context: Embodied intelligence is one of six top-priority future industries for the 15th 5-year plan period. Amidst progress in pushing global standards and a booming robotics sector, Beijing is now using provincial governments and central SOEs to turn industrial demand into training infrastructure, addressing two key barriers to embodied intelligence: scarce real-world operating data and unproven commercial viability. Scenario quotas and requirements to replicate validated solutions make this a state-led scale-up mechanism rather than another demonstration program.
MIIT (Ministry of Industry and Information Technology) and SASAC (State-owned Assets Supervision and Administration Commission) launched the 2026 humanoid robot and embodied intelligence real-world training action, seeking routine deployment in representative industrial, service and specialised settings by year-end.
The action targets
- validation and routine deployment of humanoid robots and embodied AI systems in production environments, moving them from pilot projects into regular operations
- more than 100 application scenarios with clear commercial or operational value across industry, services and specialised tasks
- ten-thousand-unit-scale deployment capacity
Beijing, Tianjin, Shanghai, Jiangsu, Zhejiang, Shandong, Hubei, Hunan, Guangdong and Sichuan must each select at least 20 real-world training spaces covering at least two of the three fields; participating central SOEs must each select at least ten. Scenarios include manufacturing, inspection, maintenance, warehousing, retail, healthcare, workplace safety and emergency response.
Each scenario should form an innovation application consortium led by user units and robot makers or service providers, together with model, component and research organisations. Consortia will
- open and minimally adapt real workplaces
- set measurable deployment goals and technical requirements
- supply workflow data and environmental information
- develop replicable skill packages, models, motion-control algorithms and high-fidelity datasets
- agree on task boundaries, intellectual property ownership and benefit sharing
User units or third-party bodies will test real-world success rates, efficiency gains, safety, reliability and economic viability. Provincial governments and central SOEs should then deploy validated systems across similar settings, applying the principle ‘validate one, deploy a batch, drive a wider field’.
The action also calls for
- cloud-edge-device coordination and offline autonomous computing
- stronger durability, thermal management, collision detection, force limits, emergency braking and black-box functions
- lifecycle identity management and implementation of emerging standards
- risk warnings, upgrade windows and exit channels for fast-changing applications
- ‘humanoid robot as a service’ through pay-per-use and operating leases
- local trials of robot insurance and other support measures
Work plans are due by 30 June and final reports by 30 November. MIIT and SASAC will track implementation, promote leading cases and favour strong performers in future policies, standards and projects.