2025 年具身智能行业投融资热度持续攀升，行业内已形成明确共识：要实现从 L1 特定任务具身智能到 L2 组合任务具身智能的技术跨越，乃至向更高阶通用能力迈进，数据采集是必须突破的核心环节。 不同于语言、图像等低维数据的获取，具身智能需要的是物理世界绝对坐标系下的精确测量数据，其获取难 度、成本投入与标注周期，都远超传统模型训练需求。 具身智能的核心目标，是让机器人在物理世界中具备 "举一反三" 的常识理解能力，面对未曾接触过的物体和 任务，能像人类一样依托过往经验推导操作逻辑。而这一能力的构建基础，正是高质量、多模态的交互数据。 当前国内头部企业仍处于 L1 前期发展阶段，可完成特定环境下的单工位操作任务；而 π0.5 模型通过融合操 作数据、网络数据、语言指令等多源信息的预训练方式，在家庭真实环境长程任务中的准确率已突破 60%， 接近 L2 水平。 行业普遍认为，预训练是具身智能技术进阶的核心，而预训练效果直接取决于数据的 "量" 与 "质"：一方面， L1 级模型已需 1 万小时 + 的数据量支撑训练，且 Scaling Law 规模定律在具身智能领域仍未见顶，数据规 模的扩大仍能持续推动模型性能提 ...

Core Viewpoint - Linghou Robot has successfully completed over 100 million RMB in Series A financing, with funds primarily allocated for R&D, laboratory construction, and capacity expansion in the fields of industrial automation and general robotics [1][3]. Company Overview - Linghou Robot, established in 2015, originated as a research department of the listed company Bozhong Precision Engineering, focusing on the development of components for intelligent manufacturing equipment. It is now part of Jiangsu Bozhong Intelligent Technology Group and is recognized as a core supplier of intelligent manufacturing components [3][5]. - The company has been acknowledged as one of the first enterprises meeting the Ministry of Industry and Information Technology's standards for the industrial robot industry and has received provincial-level recognition as a specialized and innovative small and medium-sized enterprise [3]. Team and R&D - The core team includes founder and chairman Dong Hao, who drives technological innovation, and vice presidents Wang Bihan and Chen Dexi, who oversee production and international market expansion, respectively. The company has established R&D centers in Suzhou, Shenzhen, and Silicon Valley, employing over 400 staff, with more than 200 in R&D [5][8]. - Linghou Robot has developed various products, including machine vision systems, direct-drive motors, industrial robots, and mobile robots, which are widely used in industries such as 3C electronics, semiconductors, automotive, and new energy [5][6]. Product and Market Strategy - The company is advancing into the field of embodied intelligence, launching products like deep vision modules and wheeled robot chassis modules. It aims to provide scalable ODM solutions for global clients in the robotics sector [6][8]. - Linghou Robot has established a comprehensive management system, achieving ISO9001 certification and implementing an IPD process to enhance R&D efficiency based on market demand [7]. Future Outlook - With a sales network covering major cities in China and expanding into Southeast Asia, Europe, and North America, Linghou Robot has served over 800 clients. The company plans to focus on independent R&D of key components and strengthen its core competitiveness to become a world-class supplier in the intelligent manufacturing sector over the next 3 to 5 years [8].

