Core Viewpoint - The development of embodied intelligence products, represented by humanoid robots, is gaining significant attention and is expected to grow rapidly, with the global market projected to reach 19.525 billion yuan by 2025, and China's market expected to reach 5.295 billion yuan, accounting for approximately 27% of the global market [2]. Group 1: Data and Embodied Intelligence - Embodied intelligence requires high-quality multimodal data for training and optimizing intelligent agents, which can be categorized into real data and synthetic data [2]. - Real data is collected through techniques like remote operation and motion capture, while synthetic data is generated through virtual modeling, with the latter being less costly but differing significantly from real data [2]. - The industry is employing data cleaning and labeling techniques to enhance data quality and ensure it meets the nutritional needs of robots [2]. Group 2: Market Growth and Challenges - The embodied intelligence industry is rapidly developing, with a projected market size of 19.525 billion yuan globally and 5.295 billion yuan in China by 2025 [2]. - A significant challenge for the industry is the lack of high-quality and diverse datasets, which hinders the capabilities of embodied intelligence [3]. - The National Data Standardization Technical Committee has proposed the construction of three types of high-quality datasets to address the data scarcity issue [3]. Group 3: Innovations and Future Directions - Experts suggest innovating data collection and governance models to reduce costs and ensure data compliance and usability [3]. - The integration of multimodal technologies to fuse different sensory data from robots is seen as a crucial step for future advancements in embodied intelligence [3]. - The "flywheel effect" indicates that as the AI industry consumes high-quality data, it simultaneously generates new data, enhancing the value extraction from data [3].
布局未来看产业 财米油盐|具身智能,“吃”得好、更聪明
Ren Min Wang·2025-08-30 11:42