Today, with the rapid development of robot technology, UBTech's latest open-source embodied intelligence large model, Thinker, shines like a brilliant star, illuminating the future of industrial humanoid robots. The goal of this large model is to address key issues in the robotics field, such as insufficient accuracy, excessive parameters, and poor real-time performance in tasks like spatial understanding and visual perception.

UBTech revealed on its official WeChat account that current robot models face challenges: although the Internet provides massive data, the quality of the data varies greatly, making it difficult to train high-performance models for robots. Thinker, however, offers an innovative data processing solution with a full-cycle approach of "refining and purifying - automated annotation - data-driven training." It can process 20B of raw data into just 10M, significantly improving data quality.

image.png

In addition, Thinker has a powerful automated annotation system, reducing annotation costs by an astonishing 99% through a "weakly supervised + self-supervised + minimal manual verification" approach. This means that the high cost and high error rate of traditional manual annotation will be effectively alleviated. Through a continuous "annotation - training - feedback - iteration" cycle, Thinker can continuously improve its accuracy, gradually achieving more intelligent robot operations.

image.png

UBTech's Thinker large model not only provides a powerful "brain" for robots at the technical level but also lays the foundation for the development of future embodied intelligence technologies. With the open-sourcing of this large model, more developers and researchers will be able to use it for deeper research and application development, thus promoting further progress in the robotics industry.

UBTech's innovation not only demonstrates its leading position in the industry but also provides new momentum for the global development of robot technology.