Core Viewpoint - The integration of robotics and artificial intelligence in biopharmaceutical laboratories is crucial for enhancing drug development efficiency and addressing public health crises. The article emphasizes the importance of advanced robotics in automating complex biochemical processes while highlighting existing limitations in environmental perception and adaptability [1][2]. Group 1: Overview of Biochemical Laboratories - The development of biochemical laboratories has evolved through four stages: manual operation, early automation, comprehensive automation, and intelligent automation. Each stage has progressively improved operational efficiency and reduced human error [5][10]. - Current laboratories have begun to achieve a preliminary integration of automation and intelligence, but single-robot operations still face limitations in meeting diverse experimental needs [10]. Group 2: Key Technologies in Intelligent Biochemical Laboratories - Intelligent biochemistry laboratories leverage the deep integration of AI and robotics to enhance experimental processes, significantly improving efficiency and reducing human error [11]. - Environmental perception technology is critical for robots to accurately identify and understand dynamic changes in experimental settings, ensuring safe and precise operations [11][12]. - The combination of visual perception and motion planning technologies allows robots to efficiently execute tasks in complex environments, enhancing their adaptability and robustness [12][13]. Group 3: Environmental Perception Technologies - Environmental perception in intelligent biochemical laboratories involves recognizing and locating experimental tools and reagents, understanding object characteristics, and assessing spatial relationships [14][15]. - The perception technology is categorized into three levels: target recognition, object relationship recognition, and area relationship recognition, each contributing to the robot's operational capabilities [15][16]. - Single-modal data perception methods, such as visual and LiDAR sensors, are essential for accurate pose estimation of objects in the laboratory [18][22]. Group 4: Multi-Robot Task and Motion Planning Technologies - Multi-robot collaboration is essential for complex tasks in drug development, involving task decomposition, coalition formation, task allocation, and execution [45]. - Task planning methods include predefined task planning, reinforcement learning-based planning, and large language model-based planning, each offering unique advantages for dynamic environments [46][47]. - Motion planning techniques encompass traditional path planning algorithms, imitation learning, and embodied intelligence approaches, addressing various challenges in static and dynamic environments [57].

Core Viewpoint - The "fast-slow dual system" design in robotics is gaining attention as it allows robots to balance rapid responses to environmental changes with thoughtful, complex decision-making, inspired by human cognitive processes [1][3][4]. Group 1: Fast-Slow Dual System Concept - The fast system (System 1) operates instinctively and automatically, while the slow system (System 2) engages in deliberate and conscious thought, enabling efficient decision-making [1][4]. - This decoupling of fast and slow systems allows for independent upgrades of AI algorithms without altering the stable control framework, reducing development complexity [1][4]. Group 2: Application in Robotics - The fast-slow dual system can address the challenge of balancing generality and practicality in robotics, enabling quick visual motion strategies and end-to-end training for rapid communication and interaction [3][4]. - For example, in a task like "grabbing a red block," the slow system handles perception and planning, while the fast system calculates the necessary torque for smooth movement, allowing for immediate responses to disturbances [3][4]. Group 3: Company Implementations - Various companies are adopting the fast-slow dual system, each with unique implementations but sharing the core inspiration from human cognitive dual-process theory [4][5]. - Companies like Figure AI, PI, and智平方 are developing models that emphasize the separation of planning and execution, allowing for asynchronous parallel processing and improved task execution in complex environments [6][9][12][13]. Group 4: Data Strategies and Training Methods - Companies such as 魔法原子 and 微亿智造 focus on collecting vast amounts of real-world data to minimize the gap between simulation and reality, while others like PI utilize synthetic data for efficient model training [5][28]. - The approaches vary from general models seeking ultimate generalization to specialized models targeting specific scenarios for reliability and efficiency [5][28]. Group 5: Specific Company Models - Figure AI's Helix model integrates a dual system architecture for high-frequency control of the robot's upper body, achieving end-to-end visual-language-action capabilities [6][7]. - PI's Hi Robot system combines high-level semantic planning with low-level action execution, enabling complex task management in home environments [9][10]. - 智平方's GOVLA model enhances task understanding and execution through a three-part architecture, achieving coordination in open environments [12][13]. - 星海图's G0 model focuses on real-time control and task planning in complex scenarios, supported by a high-quality dataset [15][16]. - 擎朗's KOM2.0 model emphasizes service industry applications, utilizing a dual system for environment perception and action generation [18][19]. - 星动纪元's ERA-42 model integrates high-level planning with low-level control, enhancing execution capabilities in dynamic environments [21][22]. - 节卡's JAKA EVO platform employs a dual system for task parsing and execution, facilitating rapid adaptation to new industrial scenarios [25][26]. - 微亿智造's architecture leverages a cloud-edge-end model to ensure scalable applications in complex industrial settings [28][30]. - 魔法原子’s model simulates human cognitive processes to enhance real-time response and long-term task planning [30][31]. - 灵初智能's Psi-R1 model combines planning and execution through a layered architecture, achieving high-level task management and precise control [33][34].

