News

As embodied intelligence technology continues to iterate, humanoid robots' scene recognition, command understanding, and task planning capabilities have steadily improved, and they have gradually acquired the basic capabilities for application in real industrial scenarios. However, in the dynamic and changeable production line actual operation environment, the traditional embodied solutions common in the industry still have obvious shortcomings: most models can only complete static decision-making and short-term task disassembly based on real-time scenarios. They lack the ability to scenario deduction and future state prediction of the physical world, and are difficult to adapt to the real-time changing long-distance task operation requirements of real industrial scenarios. As a result, the entire link of decision execution and iterative optimization cannot be closed.

UBTECH is based on the real operational needs of industrial scenarios and the evolution direction of embodied intelligence technology. Based on the self-embodied intelligence large model Thinker, UBTECH has launched a new embodied intelligent world model Thinker-WM to create a physical AI base and achieve an all-round upgrade of the embodied brain architecture and core capabilities. With its advantages in spatial generalization, long-range tasks and comprehensive performance, Thinker-WM topped the list of the authoritative developed embodied intelligence evaluation benchmark Libero. Also participating in the list are models from well-known companies such as NVIDIA, Physical Intelligence, and Xiaomi.