证券研究报告 具身智能科技前瞻探索(第1期) 2025/ 03 / 01/ 姓名: 王浩 (分析师) 邮箱:wanghao2@gtht.com 电话:0755-23976666 - P 证书编号： S0880513090004 证书编号: S0880124070046 姓名: 张涵 (研究助理) 邮箱: zhanghan5@gtht.com 电话:0755-23976666 《具身智能科技前瞻探索》第1期 报告要点 01 【具身大模型】HALO：面向具身多模态思维链推理的统一VLA模型 香港科技大学等研究团队提出HALO,一种统一的VLA模型，通过具身多模态思维链推理（包含文本推理、视觉子目标预测与动作预测），显著提升复杂 操作任务表现，RoboTwin2.0模拟基准上平均成功率达到80.5%，超越基线模型pi0达34.1个百分点。 02 【轻量化部署】QuantVLA：面向VLA模型的尺度校准后训练量化方法 VLA模型随规模与任务复杂度提升，存算开销激增，难以部署于资源受限的机器人平台，俄亥俄州立大学等团队提出QuantVLA，面向VLA模型的尺度校 准后训练量化框架，将模型权重量化至4比特、激活量化至8比 ...

当晚的舞台上，各家企业纷纷亮出底牌：松延动力的双足与仿生人形机器人与蔡明同台演绎小品；宇树 机器人通过腾空、出拳及高难度空翻大秀武术功底；魔法原子的人形机器人化身歌舞伴舞，银河通用则 跨界参演了贺岁微电影。 机器人登上春晚舞台可以回溯到2016年。当年，优必选以540台Alpha小型人形机器人集体表演亮相，随 后又陆续三次登上春晚舞台。此后，宇树也曾在2021年牛年春晚展示机器牛"犇犇"。 直到2025年春晚，宇树的人形机器人H1穿着花棉袄转着手绢登台，实现了现象级破圈，被行业视为人 形机器人走向大众视野的里程碑事件。 炒股就看金麒麟分析师研报，权威，专业，及时，全面，助您挖掘潜力主题机会！ （来源：智通财经） 今年的春晚，不仅是面向全民的科技秀场，更像是一场人形机器人赛道的生存暗战。 源：央视新闻 "这标志着人形机器人产业正从单纯的'技术秀场'迈入'规模化品牌竞争'的早期阶段。"CIC灼识咨询副总 监张笑璐向智通财经记者指出，企业集体登台，核心意图在于争夺大众的认知心智，为后续B端（企 业）和C端（用户）市场的客户信任建立铺垫。 这种"组团"现象背后，是资本态度的剧烈转变。一位具身智能领域投资人向智通财经记 ...

Core Insights - Embodied intelligence has gained significant traction globally, with Figure achieving a valuation of $39 billion despite zero revenue, while domestic players are securing commercial orders and projecting substantial revenue growth [1][4] - The Chinese government has integrated embodied intelligence into its key industrial strategies, indicating a robust market potential that is not merely speculative [1][9] - The competition between China and the U.S. in embodied intelligence is intensifying, with both nations striving to innovate and apply this technology across various sectors [6][11] Definition and Understanding - Embodied intelligence is recognized as a crucial development in artificial intelligence, characterized by agents that interact with their environment through a physical body, showcasing autonomy and adaptability [2] - It represents a convergence of machine learning, computer vision, and robotics, marking a significant step towards practical AI applications [2] Commercial Applications - Different forms of embodied intelligent robots are evolving to meet diverse needs across retail, dining, manufacturing, logistics, education, and healthcare [4] - Commercial applications focus on enhancing service experiences and operational flexibility in dynamic environments, while industrial applications emphasize precision and stability in structured settings [4] Strategic Importance - Embodied intelligence is pivotal for upgrading technology supply chains and fostering new industries, contributing to the competitive edge of nations [6] - The breakthroughs in this field are essential for China's long-term economic benefits and technological self-reliance [6] Policy Support - The Chinese government has actively promoted the development of embodied intelligence through various action plans and funding initiatives, facilitating industry growth [9][8] Development Stages - The evolution of embodied intelligence can be categorized into three phases: conceptual development (1950s), technological accumulation (2000-2020), and