Core Insights - A company in Hangzhou has improved the lifespan of humanoid robot lead screws to a usable level through cold heading technology, potentially reducing costs to a feasible range [1] - The industry is facing challenges with the precision and lifespan of lead screws used in humanoid robots, leading some companies to consider alternative designs [4][5] Group 1: Technology and Manufacturing - The reverse planetary ball screw structure is complex, requiring compact design and high precision, with most humanoid robots needing P5 precision, which corresponds to an ISO standard of C5 [2] - Cold heading technology has been introduced for micro planetary roller screws, achieving C3 precision and significantly higher efficiency (5 minutes per screw) compared to traditional methods [5][10] - The cold heading process reduces material waste and improves production efficiency, making it a promising solution for lowering costs in the production of micro lead screws [8][10] Group 2: Company Overview - New Coordinates, established in 2002 and listed on the Shanghai Stock Exchange in 2017, is a leader in precision cold forging and has expanded its business into automotive and robotics sectors [5][11] - The company has a strong global presence, with manufacturing bases in Europe and North America, and has shown robust revenue growth, with a CAGR of 16.0% from 2020 to 2024 [11] - New Coordinates has developed proprietary equipment for internal thread rolling and hot forming, achieving a material utilization rate of 91%, significantly higher than the industry average [10]

城市救援中，它们穿越浓烟为消防员探路，争分夺秒救人；农田上空，它们化身"空中医生"，精准喷洒农药， 大幅提升农业效率……微小型无人机在人们的生活、生产中已经扮演了重要角色。但你可曾发现，虽然这些无 人机可以单飞、可以群舞，却几乎没有两架无人机能像"叠罗汉"一样，一上一下叠飞？这是因为两架无人机交 叠时会产生强烈的"气流干扰"，极易出现"翻车"事故，更不用提精准作业。这个看似"简单"的动作，却是无人 机技术发展的一个卡点。 北京时间9月24日晚23时，Nature报道了西湖大学工学院赵世钰实验室在该领域实现的突破，他们研发了名 为FlyingToolbox（飞行工具箱）的空中协同操作系统，在国际上首次完成了多架旋翼无人机的空中工具交 换，实现了"叠式"飞行状态下的高精度协同作业，成功解决了近距离飞行与高精度操作不可兼得的关键技术难 题。 这是中国在多旋翼无人机领域的研究成果首次登上Nature。当无人机从"单兵作战"迈向"群体协作"，这些空中 作业机器人将在更高更远的应用场景中，进行人类难以完成的任务。 原文链接: https://www.nature.com/articles/s41586-025-09575 ...

Core Viewpoint - The article emphasizes the significant role of robotics in shaping the future of the nation, highlighting the contributions of professionals in the field during a time of national celebration [4]. Group 1 - The robotics industry is portrayed as a key player in national development, with professionals using technology to weave the fabric of a strong country [4]. - The article expresses gratitude towards robotics practitioners for their contributions to the nation, framing their work as a substantial gift to the country [4]. - There is a call for continued innovation and progress in robotics, likening it to the vibrancy and precision of national celebrations [4].

