SIASUN(SZ:300024)2025-02-24 07:34Summary of the Conference Call on Figure Helix Model and Robotics Industry Company and Industry Overview - The discussion centers around the Figure Helix model and its implications for the robotics industry, particularly in the context of humanoid robots and their commercialization. Key Points and Arguments 1. Figure Helix Model Architecture - The Figure Helix model utilizes a 70 billion parameter large model with a hierarchical brain architecture that does not rely on cloud computing, significantly lowering deployment barriers for commercial applications [1][4] - It features a fast-slow system (S1 and S2) that balances rapid response and generalization capabilities, achieving an operational frequency of up to 200 Hz, updating every 5 milliseconds [1][4] 2. Data Training and Generalization - The model is trained on a trillion-level text-image dataset, enhancing its ability to recognize previously unseen data and objects in home environments [1][6] - Despite having 500 hours of training data, domestic manufacturers can generate similar amounts of data within a week, indicating that data alone may not be a significant competitive advantage [10] 3. Domestic Market Dynamics - Domestic manufacturers can quickly adapt the Figure Helix system framework, achieving similar results in 2 to 3 months due to rapid data collection and synthesis capabilities [1][8] - The Chinese robotics market is characterized by diverse application scenarios, providing opportunities for various manufacturers to develop tailored solutions [16] 4. Future Trends in Robotics - The hardware design of robots is expected to converge towards a unified structure over the next 5 to 6 years, with a focus on bipedal locomotion surpassing wheeled designs in terms of control, stability, and energy efficiency [2][17] - The Shanghai government is establishing a data collection center to standardize data gathering and make it available to manufacturers, aiming to produce 20 million data points for further research and development [28] 5. Challenges and Limitations - The Figure Helix model faces challenges in data collection, relying on synthetic data which may not be as effective as real-world data, particularly in complex scenarios [19][36] - The model's high-frequency claims may be overstated, with actual performance likely between 10 to 50 Hz [21] 6. Comparative Analysis with Competitors - The Figure Helix model is positioned against competitors like Yichang, which relies on external technology, while Figure Helix emphasizes full autonomy and superior motion control capabilities [13] - Companies like ByteDance and Huawei could rapidly advance in robotics if they enter the field, but they lack manufacturing capabilities essential for hardware production [20] 7. Market Cost and Product Trends - The cost of domestic robots is expected to stabilize around 100,000 yuan for smaller models, with a shift towards miniaturization to meet market demands [31][32] - Full-sized robots (1.7 to 1.8 meters) may become less viable due to high costs, leading to a trend towards smaller, more affordable models [32] 8. Technological Innovations and Future Directions - The industry is exploring DeepMind's influence and the potential of world models, though challenges remain in accurately modeling real-world dynamics [33] - The collaboration between different manufacturers and the establishment of a cloud-based multi-agent framework could enhance cooperative robotics in the future [37] Other Important Insights - The collaborative training process for robots involves simultaneous operation by two operators, enhancing the model's ability to infer actions based on visual signals [23] - The training cost and time for expanding the Figure Helix model to handle more complex tasks remain high, indicating that current generalization capabilities are not yet optimal [18] This summary encapsulates the critical insights from the conference call regarding the Figure Helix model and the broader robotics industry, highlighting both opportunities and challenges ahead.