application expansion driven by large models (2020 onwards) [11] - The competition between China and the U.S. is evident in foundational models, computational power, and practical applications [11] Bottlenecks and Challenges - The industry faces challenges such as data scarcity, high costs of core components, and the need for improved training efficiency and commercial viability [13][16] - The lack of high-quality multimodal data and the maturity of technologies like dexterous hands are significant hurdles [13][25] Data Acquisition and Solutions - Current data acquisition methods include remote operation, simulation, motion capture, and internet video, but high-quality data remains scarce [16] - The industry is exploring solutions like "world models" and data collection training grounds to alleviate data challenges [19] Model Evolution - The VLA model is emerging as a consensus for development, integrating large language model reasoning with real-world perception and action capabilities [21] - This evolution is expected to lead to a significant leap in embodied intelligence capabilities [21] Commercialization Trends - The commercialization of embodied intelligence is progressing through various dimensions, with initial applications focusing on low-complexity, high-ROI scenarios [31] - The business model is shifting from hardware sales to service subscriptions and performance-based payments [35] Global Market Predictions - The global market for embodied intelligence is projected to reach 19.2 billion RMB by 2025, with a compound annual growth rate of 73% over the next five years [46] - China's market is expected to grow from 2.1 billion RMB in 2025 to over 280 billion RMB by 2035, indicating a hundredfold increase in a decade [50] International Expansion - Chinese companies are accelerating their international presence, transitioning from core capabilities to localized applications in global markets [53] - Successful case studies illustrate the feasibility of Chinese embodied intelligence in high-standard international markets [53] Competitive Landscape - The competition in embodied intelligence features three main players: AI-native challengers like Figure, traditional industrial players like ABB, and cross-industry giants like Tesla [55] - The industry is witnessing early signs of product homogenization, suggesting an impending consolidation phase [57] Startup Strategies - Startups must leverage their agility and innovation to survive against established giants, focusing on strategic partnerships and long-term value creation [59]

2025年接近年末，绿洲资本创始合伙人张津剑跟成立刚一年的具身智能创企HillBot联合创始人苏昊共进了一顿晚餐。 苏昊给张津剑展示了一些数据，进而抛出一个想法：他觉得具身智能即将在2026上半年走到GPT-2时刻。 这在行业内并不是共识。更多受访者认为，行业应该还没走到GPT-1。 具身智能赛道火在大模型之后，又与大模型紧密相关。尽管从技术上无法完全对标，但投资人愿意用"GPT-1"之类的表述试图对行业发展阶段进行定位 ——这从根本上影响他们选择是否加码、何时加码以及加多少码。 "GPT-1是搭建一个验证它是否可行的技术环境，GPT-2是本质上证明了某些技术路径是可行的。"张津剑对所谓具身智能的"GPT"时刻下如此定义。 这种定位十分重要。假设你在GPT-3.5和GPT-1/2两个时期投进OpenAI，那么2026年你得到的估值增长将分别是30倍和大约100倍。 奇怪的是，具身智能还远远没有走到GPT-3.5阶段，只因宇树科技在2025年春晚舞台上意外走红，就提前浮出了水面。 "好在最后磨下来了点（额度），但真的是'生磨'。" 类似的例子有很多。另一位AI领域投资人告诉智通财经记者，三年前曾有一个具身智能项 ...

