SIASUN(300024)2025-02-20 05:42Summary of Key Points from the Conference Call Industry Overview - The discussion revolves around the humanoid robotics industry, focusing on the development of control algorithms and hardware design for humanoid robots and their comparison with traditional industrial robots [2][3][6][12]. Core Insights and Arguments - Definition of the Cerebellum in Robotics: The cerebellum in robotics refers to the main CPU that controls the joints, differing from traditional control systems that follow predetermined trajectories. AI robots can autonomously generate trajectories based on real-time environmental assessments, unlike traditional industrial robots that rely on fixed algorithms [2]. - Potential of Traditional Utility Companies: Traditional utility companies have the potential to develop the underlying algorithms for humanoid robots by collaborating with local vendors to gather data and optimize algorithms, merging industrial control technology with AI applications [3]. - Control Model Standardization: Lower limb control is more likely to achieve a unified model due to its fixed motion patterns, while upper limb control remains diverse due to various technological paths, indicating a future possibility of a standardized solution if a dominant technology emerges [5]. - Execution Layer Similarities: There is no significant difference in the execution layer between humanoid robots and traditional industrial machinery, as both are based on similar control theories and algorithms. The main distinction lies in the AI components that generate trajectories in real-time [6]. - Integration of Risk Control and Algorithms: The integration of risk control and large model algorithms in the robotics industry is a trend rather than a conflict, indicating a need for talent in both AI computation and control [7]. - Advancements in Small Models: Technologies like DataStick enable smaller models to be embedded on the edge for real-time computation and control, suggesting a significant growth opportunity for the industry through cloud and edge model optimization [8]. - Development of Vertical Industry Models: The emergence of small vertical industry models that are trained on specific scene data is seen as a more practical and rapid application method compared to large general models, enhancing the data demand and generalization capabilities of humanoid robots [9]. - Feasibility of Feedback Data Collection: The model of deploying robots to terminals to collect feedback data for optimization is feasible, although challenges exist in collecting and optimizing data on a large platform [10]. - Hardware Design Requirements: Humanoid robots share no essential differences in components with traditional industrial robots but require adaptations such as low-voltage power supply and higher integration levels [12]. - Dynamic Capture Technology: Dynamic capture technology will continue to play a crucial role in data collection and optimization, utilizing external dynamic data to generate motion trajectories through reinforcement learning [14]. - Startup Strategies: Startups should focus on the specialization of humanoid robots, optimizing costs and adapting components for mass production to enhance market competitiveness [15]. - Market Demand for Humanoid Robots vs. Robotic Dogs: While robotic dogs have clear application scenarios and may see higher short-term production volumes, the long-term demand for humanoid robots is expected to surpass that of robotic dogs [16]. - Data Processing Structure in Robotic Dogs: Robotic dogs possess a data processing structure similar to that of humanoid robots, with a cloud-based control system and a local CPU for motion control [17]. - Data Acquisition for Reinforcement Learning: Data for reinforcement learning in motion control is obtained through simulation software or real-world deployment, both methods being widely utilized [18]. - Roles of Cerebellum and Brain in Robotics: In robotic systems, the brain refers to the cloud component responsible for complex computations, while the cerebellum refers to the local CPU that handles real-time tasks [19]. - Prominent Domestic Motion Control Companies: Several domestic companies excel in motion control, with those focusing on bottom-up research and development showing greater potential for technological advancement and innovation [20]. Other Important Insights - The integration of AI and traditional control methods is crucial for the evolution of humanoid robotics, emphasizing the need for a multidisciplinary approach in talent acquisition and technology development [7][19].