, public data platforms, pilot zones, computing centers, and industry clusters. If local efforts proceed without coordination, they may create incompatible practices for embodied intelligence data. If they are guided in an orderly manner, they can contribute to a more coherent national ecosystem. Local experimentation can also inform national standards for embodied intelligence by revealing what works in different industrial and regional contexts.
-
Local Coordination and Enterprise Investment in Embodied Intelligence
The National Data Administration’s commitment to guide local data systems and support enterprise data investment has direct implications for embodied intelligence. Local governments and data agencies may play a role in building data infrastructure, supporting pilot projects, and facilitating collaboration between research institutes and companies. Enterprises, meanwhile, may need to invest in data collection, cleaning, labeling, simulation, storage, security, and governance. For embodied intelligence, these investments can be substantial because data is often generated through physical processes that require equipment, facilities, and expert oversight.
Enterprise data investment is not only a matter of purchasing storage or computing power. It also includes developing data strategies, building data teams, creating data pipelines, establishing quality controls, and participating in standards development. Companies that invest in embodied intelligence data may gain advantages in model performance, safety, reliability, and deployment speed. However, they may also face uncertainty about return on investment, especially when data requirements are changing rapidly. Policy support and clear standards can reduce some of that uncertainty by making embodied intelligence data more reusable, comparable, and trustworthy.
Local coordination can help reduce duplication and improve resource efficiency. For example, different regions may develop specialized strengths in manufacturing, logistics, healthcare, agriculture, or services. These strengths can shape the types of embodied intelligence data that are most relevant locally. A coordinated approach can help local data systems share best practices, align metadata, and connect with national standards. It can also help smaller companies access data infrastructure and expertise that would be difficult to build independently.
At the same time, local coordination must respect diversity. Embodied intelligence applications vary widely. A humanoid robot in a factory may require different data from a mobile robot in a warehouse or a service robot in a public space. Standards for embodied intelligence data should therefore allow for domain-specific extensions while maintaining common foundations. The National Data Administration’s emphasis on orderly work and healthy development suggests an awareness that coordination should enable, not constrain, embodied intelligence innovation.
-
Embodied Intelligence Data Standards Could Support Interoperability, Safety, and Reuse
Standards for embodied intelligence data could have far-reaching effects. They may support interoperability by allowing datasets, models, and systems from different developers to work together more easily. They may improve safety by defining expectations for data quality, traceability, and validation. They may encourage reuse by making embodied intelligence datasets easier to discover, understand, and combine. They may also strengthen evaluation by providing common benchmarks and metrics for embodied intelligence performance.
Interoperability is especially important for embodied intelligence because the field combines hardware, software, sensors, actuators, and artificial intelligence models. A robot platform from one developer may need to use data produced by another. A simulation environment may need to exchange information with a real-world training system. A foundation model for embodied intelligence may need to consume data from multiple sources. Without common standards, each connection may require custom engineering. With standards, integration can become more scalable and less costly.
Safety and trust are also central. Embodied intelligence systems may operate in environments where errors can cause physical harm or financial loss. Data standards can help by requiring documentation of data sources, collection conditions, limitations, and intended uses. They can support auditability and traceability. They can help developers understand whether a dataset is suitable for a particular embodied intelligence application. They can also help regulators and users assess the risks associated with embodied intelligence systems.
Reuse is another potential benefit. Many organizations collect valuable embodied intelligence data but may not have the capacity to use it fully. Standards can make it easier to share data across projects, institutions, and sectors. They can reduce the need to recreate similar datasets. They can help small companies and research groups access high-quality embodied intelligence data without building every collection pipeline from scratch. In this way, standards can democratize participation in embodied intelligence development while maintaining appropriate controls.
-
Industry Voices Emphasize Quality, Diversity, and Scale for Embodied Intelligence
The symposium’s focus on data empowerment indicates that participants were asked to consider how quality, diversity, and scale affect embodied intelligence. These three dimensions are closely connected. Quality ensures that data is accurate, relevant, well-labeled, and suitable for training or evaluation. Diversity ensures that embodied intelligence systems encounter a wide range of conditions, tasks, objects, and interactions. Scale ensures that systems have enough examples to learn robust patterns. Together, they shape the capabilities and limitations of embodied intelligence.
For embodied intelligence, quality is not a simple property. A dataset may be high quality for one task but unsuitable for another. For example, data collected in a controlled laboratory may be precise but lack the messiness of real-world environments. Data collected in a busy public space may be diverse but noisy. Data generated in simulation may be scalable but may not perfectly match physical dynamics. Standards for embodied intelligence data may need to describe context, provenance, and intended use so that users can judge quality appropriately.
