1月12日，具身智能公司自变量机器人宣布近期完成10亿元A++轮融资。由字节跳动、红杉中国、北京 信息产业发展基金、深创投、南山战新投、锡创投等投资机构及地方平台联合投资。 目前，自变量已先后获得美团、阿里和字节跳动三家互联网大厂的押注。此前2025年9月初的A+轮投资 中，阿里云领投了自变量的近10亿元融资。自变量2025年5月完成的A轮融资中，则是由美团战投领 投、美团龙珠跟投。 和赛道上一些人形机器人硬件公司不同，自变量靠具身智能模型起家。公司方面在不同场合提出，具身 智能基础模型是独立于大语言模型、多模态模型等虚拟世界基础模型的物理世界的基础模型。其核心是 让机器人能够具备实时处理非结构化、动态及随机任务的能力。 在解决具身智能模型数据瓶颈问题上，自变量大规模依赖真机强化学习。该公司创始人兼CEO王潜在 2025年11月底的一场论坛上表示，外界对"数据规模"可能存在一些误解，第一反应往往是"大力出奇 迹"。但单纯堆数据规模未必能带来理想结果，只有更高质量的数据才能实现突破。数据质量优先于数 据总量，这也是为什么自变量坚持以物理世界真实数据为主的原因。 据公司方面介绍，自变量自研了遥操、外骨骼、无本体等 ...

Core Viewpoint - The article discusses the debate on whether embodied intelligence should be viewed as an application or as an independent foundational model, asserting that it is a foundational model specifically designed for the physical world, parallel to language and multimodal models [6][12][60]. Group 1: Differences Between Physical and Virtual Worlds - There is a fundamental difference between the physical world, characterized by randomness and continuous processes, and the virtual world, which is highly reproducible and low in randomness [2][10]. - Existing models based on language and visual modalities are inadequate for accurately representing the complexities and randomness of physical interactions [16][22]. Group 2: Need for a Separate Foundational Model - A separate foundational model for embodied intelligence is necessary due to the unique characteristics of the physical world, which often leads to unpredictable outcomes even under identical conditions [10][11]. - The current architectures and training methods struggle to capture the high randomness present in physical events, necessitating a new approach to model design [12][20]. Group 3: Future of Multimodal Models - Shifting the perspective to view embodied intelligence as an independent foundational model can lead to significant changes in model architecture and data utilization [9][23]. - The learning and perception processes in the physical world differ fundamentally from those in the virtual world, suggesting that future multimodal models should incorporate these differences [24][29]. Group 4: Scaling Laws and Data Utilization - The article emphasizes the importance of scaling laws in the development of large models, particularly in the context of robotics, where data acquisition and utilization are critical [46][51]. - A phased approach to training, utilizing both pre-training and post-training data, is recommended to enhance model performance [48][52]. Group 5: Hardware and AI Integration - The integration of AI in defining hardware is crucial for the development of embodied intelligence, advocating for a simultaneous evolution of both software and hardware [53][54]. - The potential for embodied intelligence to drive exponential growth in resources and capabilities is highlighted, suggesting a transformative impact on the future of artificial general intelligence (AGI) [59][60].

Core Viewpoint - The embodiment intelligence model is considered an independent foundational model parallel to language and multimodal models, specifically designed for the physical world [6][12][61] Group 1: Differences Between Physical and Virtual Worlds - The fundamental differences between the physical and virtual worlds are recognized, with the physical world characterized by continuity, randomness, and processes related to force, contact, and timing [2][10] - Existing models based on language and visual paradigms are structurally misaligned with the complexities of the physical world [3][21] Group 2: Need for a Separate Foundational Model - A separate foundational model is necessary due to the significant randomness in the physical world, which existing models struggle to accurately represent [10][17] - The current reliance on multimodal models for embodiment intelligence is seen as inadequate, necessitating a complete rethinking of model architecture and training methods [9][21] Group 3: Future of Multimodal Models - Shifting perspectives on embodiment intelligence will lead to new insights in model architecture and data utilization [24][30] - The learning processes in the physical world differ fundamentally from those in the virtual world, suggesting that future multimodal models must adapt to these differences [25][28] Group 4: Scaling Laws and Data Utilization - The concept of Scaling Law is crucial in the development of large models, particularly in robotics, where data sourcing remains a significant challenge [47][49] - A phased approach to training and data collection is recommended, emphasizing the importance of real-world data for effective learning [52][53] Group 5: Hardware and AI Integration - A new learning paradigm necessitates the redesign of hardware in the physical world, advocating for AI to define hardware rather than the other way around [54][55] - The potential for embodiment intelligence to drive exponential growth in resources and capabilities is highlighted, drawing parallels to historical industrial advancements [60][61]

