Core Insights - The key element for embodied robots currently and in the future is the robot model, which is at a developmental stage similar to the early years before the release of ChatGPT [1] - The growth rate of the intelligent machinery industry in China is estimated to be around 50% to 100% for each company this year, with industrial robot exports increasing by over 51% [1] - The progress of robot models is slower than expected, and the industry has not yet reached a critical point for embodied intelligence [2] Company Developments - Yushu Technology announced the release of the H2 bionic humanoid robot, which stands 180 cm tall and weighs 70 kg, capable of performing dance and martial arts [4] - The company has previously launched three humanoid robots: G1, H1, and R1, with the latest R1 model priced starting at 39,900 yuan, attracting market attention due to its cost-effectiveness [4] - Yushu Technology underwent a name change to Yushu Technology Co., Ltd. on October 23, with Wang Xingxing becoming the chairman [5] Industry Trends - Embodied intelligence has become one of the hottest technology sectors in recent years, with significant advancements expected in the near future [1] - The industry is currently facing challenges with the mainstream models, particularly those based on video generation, which require substantial computational power [3] - The company is focusing on deep reinforcement learning for full-body motion control, laying the groundwork for large-scale data collection [2]

Core Insights - The key element for embodied robots currently and in the future is the robot large model, with the current development stage being similar to the 1-3 years prior to the release of ChatGPT [1][3] - The industry of embodied intelligence has become one of the hottest technology sectors in recent years, with an estimated average growth of 50%-100% for intelligent machine companies in China this year [1][3] - The export growth rate of industrial robots in China has exceeded 51% this year, with expectations for even faster growth in intelligent robots [1] Industry Development - The progress of robot large models is slower than expected, and has not yet reached a critical point [3] - The company is conducting deep reinforcement learning for full-body motion control, with overall technological advancements being faster than expected but still not reaching the critical point [3] - A key capability for humanoid robots in the coming years will be the ability to quickly recover from being knocked down under strong interference [3] Model Preferences - Two mainstream models of embodied intelligence are discussed, with a preference for video-based world models [4] - The VLA+RL model is noted for its clear architecture and ability to train in simulation or real environments, but it faces challenges in generalization [5] - Video-based world models require significant computational power, which poses challenges for small and medium-sized robotics companies [5] Product Announcements - The company announced a new product on October 29, described as a compact robot with performance comparable to 2xGo2, likely a quadruped robot [7] - On October 20, the company officially launched the H2 bionic humanoid robot, which stands 180 cm tall and weighs 70 kg, capable of performing dance and martial arts [8][9] - The company has previously released three humanoid robots, with the latest model R1 launched on July 25, priced starting at 39,900 yuan, attracting market attention due to its cost-effectiveness [10] Company Background - The company recently changed its name to Yushu Technology Co., Ltd., with Wang Xingxing becoming the chairman [11] - Established in August 2016, the company has a registered capital of approximately 364 million yuan, focusing on the research and sales of intelligent robots and industrial robot manufacturing [11]

Core Insights - The development of humanoid robots is heavily dependent on the progress of embodied intelligence models, with a focus on achieving significant breakthroughs in the next one to two years [5][6][7] Group 1: Current State of Humanoid Robots - The key technology for humanoid robots is the advancement of large models, which is currently progressing slower than expected [5] - The current stage of embodied robots is likened to the early years before the release of ChatGPT, where the direction is clear but breakthroughs have not yet been achieved [6] Group 2: Future Aspirations - A significant milestone would be achieving a humanoid robot that can complete 80% of tasks in 80% of unfamiliar scenarios using natural language commands, marking a "ChatGPT moment" for the field [6][7] Group 3: Challenges in Development - There are two main areas for improvement: model architecture and data quality. Current models lack generalization capabilities, and data collection remains a significant challenge [8] - The mainstream training paths for humanoid robots face issues with generalization, where performance drops significantly when changing objects or environments [9] Group 4: Specific Technical Issues - The alignment between video generation and robot control is a critical bottleneck, as current video models can generate high-quality scenarios but struggle to translate these into precise actions in the real world [9]