Diversity in embodied intelligence data can include variation in objects, materials, lighting, weather, terrain, human behavior, robot morphology, sensor configuration, task goals, and failure modes. Capturing such diversity is difficult but essential. An embodied intelligence system that has never encountered a slippery surface, a crowded room, or an unexpected obstacle may fail in practice. Data standards can help by defining categories and metadata that make diversity explicit. They can also help organizations identify gaps in their embodied intelligence datasets.
Scale remains a major challenge. Large-scale embodied intelligence data may require many hours of operation, many environments, and many tasks. It may require simulated environments that can generate vast amounts of experience. It may require human demonstrations, teleoperation, or autonomous exploration. The National Data Administration’s recognition that demand for large-scale data continues to grow suggests that scale will remain a policy concern. Standards can help scale data efforts by making collection, annotation, and sharing more systematic.
-
Embodied Intelligence Data Governance Requires Balance and Experimentation
Data governance for embodied intelligence must balance multiple goals. It must support innovation while protecting security and privacy. It must encourage sharing while respecting intellectual property and commercial interests. It must promote common standards while allowing domain-specific flexibility. It must enable large-scale data use while ensuring that data is accurate and lawful. These balances are not easy to achieve, especially in a field as dynamic as embodied intelligence.
The National Data Administration’s problem-oriented and scientific approach suggests that governance will be informed by experimentation and evidence. Pilot projects, local trials, and industry consultations can help identify what works. Standards can be introduced in stages, starting with foundational areas such as metadata, terminology, and quality documentation, then expanding to more complex areas such as cross-platform data exchange and evaluation. This staged approach can reduce risk and allow embodied intelligence stakeholders to adapt.
Experimentation is particularly important because embodied intelligence data is still evolving. There is no single accepted taxonomy for embodied intelligence data. Different communities may use different terms for similar concepts. Different robots may produce different data structures. Different models may require different input formats. Standards development can help harmonize these differences, but it must remain open to new approaches. A balance between coordination and flexibility will be essential for healthy embodied intelligence development.
The goal of healthy industry development, as stated by the National Data Administration, reflects this balance. Healthy development implies growth that is sustainable, safe, and beneficial. For embodied intelligence, that means avoiding both fragmentation and overregulation. It means supporting enterprise investment while protecting public interests. It means encouraging data sharing while maintaining trust. The symposium can be seen as part of a broader effort to define what healthy embodied intelligence data governance looks like.
-
Table: Participants and Embodied Intelligence Data Ecosystem Roles
Participant Category Relevance to Embodied Intelligence Data Institute of Automation, Chinese Academy of Sciences Research institute Foundational research relevant to automation and embodied intelligence data Beijing Institute for General Artificial Intelligence Research institute General artificial intelligence research relevant to embodied intelligence Lightwheel Technology company Data and simulation capabilities relevant to embodied intelligence JD Group Enterprise group Commercial platform and technology ecosystem relevant to embodied intelligence data applications Shendu Jizhi Technology company Embodied intelligence and data-related development Shijingshan Humanoid Robot Data Training Center Data training center Humanoid robot data training relevant to embodied intelligence Galaxea Robotics company Embodied intelligence robotics development Zibianliang Technology company Embodied intelligence and data-related development Beijing Humanoid Robot Innovation Center Innovation center Humanoid robot innovation relevant to embodied intelligence -
Table: Policy Directions for Embodied Intelligence Data Standards
Policy Direction Meaning for Embodied Intelligence Problem-oriented approach Focus on concrete obstacles in embodied intelligence data development Industry trend monitoring Align policy with the evolving embodied intelligence landscape Scientific layout Organize data infrastructure and standards work systematically for embodied intelligence Development guidance Provide direction for stakeholders working on embodied intelligence data Timely standard building Promote embodied intelligence data standards when conditions are appropriate Local coordination Guide local data systems to carry out related work in an orderly manner Healthy industry development Support sustainable and well-regulated growth of embodied intelligence Enterprise data investment Encourage companies to increase investment in embodied intelligence data Data empowerment Use data to better enable embodied intelligence innovation -
Key Takeaways for Embodied Intelligence Stakeholders
The National Data Administration’s symposium and subsequent statement provide several important signals for stakeholders in embodied intelligence. These signals are relevant to researchers, technology companies, robotics developers, data service providers, local data agencies, and investors. Although the administration did not announce a complete regulatory framework, its emphasis on standards, coordination, and enterprise investment suggests that embodied intelligence data will receive sustained policy attention.