Core Viewpoint - The roundtable forum highlighted the importance of funding and data in advancing embodied intelligence, with participants discussing various strategies for utilizing a hypothetical budget of 10 billion yuan to drive development in the field [1][53]. Group 1: Funding and Investment Strategies - Participants expressed differing opinions on how to allocate 10 billion yuan for the advancement of embodied intelligence, with suggestions including investing in research institutions and building data engines [1][54][56]. - The CEO of Accelerated Evolution emphasized the need for collaboration, suggesting that 10 billion yuan may not be sufficient without partnerships [1][53]. - The focus on creating the largest self-evolving data flywheel was proposed as a key investment area [54]. Group 2: Data Challenges and Solutions - A significant discussion point was the scarcity of data, with varying opinions on the importance of real-world data versus synthetic data [2][29]. - The emphasis was placed on the necessity of high-quality, diverse data collected from real-world scenarios to enhance model training [30][32][36]. - The use of simulation data was also highlighted as a means to accelerate the development of embodied intelligence before sufficient real-world data can be gathered [43][44]. Group 3: World Models and Predictive Capabilities - The forum participants agreed on the critical role of world models in embodied intelligence, particularly in enabling robots to predict and plan actions based on future goals [5][12]. - There was a consensus that training data for these models should primarily come from the robots themselves to ensure relevance and effectiveness [5][12]. - The discussion included the potential for a unified architecture in embodied intelligence models, contrasting with the current fragmented approaches [7][15][27]. Group 4: First Principles and Decision-Making - Participants shared their foundational principles guiding decision-making in the development of embodied intelligence, emphasizing the importance of data scale and quality [48][49][51]. - The need for a physical world foundation model that accurately represents complex physical interactions was highlighted as essential for future advancements [26][27]. - The concept of a closed-loop model for embodied intelligence was proposed, contrasting with the open-loop nature of current language models [10][11].

Core Insights - The establishment of humanoid robot innovation centers in Zhejiang and Hubei since 2023 aims to foster communication between research outcomes and industry needs, promoting co-creation in the humanoid robot ecosystem [1][2][4] - The humanoid robot industry is transitioning into a new phase in 2024, characterized by refined algorithms and initial systematic construction, laying a solid foundation for industrial application [1][7] - The focus is on developing practical products that can be promoted, which will encourage upstream companies to participate in product refinement [4] Group 1: Industry Development - The Zhejiang humanoid robot innovation center has initiated pilot projects in the textile industry, demonstrating the practical application of robots in various manufacturing scenarios [1][2] - The center is addressing the challenges of robot adaptability and dependency on human operators by creating a general algorithm library and software for quick deployment [3][4] - The integration of data from real and simulated environments is crucial for enhancing the capabilities of humanoid robots, with a recommended data ratio of 1:9 for effective training [6] Group 2: Future Prospects - The humanoid robot industry is witnessing cross-industry collaboration, with companies from automotive and mobile sectors entering the humanoid robot space, which may lead to technological synergies [7] - The humanoid intelligent model is expected to mature over the next two to three years, although specific advancements remain unpredictable [5][6] - The development of general humanoid robots or general embodied intelligent models is anticipated, driven by existing technology and future demand [7]

Group 1 - The year 2025 is referred to as the "Year of Robot Commercialization" or the "Year of Robot Cash Cows," with significant investment in the embodied intelligence sector in China, totaling 11.037 billion yuan in the first half of the year, surpassing the total for 2024 [1] - The application side of the industry has seen a staggering 17-fold year-on-year increase in transaction volume, indicating a trend of explosive growth [1] - The industry has transitioned from a "sci-fi concept" phase to a period of realizing large-scale commercial value, driven by a technology monetization loop of "data collection - model training - commercial transformation" [1] Group 2 - The development of large models has given robots a "human touch," with natural language understanding accuracy reaching 92.3%, approaching human levels [3] - Traditional industrial robots have limitations in adaptability and task flexibility, while breakthroughs in embodied intelligence systems enable robots to achieve human-like environmental understanding through deep reinforcement learning [3][4] - The transition from "program-driven" to "cognition-driven" robots is accelerated by advancements in technology that allow for multi-modal task execution [3] Group 3 - High-quality, large-scale datasets are crucial for building accurate cognitive models for robots, requiring significant investment in data collection and annotation [4] - Many companies, including traditional industries, are participating in the data annotation sector, providing customized services for various AI and robotics applications [4] Group 4 - Self-learning and self-correction are key for achieving higher levels of robot intelligence, allowing models to quickly adapt to new tasks based on past experiences [6] - The transparency and open-source nature of models facilitate performance optimization and accelerate model iteration through collective developer contributions [6] Group 5 - The use of edge large models allows robots to perform critical computations locally, enhancing real-time decision-making capabilities even in poor network conditions [7] - The global humanoid robot market is projected to exceed $5 trillion by 2050, with most applications in industrial and commercial sectors, while only about 10% are expected to enter household environments [9] - The shift from traditional automotive industries to the rapidly growing robotics sector indicates a significant transformation in market dynamics, with China expected to maintain a 5%-10% annual growth rate in the service robot field [9]