Core Viewpoint - The robotics industry in China is experiencing significant growth, with an optimistic estimate of at least 50%-100% average growth for each company in the smart robotics sector this year [3][5]. Industry Growth - The export of industrial robots in China has increased by 54.9%, indicating strong growth in the industrial robotics sector, which suggests that the smart robotics sector may also see an average growth rate of around 100% this year [5]. Technological Progress - The progress of robotic large models is slower than expected, not yet reaching a critical point. However, advancements in deep reinforcement learning for full-body motion control have been significant this year compared to last year [6]. - The current state of robotic large models is likened to the period 1-3 years before the release of ChatGPT, where the direction is recognized but not yet at a breakthrough point [6]. Model Development - Both the structure of models and the quality of data need improvement. The industry faces challenges in data collection and quality assessment, which are crucial for enhancing model capabilities [6]. - Achieving true embodied intelligence, akin to a "ChatGPT moment," is anticipated to allow robots to complete about 80% of tasks in unfamiliar scenarios, which would be a significant technological breakthrough [7]. Model Preferences - Among the two mainstream models for embodied intelligence, the video generation-based world model is favored due to its clear architecture, despite its high computational demands, which may limit its use by smaller robotics companies [8]. - The VLA+RL model is also popular but faces challenges in generalization capabilities [8].

Core Viewpoint - The core issue in the development of large-scale AI models is the need for improvements in both model architecture and data quality, which should be addressed in a coordinated manner [1] Group 1: Model Architecture - Current model structures lack sufficient generalization capabilities, indicating a need for structural innovation [1] - The industry is urged to focus on innovative model designs rather than merely increasing model size [1] Group 2: Data Quality - The collection and evaluation of large-scale, high-quality data remain significant bottlenecks in the industry [1] - There is a call for the industry to invest more rational thought into data quality rather than just accumulating vast amounts of data [1]

Investment Rating - The report maintains an "Overweight" rating for the computer industry, indicating a positive outlook for the sector in the next 6 months [1]. Core Insights - The brain is identified as the key to achieving embodied intelligence, with humanoid robots consisting of a brain, cerebellum, and body, where the brain utilizes AI large models for optimal decision-making based on environmental inputs [3][8]. - The technology path for embodied intelligence is still in the exploratory phase, with three main technical routes: end-to-end VLA, brain+cerebellum layered approach, and world model technology [3][13]. - The development of the robot brain is seen as a critical bottleneck for the advancement of embodied intelligence, with current challenges primarily related to model issues rather than data [3][18]. - The global landscape features a diverse array of robot brain manufacturers, with several companies achieving valuations exceeding $1 billion, such as Galaxy General at 11.5 billion yuan and Physical Intelligence at $2.4 billion [3][17][46]. - Investment suggestions highlight the rapid growth of the embodied intelligence market, emphasizing the importance of the brain in this development, and recommend focusing on related companies [3][56]. Summary by Sections 1. Definition and Capabilities of Robot Brain - The humanoid robot's brain is based on AI large models, enabling autonomous navigation, task execution, and human-robot interaction [8][9]. 2. Parallel Exploration of Humanoid Robot Large Model Technologies - The report outlines three main technical routes for humanoid robot brain development: end-to-end VLA, brain+cerebellum layered approach, and world model technology, with ongoing research in each area [13][14]. 3. Industry Participants in Humanoid Robot Brain - Participants are categorized into three types: companies specializing in robot brains, general large model companies, and companies developing their own solutions [17]. 4. Key Bottleneck in Embodied Intelligence Development - The report emphasizes that the robot large model is the most significant challenge facing intelligent agents, limiting the large-scale application of humanoid robots [18][19]. 5. Challenges in Developing Embodied Intelligence Large Models - The report discusses the difficulties in obtaining training data for embodied intelligence, highlighting the scarcity of interaction data with the physical world [20][21]. 6. Diverse Global Robot Brain Manufacturers - The report details various companies in the sector, including Tesla, Figure AI, Physical Intelligence, and others, each with unique approaches and technological advancements [22][31][36].

