中信证券:具身模型更新潮 “UMI数采+视频学习”推动世界模型发展
Di Yi Cai Jing·2026-01-16 00:31

Core Insights - The report from CITIC Securities indicates that the embodied intelligence large model sector is experiencing intensive updates, with the "low-cost + high-efficiency" characteristics of UMI data collection being widely adopted [1] - The large-scale use of first-person video data has received phased certification from leading manufacturers, suggesting that diverse and rich video learning is expected to nurture world models based on video generation models [1] - The year 2026 is projected to be the production year for the embodied intelligence robotics industry, with the level of embodied models still lagging behind hardware foundations, making their iterative evolution crucial [1] - As companies like Skild AI, Physical Intelligence, and Galaxy General continue their capitalization processes, the pricing logic for embodied model enterprises is expected to become clearer and more consistent [1]