Core Viewpoint - The humanoid robot industry is transitioning from entertainment-focused applications to practical, value-creating roles in various sectors, with a significant increase in production expected in the coming years [1][2][3]. Industry Outlook - The humanoid robot shipment in China is projected to reach 18,000 units in 2025, a surge of over 650% compared to 2024, and is expected to rise to 62,500 units in 2026 [2]. - The industry is moving towards practical applications, with robots expected to perform tasks in factories, construction sites, and logistics warehouses, rather than just serving as performers [2][3]. Investment Trends - Investors are now prioritizing companies that can demonstrate real-world applications and stable products, moving away from those that lack a solid business model or rely on minimal teams [3][4]. - The focus has shifted from merely having advanced technology to ensuring that robots can effectively operate in real-world scenarios and generate economic value [4][12]. Technological Development - Three main technical paths are emerging in the humanoid robot sector: VLA (Visual Language Action) model, world model, and layered decision-making with hardware-software collaboration [6][8]. - The VLA model aims for general intelligence, allowing robots to understand and execute complex commands, but faces challenges in computational demands and data requirements [6][7]. - The world model approach, exemplified by Tesla, focuses on creating a digital simulation of the physical world to predict actions and outcomes, reducing reliance on real-world data [8]. - The layered decision-making approach breaks down tasks into manageable components, enhancing reliability and efficiency in real-world applications [8][15]. Market Dynamics - The industry is witnessing a shift towards practical applications, with a growing demand for robots that can operate in specific environments and perform tasks like assembly and logistics [12][16]. - The market is increasingly focused on B2B solutions, where robots can work alongside humans without requiring significant infrastructure changes [16][18]. Future Trends - The next 3 to 5 years are critical for the deployment of robots in specific scenarios, with an emphasis on enhancing their operational capabilities and reliability [12][17]. - The industry is expected to see a convergence of technology paths, with a focus on integrating hardware and software to improve performance and adaptability [17][18]. - There is a growing trend towards domestic production of key components, which will support the development of more cost-effective and efficient robotic solutions [18].

Core Insights - Embodied intelligence has gained significant traction globally, with Figure achieving a valuation of $39 billion despite zero revenue, while domestic players are securing commercial orders and projecting substantial revenue growth [1][4] - The Chinese market is integrating embodied intelligence into its strategic development plans, indicating a shift towards a trillion-dollar market landscape [1][9] Definition and Understanding - Embodied intelligence is recognized as a crucial development in artificial intelligence, characterized by agents that interact with their environment through a closed-loop of perception, understanding, decision-making, and action [2] - It represents a convergence of machine learning, computer vision, and robotics, marking a significant step towards practical AI applications [2] Commercial Scene Classification - Different forms of embodied intelligence robots are evolving to meet diverse needs across retail, dining, manufacturing, logistics, education, and healthcare [4] - Commercial applications focus on enhancing service experiences and operational flexibility in dynamic environments, while industrial applications emphasize precision and stability in structured settings [4] Strategic Significance - Embodied intelligence is pivotal in narrowing the technological gap between China and the U.S., driving innovation across various sectors including manufacturing, transportation, and healthcare [6] - The competition in advanced technology between the two nations highlights the importance of breakthroughs in embodied intelligence for economic and competitive advantages [6] Policy Incentives - The Chinese government is actively promoting the development of embodied intelligence through various policies, funding, and standardization efforts [8][9] - Local governments are also implementing initiatives to support industry growth, including funding for humanoid robots and establishing collaborative platforms [9] Development Stages - The evolution of embodied intelligence can be categorized into three phases: conceptual development (1950s), technological accumulation (2000-2020), and application expansion driven by large models (2020 onwards) [11] - The current phase is marked by rapid advancements, with the U.S. leveraging its computational resources and China accelerating its catch-up through policy support and industry collaboration [11] Bottlenecks and Challenges - The transition from experimental to commercial applications faces challenges, including data scarcity, technological maturity, high costs, and long ROI cycles [13][16] - Key issues include