Core Insights - The founder of Yushu Technology, Wang Xingxing, discussed the current state of robotics at the 2025 Inclusion Bund Conference, emphasizing that while language models excel in text and image domains, the practical application of AI in robotics is still in its infancy [3][4] - Wang highlighted the challenges in data quality and model capabilities as significant barriers to the advancement of humanoid robots, noting that the focus has traditionally been on data rather than model development [4] Data Challenges - The core issue in data utilization for robotics is the difficulty in defining what constitutes high-quality data, including the types of actions and scenarios that need to be captured [4] - There is a lack of clarity on the scale and types of data required for effective robot training, indicating that the field is still in a developmental phase regarding data standards [4] Model and Hardware Limitations - Despite advancements in hardware, the primary challenge remains the inadequacy of AI models, which struggle to effectively control robotic movements and integrate multimodal inputs such as video and language [4] - Wang expressed optimism about the future of AI models, suggesting that a more aggressive approach to understanding and utilizing these models could lead to significant advancements in robotics [4] Talent and Management Issues - The industry faces a shortage of top-tier talent, which poses a challenge for technological development, alongside management difficulties that can arise as teams grow larger [4]

Group 1 - The core viewpoint of the article highlights the rapid growth in the robotics sector, particularly focusing on humanoid robots, with significant stock price increases among key companies [1] - The Guozheng Robotics Industry Index rose by 2.3% by 10:00 AM, with notable gains from companies such as Liyuanheng (over 12% increase), Hanwei Technology (over 7% increase), and Huazhong CNC and Zhaowei Electromechanical (over 4% increase) [1] - UBTECH announced a record procurement contract worth 250 million yuan for humanoid robot products and solutions, following a previous major order of nearly 100 million yuan [1] Group 2 - Huatai Securities noted that Chinese companies are significantly investing and acquiring in the humanoid robot sector, which is expected to lead to a rapid decrease in hardware costs [1] - The innovation in robotic models powered by AI is anticipated to accelerate the overall robotics industry trend [1] - The Guozheng Robotics Industry Index emphasizes humanoid robots and core components, with nearly 80% of related stocks, making it a leading index for humanoid robots [1]

Group 1 - The China Robot Index decreased by 1.71% as of September 4, 2025, with mixed performance among constituent stocks, led by Dongjie Intelligent up 3.99% and Kewos down [1] - The Robot ETF by Jiashi (159526) has seen a 12.80% increase over the past month as of September 3, 2025 [1][3] - The Jiashi Robot ETF recorded a turnover rate of 6.42% and a transaction volume of 38.57 million yuan, with an average daily transaction of 62.59 million yuan over the past week [3] Group 2 - The Jiashi Robot ETF's net asset value increased by 75.41% over the past year, with a maximum single-month return of 25.78% since inception [3] - The top ten weighted stocks in the China Robot Index account for 50.36% of the index, with Huichuan Technology and Keda Xunfei being the top two [3][5] - On September 3, 2025, UBTECH secured a 250 million yuan contract for humanoid robot products and solutions from a well-known domestic company [5] Group 3 - Huatai Securities suggests that the initial delivery order numbers are not the key indicators; the focus should be on forming correct model paradigms and data flywheels [6] - The large-scale layout of humanoid robots by Chinese manufacturing companies is expected to lead to a non-linear decrease in hardware costs, with AI-driven innovations being crucial for market demand expansion [6]

Core Viewpoint - The domestic humanoid robot industry is experiencing significant growth, with multiple companies securing large orders, indicating a shift towards commercialization and increased demand for humanoid robots [1][3][4]. Group 1: Company Developments - UBTECH Robotics announced a procurement contract worth 250 million yuan for its humanoid robot Walker S2, which features an autonomous hot-swappable battery system, with delivery expected to start this year [1]. - Excluding the Tian Gong Walker, UBTECH's Walker series has secured nearly 400 million yuan in contracts [1]. - UBTECH expects to deliver over 500 industrial humanoid robots this year, with an annual production capacity reaching 1,000 units [2]. Group 2: Industry Trends - The humanoid robot sector is entering a phase of large-scale commercialization, with companies like Tiantai Robotics and Zhiyuan Robotics announcing substantial orders, including a global first order for 10,000 humanoid robots [3]. - As of August 4, 2023, Zhiyuan Robotics has won nine projects, Yushu Technology has won 68 projects, and UBTECH has won 16 projects, indicating a strong order intake compared to last year [4]. - The commercialization of humanoid robots is expected to lead to a non-linear decrease in hardware costs, driven by innovations in AI and large model development [4]. Group 3: Investment Insights - The application of humanoid robots is expanding from education and research to industrial manufacturing, commercial operations, and government services, suggesting a broader range of deployment scenarios [5]. - Investment firms recommend focusing on new technology iterations and component market trends, anticipating a restart of the supply chain with various components like tactile sensors and high-power density motors becoming available in the second half of the year [5].