On 22 October 2026, People's Daily Online hosted a conference in Beijing under the banner "Cognition · Embodiment · Future" — formally an exchange meeting on embodied cognition and AI evolution that doubled as a results release. The item with the longest reach was the founding of an alliance for embodied intelligence applications and safety ethics, with 18 initial member units represented on stage to start it.
Two national key laboratories guided the event: the National Key Laboratory of Communication Content Cognition and the National Key Laboratory of Autonomous Intelligent Unmanned Systems. The programme was split into two halves. The first, "Four-Dimensional Evolution", ran through four themes — the cognition base, model evolution, the data engine and deployment in real settings. The second, "Laying the Foundation", is where the institutional pieces were announced.
What the four dimensions covered
Fan Zhengwei, Party secretary and president of People's Daily Online, opened the cognition-base segment with a talk on how a "mainstream value corpus" could support embodied intelligence. In the model segment, He Bin of Tongji University described agents moving into the physical world, arguing that multi-agent platforms and industry alliances are how agent technology shifts from isolated breakthroughs to system-level coordination. That is his formulation of a direction, not a demonstrated result.
Yang Haibo, co-founder and president of Guanglun Zhineng, made the data case. Embodied AI, he said, generates substantially larger data demands than large language models or autonomous driving, and robot learning needs extensive trial and error inside simulation. His company's proposal is an evaluation pyramid — simulation-based assessment that is large-scale, quantifiable, comparable and traceable — tied to a data, evaluation and deployment chain. That is a company's stated approach rather than a published benchmark.
JD Retail's embodied intelligence chief Zheng Xiaodan described an ambition to be "an operator of the physical world": robots being deployed in retail, logistics and health settings, after-sales repair arrangements with several robot makers, and a wish to work with People's Daily Online on value alignment, safety and deployment.
In the second half, Li Jun, deputy director of the communication-content laboratory, and He Bin, deputy director of the autonomous-systems laboratory, jointly started a collaborative innovation cooperation.
The alliance and its four platforms
The alliance was assembled by People's Daily Online drawing on the communication-content laboratory, together with universities, research institutes and companies along the industrial chain. It is described as an extension of an existing "mainstream value corpus ecosystem alliance" and will be folded into that ecosystem's construction.
Its stated remit is to build four platforms: a practice platform for safety governance, a leading platform for standards and rule-making, a hub platform for industrial coordination, and a support platform for talent development.
The founding list mixes public research infrastructure, a ministry-affiliated institute, a Beijing industry association, a district-level technology group, and robotics and AI companies. The 18 initial members are the National Key Laboratory of Communication Content Cognition; the National Key Laboratory of Autonomous Intelligent Unmanned Systems; the People's Daily Online AI Research Institute; the AI Center of the Fifth Electronics Research Institute of the Ministry of Industry and Information Technology; the Beijing Artificial Intelligence Association; Beikong Zhike Digital Intelligence Research Institute; Chaoyang Technology Group; Horizon Robotics; Digua Robotics; Guanglun Zhineng; JD; JAKA Robotics; Lingxin Qiaoshou; Leju Robotics; ModelBest; Shangpin Home Collection; Yingli Energy; and Zhongguancun Robot Industry Innovation Development Co.
That roster runs from perception and computing hardware through robot arms, humanoids, dexterous hands and models to simulation tooling, retail and logistics, alongside companies from home furnishing and energy. The breadth matters for the platform that is supposed to set standards: rule-making in embodied AI touches hardware safety, data provenance and deployment practice at once.
The training system behind the announcement
The People's Daily Online AI Research Institute released an AI communication-cognition and safety training system together with a talent programme called "Zhi Xing", roughly knowing and doing. Zhao Xiumei, who heads the institute, said the curriculum is designed around "technical capability plus safety governance", with eight core course modules running from basic understanding through technical architecture and safety ethics to practical deployment. Zhi Xing is structured as a learn, practise, evaluate and apply model, with three product lines aimed at the general public, at enterprises and at key industries.
The announcement named no module titles, teaching hours, assessment criteria or partner institutions for delivery, so the eight-module structure is currently an outline rather than a syllabus that outside organisations can plan around.
Signings, and what to watch
Three agreements were signed. The AI Research Institute signed an embodied-AI industry ecosystem agreement with JD Retail and an ecosystem partnership with Zhiyuan Robotics. Separately, the autonomous-systems laboratory, Chaoyang Technology Group and Renmin Chuangtou agreed to co-build a "Chaoyang intelligent native AI evolution innovation platform".
The alliance announcement did not state a budget, membership obligations, a timetable for standards work or who staffs the four platforms. Those are the practical questions that decide whether a four-part mandate becomes a working body or stays a launch photograph. For now what exists is a named membership, a four-platform remit and a curriculum outline. The safety-governance and standards platforms are the two to track, because those are where deliverables would eventually be visible — a draft rule, a testing arrangement, a published evaluation method. Until something of that kind appears, the most defensible description of the 22 October event is that a convening institution has put its name and its laboratory behind an industry coalition, and invited the hardware, model and application sides of Chinese embodied AI to sit inside one framework.