- The National Data Administration is actively examining how data can empower embodied intelligence development.
- Embodied intelligence is recognized as a data-driven industry with growing demand for high-quality, diverse, and large-scale data.
- The administration plans to promote embodied intelligence data standards at an appropriate time.
- Local data systems will be guided to carry out related work in an orderly manner.
- Enterprises are encouraged to increase data investment to support embodied intelligence.
- Healthy industry development is a stated goal for embodied intelligence data policy.
- The symposium brought together research institutes, technology companies, robotics ventures, a data training center, and an innovation center.
- Standards for embodied intelligence data may address interoperability, quality, diversity, safety, reuse, and evaluation.
- Policy implementation is likely to be problem-oriented, trend-aware, and scientifically laid out.
- Embodied intelligence data governance will need to balance innovation, security, privacy, and commercial interests.
For researchers, the policy direction may create opportunities to contribute to benchmarks, datasets, and evaluation methods for embodied intelligence. For companies, it may create incentives to invest in data infrastructure and participate in standards development. For local data agencies, it may require coordination with national priorities while supporting regional strengths. For the broader embodied intelligence ecosystem, it may help reduce fragmentation and improve trust in data-driven development.
-
Outlook: Embodied Intelligence Data Standards as an Evolving Foundation
The outlook for embodied intelligence data standards is likely to be evolutionary rather than instantaneous. The National Data Administration’s statement does not suggest that a complete set of standards will appear immediately. Instead, it points to a process of consultation, scientific layout, local coordination, and timely action. That process will need to adapt as embodied intelligence technology advances. New robot designs, learning methods, sensors, and applications will continue to change data requirements.
Embodied intelligence data standards may begin with foundational elements such as terminology, metadata, documentation, and quality reporting. These elements can help different organizations describe embodied intelligence data in compatible ways. Over time, standards may expand to include data exchange protocols, annotation guidelines, safety and privacy requirements, benchmark suites, and evaluation procedures. Each stage can build on previous work, allowing the ecosystem to mature without being frozen in place.
International alignment may also become relevant as embodied intelligence develops across borders. Although the National Data Administration’s focus is domestic, embodied intelligence research and commerce are global. Standards that are compatible with international practices may help domestic organizations participate in global embodied intelligence ecosystems. At the same time, local priorities around security, privacy, and industrial development will shape how standards are implemented.
The administration’s support for enterprise data investment suggests that private-sector innovation will remain central. Standards can provide a common language, but enterprises will continue to build proprietary datasets, models, and applications. The challenge will be to encourage interoperability and sharing where beneficial while preserving incentives for investment. A healthy embodied intelligence data ecosystem may therefore combine public standards, private data assets, shared benchmarks, and local pilot programs.
-
Conclusion: Embodied Intelligence Enters a Data-Governance Phase
The National Data Administration’s symposium on data-enabled embodied intelligence development marks a notable step in the policy attention given to embodied intelligence. By bringing together research institutes, technology companies, robotics ventures, a humanoid robot data training center, and an innovation center, the administration has acknowledged that embodied intelligence depends on a broad and coordinated data ecosystem. The meeting’s focus on high-quality, diverse, and large-scale data reflects the practical realities of building embodied intelligence systems that can operate reliably in complex environments.
The administration’s statement that it will promote embodied intelligence data standards at an appropriate time is carefully phrased. It signals both direction and caution. The direction is toward standards, coordination, and data empowerment for embodied intelligence. The caution is that standards should be developed when conditions are suitable and with attention to industry trends, scientific layout, and healthy development. This balanced approach may help avoid the risks of premature regulation while addressing the risks of fragmentation.
Guidance for local data systems and support for enterprise data investment are also significant. They suggest that embodied intelligence data policy will not be limited to national-level standard-setting. It will also involve local implementation, pilot projects, infrastructure, and private-sector participation. For embodied intelligence to develop healthily, data must flow across organizations and applications in ways that are secure, trustworthy, and efficient. That requires both standards and investment.
As embodied intelligence continues to advance, data will remain one of its most important foundations. The National Data Administration’s symposium and its stated commitment to timely standards for embodied intelligence data indicate that this foundation is now receiving serious policy consideration. The next phase will likely involve further consultation, technical work, local coordination, and enterprise engagement. If managed well, these efforts can help embodied intelligence realize its potential while supporting a healthy, innovative, and trustworthy data ecosystem.