Core Viewpoint - The article discusses the emergence of embodied intelligence in China, highlighting the rapid growth and investment in the sector, particularly focusing on the company "Self-Variable Robotics" founded by Wang Qian, which has raised over 1 billion yuan in funding within a year and a half [5][12]. Group 1: Company Overview - Wang Qian, the founder of Self-Variable Robotics, has a strong academic background and prior experience in the U.S. quant fund industry, which he left to pursue robotics [2][5]. - Since its establishment in 2023, Self-Variable Robotics has completed seven rounds of financing, with a total amount exceeding 1 billion yuan [5]. - The company has adopted an "end-to-end unified VLA model" technology route, updating its model every 2-3 months [7][12]. Group 2: Industry Context - 2023 is marked as a significant year for the domestic embodied intelligence sector, with major players like Nvidia's founder predicting it as the next tech wave [5]. - The domestic humanoid robotics startup landscape has formed a clear hierarchy, with Self-Variable Robotics moving from a secondary to a quasi-first-tier position due to its funding achievements [5]. - There are contrasting views on the commercial viability of humanoid robots, with some investors skeptical about their practical applications, while others continue to invest heavily [5][10]. Group 3: Technological Development - Self-Variable Robotics has developed the WALL-A model capable of performing complex tasks beyond simple operations, positioning itself at the forefront of the industry [8][12]. - Wang Qian anticipates that a GPT-3 level embodied intelligence model could emerge within a year, with commercial applications expected to materialize in one to two years [10][21]. - The company prioritizes enhancing model capabilities over immediate commercialization, with two-thirds of its expenditures directed towards model development [12][30]. Group 4: Market and Commercialization - Current commercial applications for embodied robots are primarily in research education and hospitality, which Wang Qian believes are not the ultimate target markets for long-term growth [10][31]. - The company has already developed a physical product, although it has not yet been widely released, and is currently in the proof of concept stage with seed customers [27][29]. - Wang Qian expresses skepticism about the long-term value of current commercial scenarios, suggesting they may be more about meeting investor expectations than achieving substantial market impact [31][32]. Group 5: Competitive Landscape - The article notes that while domestic companies are catching up, there remains a significant gap between Chinese and U.S. companies in terms of overall capabilities [37]. - Self-Variable Robotics claims to be on par with international leaders like Physical Intelligence and Google in certain aspects, despite the general perception of being behind [38]. - The challenges of open-source models in the embodied intelligence space are highlighted, with Wang Qian arguing that commercial success cannot rely solely on open-source strategies [43][44].

2025年会是AI应用全面落地的大年 近两年 人工智能市场最热闹的是 大模型领域， 我们 已投资 了 14 家 大语言模型、多模态模型 、 具身智能 模型或端到端智驾模型的领军企业 ，这个数量在亚洲位居前列。同时我们 协助 管理着规 模达 100亿 元 的 北京市人工智能产业投资基金。 这些 都是 "触点"，为 我们 判断 AI行业的发展 脉络 提供了 更多的数据，能够 更好地训练我们的投资 思维模 型 。 任何一轮科技浪潮，都开始于底层基础技术的耕耘。 本文为IPO早知道原创 作者｜ Stone Jin 过去几年，启明创投 一直把 AI的投资分成三个层次 ： 微信公众号｜ipozaozhidao 据 IPO早知道消息， 启明创投主管合伙人周志峰 日前 发表了题为 "2025，AI照进现实之旅"的主旨 演讲，分享了对AI投资的见解，和对AI市场演进路径的推演与预判。 以下系演讲精选： 为什么不是去年 或 前年？ 原因是 任何 一轮科技 浪潮 ，都开始于底层基础技术的耕耘，其中有两个核心技术指标，一是性 能，从凑合用到真正好用，二是成本，从 "高不可攀"到"轻松消费"，当这两个核心指标均达到临界 点时，应用就会 ...