the lack of high-quality multimodal data, underdeveloped technologies for dexterous manipulation, and ethical considerations [13] Data Challenges - The industry relies on various data acquisition methods, but high-quality data remains scarce, posing a significant bottleneck for development [16] - Efforts are underway to alleviate data challenges through innovative solutions like "world models" and data collection training grounds [19] Model Evolution - The VLA model is emerging as a consensus for the development of embodied intelligence, integrating reasoning capabilities with real-world perception and action [21] - This evolution is expected to lead to a breakthrough similar to the GPT moment in AI, with significant implications for cross-scenario learning and application [21] Commercialization Breakthroughs - The path to large-scale commercialization of embodied intelligence hinges on overcoming challenges in endurance, latency, execution, reliability, and economic viability [29] - Current applications are focusing on low-complexity, high-ROI scenarios, with future expansions into more complex environments as technology matures [31] Global Market Predictions - The global market for embodied intelligence is projected to reach 19.2 billion RMB by 2025, with a compound annual growth rate of 73% over the next five years [46] - China's market is expected to experience significant growth, potentially reaching over 280 billion RMB by 2035, driven by a robust industrial ecosystem [50] International Expansion - Chinese companies are accelerating their international presence, transitioning from core capabilities to localized applications in global markets [53] - Successful case studies illustrate the feasibility of Chinese embodied intelligence in meeting high international standards [53] Competitive Landscape - The competition in embodied intelligence features three main players: AI-native challengers like Figure, traditional industrial players like ABB, and cross-industry giants like Tesla [55] - The market is witnessing early signs of product homogenization, suggesting an impending consolidation phase [57] Startup Strategies - Startups must leverage their agility and innovation to survive against established giants, focusing on strategic partnerships and long-term value creation [59]

Core Insights - Embodied intelligence has gained significant traction globally, with Figure achieving a valuation of $39 billion despite zero revenue, while domestic players are securing commercial orders and projecting substantial revenue growth [1][9] - The Chinese government has integrated embodied intelligence into its key industrial strategies, indicating a robust market potential [1][9] Definition and Understanding - Embodied intelligence is recognized as a crucial development in artificial intelligence, characterized by agents that interact with their environment through a physical body, showcasing autonomy and adaptability [2] - It represents a convergence of machine learning, computer vision, and robotics, marking a significant step towards practical AI applications [2] Commercial Scene Classification - Different forms of embodied intelligent robots are evolving to meet diverse needs across retail, dining, manufacturing, logistics, education, and healthcare [4] - Commercial applications focus on enhancing service experiences in dynamic environments, while industrial applications emphasize precision and stability in structured settings [4] Strategic Significance - Embodied intelligence is pivotal in narrowing the technological gap between China and the U.S., driving innovation across various sectors [6] - It plays a vital role in upgrading the technology supply chain and fostering new industries, impacting long-term economic benefits and national competitiveness [6] Policy Incentives - The Chinese government is actively promoting the standardization and implementation of embodied intelligence through various supportive policies and funding initiatives [9] Development Stages - The evolution of embodied intelligence can be categorized into three phases: conceptual development (1950s), technological accumulation (2000-2020), and application expansion driven by large models (2020 onwards) [11] - The competition between China and the U.S. is intensifying, with both countries leveraging their unique strengths to advance in foundational models and application deployment [11] Bottlenecks and Challenges - The industry faces significant challenges, including data scarcity, technological maturity, high costs, and long ROI cycles, which hinder large-scale commercialization [13] - Data collection methods are varied but still insufficient for driving model generalization and practical applications [16] Data Breakthroughs - The industry is exploring solutions to data challenges through innovative approaches like "world models" and data collection training grounds, which are expected to alleviate data scarcity issues [19] Model Evolution - The VLA model is emerging as a consensus for development, integrating large language model reasoning with real-world perception and action capabilities [21] - This evolution is expected to lead to a significant leap in embodied intelligence capabilities, akin to the breakthroughs seen with large language