Group 1 - The core viewpoint is that physical intelligence will be a key direction for the development of AI large model architecture, with computing power being the foundation for embodied intelligence [1] - The rapid development of embodied intelligent robots is closely linked to the rise of AI large models, with computing power as a critical support point for this development [1] - Nvidia's release of the Cosmos world model platform at CES 2025 aims to teach AI to understand the physical world, which is expected to revolutionize the $50 trillion manufacturing and logistics industries [1] Group 2 - There are currently two mainstream architectures for embodied large models: end-to-end embodied models and hierarchical decision models, with the latter being favored by most domestic startups due to cost and technical feasibility [2] - The end-to-end model integrates "brain" and "cerebellum," reducing error transmission but requiring massive data and resources [2] Group 3 - The global humanoid robot market is projected to grow from approximately $2.16 billion in 2023 to $32.4 billion by 2029 [3] - The humanoid robot market in China is expected to reach a scale of approximately 20 to 50 billion yuan by 2028, and could grow to 500 billion yuan by 2035 [3] - By 2045, the humanoid robot market could reach a scale of approximately 10 trillion yuan, with over 100 million humanoid robots in use across various industries [3]