人形机器人传感器行业深度 摘要 人形机器人正从"动起来"向"用起来"的关键阶段过渡，传感器成为 实现更复杂交互能力的关键瓶颈，如视觉避障和触觉反馈，对外界信息 的采集至关重要。 未来半年至一年内，预计更多公司将向特斯拉送样，引发新一轮行情。 市场正从主题投资转向对量产的预期，TR1 及灵巧手等确定性高且价值 量大的环节将持续表现良好。 2023 年 12 月新技术发布后，汉威科技（触觉传感器）、北特科技（丝 杠）和中研股份（PEEK 材料）等公司表现出色，硬件创新如灵巧手和 微型丝杠推动了股价上涨。 视觉传感器在机器人中应用广泛，通常配备三个视觉传感器和一个激光 雷达，用于路径规划和环境识别。奥比中光市场份额逐渐提高，业绩拐 点显现，预计未来季度将快速增长。 触觉传感器中，电容式用量较多，电阻式已实现量产。威科技和弗莱信 与多家公司合作，产品验证稳步推进。电子皮肤技术应用范围从手指扩 展到手掌、手背甚至脸部。 根据我们的测算，一个人形机器人所需的传感器总价值约为 24,000 元人民币。 这一数据是以特斯拉为标准进行估算的。在这 24,000 元中，视觉传感器占 30%，约 7,200 元；六维力传感器占 2 ...

Group 1 - The conversation highlights the increasing popularity of robotics, with significant attendance at events like WAIC, indicating a growing interest in the field [2][11] - There is a notable difference between domestic and international teams in robotics, with international teams focusing more on research and algorithms, while domestic teams emphasize hardware and differentiation [3][4] - The perception of tactile sensors has shifted, with current robotic hands being expected to include tactile feedback, as relying solely on vision is now seen as insufficient [6][8] Group 2 - The decline in the demand for high precision in tactile sensors has led to a reduction in costs, as the industry recognizes that human touch is not as precise as previously thought [9][31] - The concept of "bubble bursting" in the industry is viewed positively, as it may lead to technological convergence and closer alignment between hardware and market needs [11][13] - The development of humanoid robots is still in its early stages, with current capabilities estimated between levels L1.5 to L2, and future advancements are anticipated [18][19] Group 3 - The potential applications of robotics are diverse, with industrial settings being more suitable for immediate deployment compared to household or service robots [21][22] - Military applications for robotics are seen as having significant potential due to lower algorithmic requirements and the ability to perform tasks that may not be feasible for traditional robots [24] - The industry is currently focused on balancing cost and efficiency, with a consensus emerging around reducing costs while maintaining functionality [31][32] Group 4 - The challenges of data collection and model training for robotic systems are highlighted, with current methods lacking generalization capabilities [26][28] - The exploration of new training methods, such as imitation learning, is underway, aiming to improve the adaptability of robotic systems [36][37] - The hardware landscape remains fragmented, with ongoing debates about the best configurations and designs for robotic systems, impacting data accumulation and development speed [30]

Core Viewpoint - The smart industry is evolving from a single line to cross-domain integration, driven by advancements in artificial intelligence, sensors, computing power, and robotics, leading to the emergence of "aggregated intelligence" as a new direction for industrial development [1] Industry Overview - The integration of various industrial chains, such as smart cars, smart robots, and low-altitude economy, is creating numerous intersections in technology, component supply, application scenarios, and regulatory policies [1] - The ability to collect and train high-quality data in complex real-world scenarios is crucial for achieving large-scale commercial applications [1] Company Profile - HeShan Technology, founded in 2017, focuses on the "tactile" perception aspect of robotics and holds over 80% market share in the tactile sensing field [2][3] - The company provides tactile sensors and electronic skin for humanoid robots and has partnerships with major automotive brands like Mercedes-Benz, BMW, and BYD [3] Technological Challenges - The main bottleneck for robots transitioning from laboratory to large-scale application is the cost of data and training [2] - HeShan Technology aims to address this by building and open-sourcing tactile simulation models on platforms like MuJoCo and NVIDIA's Isaac Sim to reduce data scarcity and costs [2][6] Market Applications - The company targets scenarios that are easy to replicate and allow for rapid learning, such as hotels and logistics, to facilitate the commercialization of embodied intelligence [2][4] - Robots are expected to achieve a high success rate of approximately 99.7% in commercial applications, with a focus on flexible and generalized tasks [2][4] Future Outlook - Industry experts predict that robots may achieve generalization within 5 to 10 years, with optimistic views suggesting significant advancements in service sectors like elderly care [5][7] - The philosophical debate on whether robots will replace humans centers on the idea that robots will take over repetitive, dangerous tasks, allowing humans to engage in more creative and imaginative work [7]

