Everbright issued a Research Report stating that humanoid robots currently have reached a relatively mature level of balance and movement capabilities. The core factor restricting the further scaling of the Industry lies in training, particularly in achieving further capability generalization through training. Generalization training relies on a large amount of real-world data, which is costly to obtain directly through human demonstration, and high-quality robotic movement materials are relatively scarce. The NVIDIA Cosmos model can assist developers in generating synthetic movement data at an exponential scale, potentially ushering in the era of physical AI for robots.