models [21] Commercialization Trends - The commercialization of embodied intelligence is progressing through various application scenarios, with initial focus on low-complexity, high-ROI environments [31] - The industry is transitioning from hardware sales to service subscription models, indicating a shift in business strategies [35] Global Market Predictions - The global market for embodied intelligence is projected to reach 19.2 billion RMB by 2025, with a compound annual growth rate of 73% over the next five years [46] - China's market is expected to experience significant growth, potentially exceeding 280 billion RMB by 2035 [50] International Expansion - Chinese companies are accelerating their international presence, demonstrating the feasibility of their technologies in global markets [53] - Successful case studies highlight the adaptability and competitiveness of Chinese firms in high-standard international markets [53] Competitive Landscape - The competition in the embodied intelligence sector is characterized by three main forces: AI-native challengers, traditional industrial players, and cross-industry giants [55] - The market is witnessing early signs of product homogenization, suggesting an impending consolidation phase [57] Startup Strategies - Startups must leverage their agility and innovation to survive against established giants, focusing on strategic partnerships and long-term value creation [59]

Core Viewpoint - The next wave of artificial intelligence (AI) will transition from virtual content generation to Physical AI, enabling autonomous completion of complex tasks in the real world by 2026 [1] Group 1: Trends in Physical AI - Trend 1: Robotaxi will move from regional trials to large-scale operations, driven by reduced costs of core hardware and supportive regulations in major markets [3][4] - Trend 2: The shipment volume of humanoid robots is expected to double, leading to a competitive landscape where companies lacking closed-loop evolution capabilities may face challenges [5][7] - Trend 3: AI Agents will evolve from simple Q&A bots to personal intelligent partners, increasing interaction with the real world and intensifying competition between terminal and application manufacturers [8][10] Group 2: Developments in Wearable and Cleaning Devices - Trend 4: A surge of innovative wearable devices will emerge, focusing on specific functions and seamless AI integration for health and interaction [11][13] - Trend 5: AI toys will enhance emotional companionship capabilities, moving away from mere conversation to empathetic interactions [14][15] - Trend 6: The cleaning appliance sector will see accelerated embodiment, with products gaining advanced capabilities to perceive and adapt to their environments [16][17]

Core Insights - Embodied intelligence has gained significant traction globally, with Figure achieving a valuation of $39 billion despite zero revenue, while domestic players are securing commercial orders and projecting substantial revenue growth [1][4] - The Chinese market is integrating embodied intelligence into its strategic development plans, indicating a shift towards a trillion-dollar market landscape [1][9] Definition and Understanding - Embodied intelligence is recognized as a crucial development direction in artificial intelligence, characterized by agents that interact with their environment through a closed-loop of perception, understanding, decision-making, and action [2] - It represents a convergence of machine learning, computer vision, and robotics, marking a significant step towards practical AI applications [2] Commercial Scene Classification - Different forms of embodied intelligence robots are evolving to meet diverse needs across retail, dining, manufacturing, logistics, education, and healthcare [4] - Commercial applications focus on enhancing service experiences and operational flexibility in dynamic environments, while industrial applications emphasize precision, load capacity, and stability [4] Strategic Significance - Embodied intelligence is pivotal in upgrading the technology industry and fostering new sectors, contributing to the collaborative innovation of advanced components like chips and sensors [6] - The competition between China and the U.S. in this field is critical for economic benefits and national competitiveness, positioning it as a key battleground for the next decade [6] Policy Incentives - The Chinese government is actively promoting the development of embodied intelligence through various policies, funding, and standardization efforts [9] - Local governments are also implementing plans and initiatives to support the industry, including establishing funds for humanoid robots and fostering collaboration [9] Development Stages - The evolution of embodied intelligence can be divided into three phases: conceptual development from the 1950s, a technology accumulation phase from 2000 to 2020, and an application expansion phase post-2020 [11] - The U.S. has a head start due to its computational resources and capital ecosystem, while China is rapidly catching up through policy support and industry collaboration [11] Bottlenecks and Challenges - The industry faces significant challenges, including data collection difficulties, high costs of core