Summary of the Conference Call on Humanoid Robots and Dexterous Hands Industry Overview - The humanoid robot market is optimistic, with expectations for mass production to begin in 2025, particularly in Q4, driven by the IPO of Yuzhu Technology and the potential release of Tesla's next-generation humanoid robot [1][2] - The dexterous hand, a key component of humanoid robots, is expected to experience explosive growth due to the increasing demand for high-quality data from embodied intelligent models, with a projected global market capacity of 7.44 million units and a market size nearing 100 billion yuan by 2035 [1][4] Key Points and Arguments - Dexterous hands outperform traditional grippers in flexibility, task operation categories, and human-machine interaction, making them the ideal choice for embodied intelligent end-effectors [1][5] - The technology route for dexterous hands primarily involves five-finger configurations, 16-20 degrees of freedom, motor drives, and cable-driven transmission [1][6] - Motor drives are the most widely used method for dexterous hands, requiring high integration and size specifications, while cable-driven transmission mimics human tendon structures [1][8] - Tesla's dexterous hand has evolved through three generations, with the third generation achieving 22 degrees of freedom and incorporating tactile sensors [1][9] Market Dynamics - The humanoid robot sector has not seen significant relative gains in the current bull market, but increasing information and funding catalysts are expected to enhance market performance in Q4 [2] - The potential for product iterations from both domestic manufacturers and leading North American firms indicates a promising future for dexterous hands [3] Competitive Landscape - Key components of dexterous hands include motors (hollow cup motors and brushless DC motors), reducers (planetary reducers), six-dimensional force sensors, tactile sensors, and cables [11] - The market for hollow cup motors is dominated by German and Swiss manufacturers, while American and Japanese firms lead in brushless DC motors [11] - Domestic companies like Yuli and Kunwei are positioned to achieve domestic substitution in the six-dimensional force sensor market [11] Investment Recommendations - Recommended companies include Zhaowei Electromechanical for its comprehensive production capabilities in dexterous hands, and Ankuo Technology and Kexin New Materials as significant suppliers of six-dimensional force sensors [12] - Other notable mentions include Mingzhi Electric for precision motion control, Beite Technology and Wuzhou New Spring for screw components, and Hanwei Technology and Fule New Materials for flexible tactile sensing systems [12]

Summary of Conference Call on Humanoid Robot Sensors Industry Overview - The humanoid robot market is optimistic, with sensors representing a high-value and high-barrier segment, accounting for over 20% of the total value. The global market size is expected to reach 85 billion yuan by 2035, with a compound annual growth rate (CAGR) of approximately 47% from 2025 to 2035, driven by intelligent upgrades of sensors [1][3]. Key Points and Arguments - **Main Sensor Configurations**: The primary sensor configurations in humanoid robots include force sensors, tactile sensors, IMUs (Inertial Measurement Units), and visual sensors. For instance, Tesla's humanoid robot requires 28 one-dimensional force sensors and 4 six-dimensional force sensors, with a total sensor value of approximately 90,000 yuan per unit [5]. - **Force Sensors**: Force sensors are categorized from one-dimensional to six-dimensional. Humanoid robots are expected to become the largest downstream application for six-dimensional force sensors. Cost control is crucial, with Tesla's target price for the complete robot set at 20,000 to 30,000 USD, implying a price cap of about 2,000 to 3,000 USD for the six-dimensional force sensors [6]. - **Market Potential for Tactile Sensors**: The tactile sensor market is projected to undergo a technological upgrade from rigid to flexible sensors, with a market size expected to reach 13.4 billion yuan by 2035 and a CAGR of about 60% from 2025 to 2035 [4][11]. - **IMU Market Growth**: The IMU market for humanoid robots is anticipated to reach 7.5 billion yuan by 2035, with a CAGR of approximately 65% from 2025 to 2035. IMUs are critical for posture control and have broad downstream applications [4][13]. - **Visual Sensors**: The visual sensor market, particularly 3D vision, is expected to grow significantly as humanoid robots become more widely adopted. The 3D vision market is projected to reach 13.4 billion yuan by 2035, with a CAGR of about 65% from 2025 to 2035 [14]. Additional Important Insights - **Technical Barriers**: The main technical barriers for six-dimensional force sensors include algorithms, calibration, and manual assembly. These barriers require significant investment and long-term experience to overcome [9]. - **Cost Reduction Strategies**: One potential path for reducing the cost of six-dimensional force sensors is the use of silicon strain gauge sensors, which can facilitate mass production and lower costs. Domestic company Anpeilong is currently developing this technology [10]. - **Key Players in Sensor Market**: Recommended companies with advantages in various sensor fields include Anpeilong and Keli for six-dimensional force sensors, Hanwei Technology and Fulaixin Material for tactile sensors, Xindong Lianke for IMUs, and Aopu Zhongguang and Aopute for visual sensors. These companies are positioned to benefit from industry growth and the acceleration of domestic replacements [15].

