盘核桃、叠衣服、穿烤肠，银河通用春晚机器人是怎么“练”成的？

Core Viewpoint - The performance of the robot "Xiao Gai" during the CCTV Spring Festival Gala showcases its advanced capabilities in autonomous perception, decision-making, and execution, moving beyond traditional pre-programmed actions [1][12]. Group 1: Robot Capabilities - "Xiao Gai" demonstrated its ability to perform complex tasks such as walnut shelling, folding clothes, and cooking sausages, highlighting its dexterity and adaptability [1][12]. - The robot's core competency is derived from the self-developed embodied large model "Galaxy Brain AstraBrain," which enables it to generalize skills to new tasks and scenarios [2][12]. Group 2: Training and Learning - AstraBrain utilizes a unique training approach that combines a self-constructed dataset of over 10 billion embodied intelligent data points with large-scale virtual and real data for end-to-end training [2][3]. - The robot undergoes extensive virtual practice, performing millions of trial-and-error attempts to refine its skills, such as carefully retrieving items from tightly packed shelves [6][11]. Group 3: Real-World Application - The robot's ability to adapt its actions in real-time is exemplified in tasks like folding clothes, where it has been trained on thousands of flexible object deformation data to handle various shapes and wrinkles [11][12]. - The technology behind AstraBrain not only enhances the robot's performance in entertainment settings but also proves its potential for industrial applications, such as precision assembly in factory environments [12].