Core Insights - The article emphasizes that robots are poised to become a new entry point for trillion-level traffic, and the first company to bridge the gap between robots and traffic will define the next generation of business ecology [1][4][19] - The opening of the first open immersive intelligent experience center in Wuxi marks a significant step in commercializing embodied intelligence technology for consumer scenarios [1][4] Group 1: Experience Center Overview - The experience center spans 1,300 square meters and features eight thematic spaces that blend traditional culture with future technology, offering affordable ticket prices for consumers [3][4] - The center aims to collect real user feedback to enhance product iterations and ensure that advanced technology serves real products and scenarios [3][4] Group 2: Market Dynamics - The robot industry is shifting from hardware competition to ecosystem competition, with the consumer market representing a largely untapped billion-level market [4][6] - The core of the robot traffic economy is to create continuous interactions with consumers, transforming the experience center into a venue for traffic accumulation [6][7] Group 3: Technical Deployment - The experience center successfully deployed 40 robots within three days, showcasing the technical capabilities and operational efficiency of the robots [9][11] - The robots are designed for high autonomy, allowing for efficient management and operation with minimal personnel [11][13] Group 4: Consumer Engagement - Young consumers are willing to pay for new experiences and share them on social media, which enhances brand visibility and engagement [7][19] - The experience center is not just an exhibition but a platform for immersive interaction, allowing consumers to engage with robots in various activities [6][19] Group 5: Commercialization Pathways - The experience center serves as a showcase and sales platform for robots, reducing the cost for enterprise clients to evaluate robotic solutions [14][16] - Eight commercial pathways have been identified, including entertainment performances and data collection training, with a focus on leveraging user interaction data for algorithm improvement [16][19] Group 6: Future Outlook - The company plans to replicate the Wuxi model in other cities through partnerships with local governments, aiming for scalable expansion of the robot traffic economy [17][19] - The article concludes that the robot industry is in its early stages, with significant potential for growth as technology evolves and consumer engagement deepens [19][20]

在人与机器不断走近的今天， 人机交互却常 常面临这样的困境：当桌上放着多个 不同 杯子时，如果对设备 说 "把杯子给我"，它可能会随意拿起一个 （ 但不是那个最合适的 ） ，或者干脆无法响应。这类场景，正 反映出当前智能系统在理解模糊指令时的能力局限。 人类之间的交流，往往依赖语境与默契，大量表达在字面上并不完整。例如我们常说 "一会儿提醒我"（具体 什么时间？）、"把空调开大点"（开大多少？）、"找个最近的店"（依据步行还是驾车？）。 对人而言，这 类信息缺口很容易通过追问或推断来补全；但对机器而言，模糊即意味着无法执行。 传统自动化系统往往只 能响应明确、结构化的指令，而未来的智能体，必须学会在模糊中主动沟通，在协作中完成任务。 这一挑战也推动着智能体从 "被动执行者"向"主动协作者"的角色转变 。 具身智能体的发展目标，是成为人 类在物理世界中的合作伙伴，而非停留在脚本定式的工具。与虚拟环境中的 AI 不同，具身智能体必须应对现 实场景的开放性与不确定性。无论是在家庭、工厂还是公共空间中，它们都面临一个根本性的新课题：如何与 人进行高效、自然、双向的协同作业。 近年来，视觉 - 语言 - 动作模型（ V ...

Group 1 - The first open embodiment intelligent experience center by Zhiyuan Robotics has opened in Wuxi, focusing on humanoid robots and creating a unique consumer experience that combines traditional culture with future technology [1][3] - The experience center features eight themed spaces, including a performance stage and interactive projects, with daily shows in the robot theater [3] - Zhiyuan Robotics and Liangxi Cultural Tourism Development Group aim to develop more tech-driven cultural tourism products and explore new commercialization paths for embodied intelligence in the tourism sector [3] Group 2 - Linghou Robotics has successfully completed over 100 million yuan in Series A financing, led by prominent investment firms, which highlights the market's recognition of its self-research capabilities and future plans [4][6] - The funds from this financing round will primarily be used for R&D of core components in industrial automation and general robotics, as well as for laboratory construction and capacity expansion [6] Group 3 - Tesla plans to start mass production of its third-generation humanoid robot in 2026, with an expected annual production target of 1 million units by 2030 [7][9] - The company is actively expanding its humanoid robot production scale, indicating a strong commitment to this segment of the robotics market [9] Group 4 - The 2025 World Robot Contest has commenced in Mianyang, Sichuan, attracting over 1,600 teams and nearly 3,000 elite competitors, marking the first time the contest is held in Mianyang [10][12] - The contest aims to host over 500 events globally in 2025, with participation expected to exceed 350,000 individuals across all age groups [12] Group 5 - Contemporary Amperex Technology Co., Ltd. (CATL) has acquired a stake in Galaxy General Robotics, which focuses on the R&D and manufacturing of intelligent and industrial robots [13][15] - The increase in registered capital to 422,600 yuan will support Galaxy General Robotics in its research and market expansion in the field of general robotics [15]