Group 1: Humanoid Robots - The humanoid robot industry requires integration of AI and ML technologies on the software side, while the hardware side focuses on breakthroughs in joint actuators and other key components [2][9] - Key components in the humanoid robot industry include harmonic reducers, frameless torque motors, planetary roller screws, and force sensors, which have a high value share [9][15] - The humanoid robot market in China is expected to grow significantly, with a projected market size of approximately 27.6 billion yuan in 2024, increasing to 750 billion yuan by 2029 [2][4] Group 2: Robot Bearings - The global industrial robot bearing market is projected to reach 9.004 billion yuan in 2023, with an expected growth to 13.585 billion yuan by 2029 [2][31] - The localization rate of high-end bearings for humanoid robots is currently below 20%, indicating significant room for domestic companies to break through [2][4] - Major international players dominate the high-end bearing market, while Chinese companies are gradually making progress in this sector [2][4] Group 3: Humanoid Robot Sensors - Sensors are critical components for human-robot interaction, with a focus on multi-modal perception systems covering force, vision, and touch [2][3] - The global market for six-dimensional force sensors is expected to reach 13.84 billion yuan by 2030, highlighting the importance of these sensors in dynamic control of humanoid robots [2][3] - Visual sensors are evolving from 2D to 3D technology, enhancing the environmental perception capabilities of humanoid robots [2][3] Group 4: Applications of Humanoid Robots - The application scenarios for humanoid robots are expected to evolve from industrial services to household and commercial services, driven by the standardization and complexity of tasks [4][5] - In China, humanoid robots are primarily applied in industrial production, accounting for 29% of the market, as they meet the automation and intelligence needs of the manufacturing sector [4][5] - Global trends show that North America focuses on warehouse logistics and industrial production, while Europe emphasizes household service applications [4][5] Group 5: Key Companies in the Sector - Key companies in the humanoid robot industry include Kete Co., which plans to establish a joint venture with Xutong Electronics to enter the embodied intelligence sector [5][8] - Audiwei specializes in sensor technology, launching underwater ranging sensors and flexible sensors [5][8] - Dingzhi Technology is a rare player in motion control solutions, diversifying into ball screws and hollow cup motors [5][8]

Core Viewpoint - Yuan Sheng Intelligent Robotics has completed a multi-million yuan financing round, focusing on the development of dexterous robotic hands, with a commitment to practical and reliable designs that prioritize usability over theoretical performance [1][2]. Group 1: Company Overview - Yuan Sheng Intelligent Robotics was established in November 2024, with a founding team from Tencent Robotics X, having over ten years of experience in dexterous hands and tactile perception, with nearly 50 published papers and over 100 patents [1]. - The company emphasizes the dexterous hand as the "soul" of humanoid robots, which directly influences the robot's ability to perform fine motor tasks such as grasping and manipulation [1]. Group 2: Technical Approach - Unlike competitors claiming numerous degrees of freedom, Yuan Sheng focuses on a practical design with approximately 22 degrees of freedom, based on human hand anatomy, prioritizing reliability over maximum performance [2]. - The company aims to create a "hexagonal warrior" product that balances flexibility, response speed, load capacity, robustness, and tactile perception, avoiding a singular focus on extreme performance [2]. Group 3: Market Strategy - Yuan Sheng targets semi-structured environments, such as flexible assembly tasks in factories, with plans to expand into home applications as technology matures [3]. - The business model aims to serve as a "universal platform," providing complete hardware, software systems, and demo cases to lower entry barriers for clients, allowing them to focus on algorithm and model development [3]. - The company believes that hardware stability is essential for enabling self-exploration training in real-world environments, which is crucial for the development of embodied intelligence [3].

