Business+IT Publishes Experts’ Discussion of Physical AI “World Models”

Business+IT published an article on Nov. 13, 2026, covering a discussion of “world models” among experts from NVIDIA, the University of Tokyo and other organizations. In the excerpt, Mabuchi says the scope of physical AI is “somewhat vague” and includes multiple layers, such as AI, humanoid robots, simulation and models that operate robots.

The excerpt conveys Mabuchi’s view that physical AI includes multiple layers, such as AI, robots, simulation and control models. Assessing discussions of world models requires distinguishing which technologies or applications are being addressed, but the excerpt does not provide specific proposals or conclusions.

The excerpt presents physical AI not as a single product or model but as a concept that encompasses multiple technology layers. Mabuchi’s explanation includes AI and robots, as well as simulation and models that operate robots. However, the excerpt does not say what technologies or implementations the “world models” in the headline refer to, or what conclusions the experts reached. Therefore, the material alone does not provide enough information to assess Japan’s specific competitive advantages or technological path to success. The excerpt also cites 51 competitions, including events focused on social implementation in settings such as factories, logistics and hotels. This suggests that discussions of physical AI should consider not only robot demonstrations but also challenges tailored to specific applications. However, the description provided does not establish the competition results, evaluation criteria or relationship to continued operation in real-world environments. The distinction between demonstrations at exhibitions or competitions and verification of reproducibility or real-world operation is also relevant to leaderobot.com’s report on a humanoid robot table tennis demonstration at the China International Import Expo, but the information provided about that article is limited to the point that exhibition coverage should be distinguished from verification; it does not show a direct connection to the competitions discussed here. Related coverage from this newspaper includes Runway releases “Praxis-1,” connecting video pretraining to robot control, which covers the release of model weights for robot control, and BitRobot releases a 100-hour 3D task dataset, which covers the release of 3D task data. Those articles address separate issues—models and data—and cannot be assumed to be technically connected to the discussion of world models in this excerpt. The points to check in the full article are whether it gives a specific definition of world models, identifies the data and evaluation methods involved, and details the factors said to be strengths for Japan. The excerpt alone does not establish those details.