Workflow
东吴证券:距离真正的具身智能大模型有多远?
SCSSCS(SH:601555) 智通财经网·2025-08-09 14:20

Core Viewpoint - The future of embodied large models will continue to evolve in three areas: modality expansion, reasoning mechanisms, and data composition [1][4] Group 1: Importance of High-Intelligence Large Models for Humanoid Robots - The key to the industrialization of humanoid robots lies in overcoming the limitations of traditional industrial robots, which are based on deterministic control logic and lack perception, decision-making, and feedback capabilities [2] - Humanoid robots aim to be "general intelligent agents," emphasizing a complete link of perception, reasoning, and execution, which requires support from large models for multi-modal understanding and generalization capabilities [2] - The rise of multi-modal large models provides humanoid robots with a "primary brain," initiating an intelligent evolution from 0 to 1, although overall intelligence is still at the L2 initial stage [2] Group 2: Progress of Large Models in Robotics from Architecture and Data Perspectives - The rapid evolution of large models in robotics is driven by breakthroughs in both architecture and data [3] - Current models have developed from early language planning models to end-to-end action output, integrating multi-modal perception capabilities into a unified model space [3] - A structured system supporting pre-training and practical capabilities has emerged, relying heavily on high-precision motion capture equipment for real-world data collection [3] Group 3: Future Development Directions of Large Models - Future embodied large models are expected to expand modalities by incorporating tactile and temperature perception channels [4] - Architectures like Cosmos aim to endow robots with "imagination" through state prediction, enhancing environmental modeling and reasoning capabilities [4] - The integration of simulation and real data for training is becoming mainstream, with high-standard, scalable training environments being crucial for the general robot training system [4] Group 4: Investment Recommendations - Companies to focus on in the model sector include Galaxy General, Star Motion Era, and Zhiyuan Robotics [5] - In the data collection field, attention should be given to Qingtong Vision, Lingyun Light, and Obsidian Zhongguang [5] - For data training environments, Tianqi Co., Ltd. is recommended [5]