Core Insights - The humanoid robot industry is experiencing significant growth, with a reported eightfold increase in order volume compared to previous years, indicating a robust demand for core components [1] - The domestic production rate of core components for humanoid robots in China has surpassed 70%, with the industry scale expected to exceed 37.9 billion yuan this year [1] - The advancement of the humanoid robot industry is closely linked to the maturity of the automotive supply chain, which has contributed to lower costs and improved quality control [2][3] Industry Development - The automotive industry has played a crucial role in the development of humanoid robots, with many components being repurposed from automotive technology [3] - Major automotive manufacturers, including Tesla and NIO, are entering the humanoid robot sector, highlighting the technological convergence between robotics and electric vehicles [3] Challenges and Opportunities - Despite the reduction in supply chain costs, high product prices remain a significant barrier to the widespread application of humanoid robots [5] - The development of core components specifically designed for humanoid robots is still in progress, as existing components are often adapted from other industries [4] - Stability and consistency of humanoid robots are critical concerns for application developers, with ongoing efforts to enhance these aspects for practical use [6][7]

Investment Rating - Industry Rating: Outperform the market [5] - Rating Change: Maintain [5] Core Insights - Dexterous hands are the core solution for humanoid robot end effectors, with current technology routes not converging and significant customization demands from downstream customers, giving an advantage to whole hand and module manufacturers [1][3] - Dexterous hands mimic human hands, integrating drive systems, transmission systems, and sensing systems, offering high flexibility but at a higher cost and lower reliability and maintainability compared to traditional grippers [1][3] - The report identifies three main trends in the development of dexterous hands: the coexistence of multiple technology routes, an increase in degrees of freedom and sensor usage, and the necessity for collaboration between manufacturers and external suppliers due to high R&D costs and long cycles [3][27] Summary by Sections Section 1: Dexterous Hands as Core Solutions - Dexterous hands are designed to flexibly manipulate objects and meet various operational needs, with Tesla's dexterous hand currently in its third iteration capable of performing complex tasks [13][15] - The key difference between dexterous hands and traditional grippers is the higher degree of freedom in dexterous hands, allowing for more complex grasping operations [1][19] Section 2: Core Systems of Dexterous Hands - The dexterous hand consists of three core systems: drive system, transmission system, and sensing system [27] - The drive system primarily includes electric, hydraulic, pneumatic, and shape memory alloy drives, with the electric drive being the mainstream solution [27][30] - The transmission system includes reducers, screws, and tendon systems, with planetary reducers and worm gearboxes being the most commonly used [27][46] - The sensing system comprises force/moment sensors, position sensors, and tactile sensors, which are crucial for providing feedback and enhancing the dexterous hand's functionality [27][75] Section 3: Development Trends of Dexterous Hands - The technology routes for dexterous hands are expected to remain diverse in the short term, with no convergence anticipated [3][92] - Future improvements in dexterous hands will likely include increased degrees of freedom and a rise in the usage of tactile sensors, which will enhance the overall value of the dexterous hand [3][92] - The high R&D investment and long development cycles necessitate collaboration between manufacturers and external suppliers, as dexterous hand development constitutes a significant portion of humanoid robot engineering [3][92] Section 4: Investment Recommendations - Companies such as Jiechang Drive, Longsheng Technology, Zhaowei Electromechanical, and Jiangsu Leili are highlighted for their potential in the dexterous hand market due to their innovative approaches and collaborations [8][4]

Core Insights - The event highlighted the significance of embodied intelligence as a frontier in artificial intelligence, emphasizing its potential to transform various industries [3]. Group 1: Embodied Intelligence - Embodied intelligence is closely related to the physical environment, with core capabilities including perception, decision-making, action, learning, and application [3]. - This technology allows robots to not only perceive the present but also to "imagine the future," which is seen as a breakthrough for industrial transformation [3]. - The integration of embodied intelligence is expected to enhance automation across manufacturing, logistics, and healthcare, transitioning factories to flexible production spaces [3]. Group 2: Design and Functionality - The design of embodied intelligent systems should not be limited to humanoid forms but should be function-driven and vary by application scenario [3]. - In industrial contexts, specialized robotic arms may offer greater efficiency than humanoid robots [3]. - Future embodied intelligent systems are anticipated to exhibit more diverse forms and characteristics [3]. Group 3: Collaboration and Development - The event included an interactive session where industry representatives engaged with experts on key technologies such as embodied intelligence, perception chips, tactile sensors, and underwater robots [5]. - Over 40 companies, universities, and research institutions participated in discussions aimed at fostering collaboration in joint laboratories, technical challenges, and talent development [5].