On October 11, the Beijing Key Laboratory of General World Models was officially unveiled at the 2026 World Model Summit Forum. The laboratory is hosted by Beijing Jijia Shijie Technology Co., Ltd. According to the on-site introduction, the laboratory was approved in May this year and held its unveiling ceremony in October, with the two time points corresponding respectively to approval and unveiling.
At the unveiling ceremony, Lu Jiwen, tenured professor and deputy head of the Department of Automation at Tsinghua University, was appointed director of the laboratory's academic committee, and Zhu Zheng, co-founder and chief scientist of Jijia Shijie, was appointed director of the laboratory. The laboratory plans to conduct research around the fundamental theories, key technologies and application systems of general world models, and to advance technological exploration for scenarios including embodied intelligence, autonomous driving and content creation.
World models focus on how the environment changes with actions. Lu Jiwen said at the forum that such models compute and predict how things evolve under specific actions, involving cross-modal, cross-spatiotemporal and cross-scenario information. For physical AI, understanding the current state, predicting changes and judging the consequences of actions is an important research direction connecting environmental perception with actual action.
According to the announced construction arrangements, Jijia Shijie will provide the laboratory with support in talent, funding, computing power, space and data, and will promote the opening and sharing of resources with universities and research institutes. These are follow-up construction plans, and their scope of implementation and actual results still require continuous observation.
The forum discussion also touched on the practical constraints on the development of world models. According to on-site reporting by Yicai, participants focused on the scale and information quality of action data, as well as the computing resources required for training and inference. The related discussion suggests that expanding investment in data and computing power needs to be combined with research goals, data representation and validation methods.
From the perspective of research and application, generating visually plausible scenes and supporting reliable action are capability requirements at different levels. The value of a model in a physical environment needs to be tested through specific tasks and actual results.
This unveiling marks the further implementation of a research platform oriented toward general world models. Progress worth watching going forward includes research results, the open sharing mechanism, and the model's verification performance in real scenarios; the establishment of the laboratory itself does not mean that the related technical challenges have been solved.
Link: https://www.hqkjw.cn/index.php/zhinenghuanqiu/31053.html