components, and the need for improved training efficiency [13][16] - The lack of high-quality, real-world data and the immaturity of certain technologies are major constraints on large-scale commercialization [13] Data Acquisition and Breakthroughs - Current data acquisition methods include remote operation, simulation, motion capture, and internet video, but high-quality data remains scarce [16] - Efforts are underway to alleviate data challenges through innovative solutions like "world models" and data collection training grounds [19] Model Evolution - The VLA model is emerging as a consensus for development, integrating large language model reasoning with real-world perception and action capabilities [21] - This evolution is expected to lead to a significant leap in embodied intelligence capabilities, akin to the breakthroughs seen with large language models [21] Commercialization Trends - The commercialization of embodied intelligence is progressing through various application scenarios, with initial focus on low-complexity, high-ROI environments [31] - The industry is transitioning from hardware sales to service subscription models, indicating a shift towards more integrated business approaches [35] Global Market Predictions - The global market for embodied intelligence is projected to reach 19.2 billion RMB by 2025, with a compound annual growth rate of 73% over the next five years [46] - China's market is expected to experience significant growth, potentially exceeding 280 billion RMB by 2035, driven by a robust industrial ecosystem [50] International Expansion - Chinese companies are accelerating their international presence, moving from core capabilities to localized applications in global markets [53] - Successful case studies highlight the feasibility of Chinese embodied intelligence in meeting high international standards [53] Competitive Landscape - The competition in embodied intelligence features three main players: AI-native challengers like Figure, traditional industrial players like ABB, and cross-industry giants like Tesla [55] - The market is showing signs of product homogenization, suggesting an impending consolidation phase as competition intensifies [57] Startup Strategies - Startups must leverage their agility and innovation to survive against established giants, focusing on strategic partnerships and long-term value creation [59]

Core Insights - The article discusses the advancements in robotics, particularly focusing on the emergence of foundational models in robotics, which are expected to revolutionize the industry by 2025 [6][23][35]. Group 1: Robotics Developments - Figure AI released its third-generation robot capable of performing various household tasks, but its success rate is questioned due to design issues [1]. - Tesla's robot has faced significant challenges in mass production, leading to a pause in production for hardware redesign [3]. - The article emphasizes the importance of foundational models in robotics, likening them to the capabilities of large language models [6][17]. Group 2: Historical Context of Robotics - The evolution of robotics is categorized into five generations, starting from programmed robots in the 1960s to the current vision-language-action (VLA) models [6][8][17]. - The first generation relied on strict programming, while the second introduced environmental perception through SLAM technology [9][11]. - The third generation utilized behavior cloning, allowing robots to learn from human demonstrations, but faced data efficiency issues [13][15]. Group 3: The Rise of VLA Models - The VLA model integrates vision, language, and action into a single neural network, enabling robots to understand complex instructions and perform tasks more efficiently [18][19]. - The emergence of VLA models is attributed to the maturity of large language models, which provide the necessary capabilities for understanding commands and reasoning [24][26]. - The article identifies three key factors contributing to the rise of foundational models in 2025: the maturity of large language models, reduced computing costs, and a mature hardware supply chain [27][31][33]. Group 4: Market Dynamics and Competition - The market for humanoid robots is projected to be massive, with estimates suggesting a $5 trillion market and the potential for one billion robots globally by 2025 [35]. - Dyna Robotics, a notable player in the field, has secured significant funding and aims to deploy robots in commercial settings, focusing on specific tasks like folding towels [37][56]. - The competition among robotics companies is categorized into three factions: full-stack integrators, vertical breakthrough specialists, and ecosystem platform developers, each with distinct strategies for achieving general-purpose robotics [41][72][81]. Group 5: Future Outlook - The article concludes that while impressive demonstrations have been made, the practical deployment of these technologies remains uncertain, with companies like Tesla and Figure AI still facing challenges in commercialization [82][85]. - The potential for household robots to assist with mundane tasks is highlighted as a near-future possibility, with companies aiming to introduce robots capable of performing specific functions in homes [85][86].