【环球时报报道 记者 杨沙沙 涂蕾 张蔚蓝】编者的话：亚太经合组织（APEC）第三十二次领导人非正 式会议将于10月底在韩国举办。近日，《环球时报》中英文版联合韩国《每日经济新闻》记者，就两国 交流最热切的领域进行采访调查。此次中韩记者联合采访首先从机器人领域入手，通过多家机器人企业 的讲述，了解中韩产业最前沿的竞争以及合作态势。联合采访后，韩国《每日经济新闻》9日以"韩中机 器人合作进入'倒计时'"为题报道，重点关注了中国机器人企业今年年底将针对韩国市场的需求量产相 关机器人产品。 " 中国的企业，中国的技术已经发展到这个地步了！ " 韩国本土也不乏优秀的机器人企业。2020年12月，韩国现代汽车集团以8亿美元收购美国波士顿动力公 司80%股权，后者开发了四足机器人Spot以及人形机器人Atlas。三星在2023年斥资约4.32亿元人民币收 购韩国Rainbow Robotics公司14.7%股份。据李昌波介绍，韩国对于机器人产业抱有非常积极的态度。韩 国新一届政府上台后，总统李在明对于人工智能以及机器人产业非常重视，希望能将机器人尽快部署到 建设工地。在一些重大投资政策指挥棒下，韩国政府机关或企业对机器人 ...

Core Viewpoint - The development of humanoid robots requires them to perform multiple tasks, including manipulating various objects and maintaining balance in unexpected situations. Large Behavior Models (LBM) are crucial for cultivating these core capabilities in humanoid robots [1][2]. Group 1: Collaboration and Development - Boston Dynamics has partnered with Toyota Research Institute (TRI) to develop LBM for the Atlas humanoid robot, utilizing end-to-end language modulation strategies to assist in long-term manipulation tasks [2]. - The strategy consists of four processes: collecting behavior data through remote operation, processing and annotating the data, training neural network strategies, and evaluating these strategies with a testing suite [3]. Group 2: Strategy Principles - Boston Dynamics follows three core principles in strategy formulation: maximizing task coverage through a remote operation system, adopting a multi-task training policy for better generalization, and building a robust infrastructure for rapid iteration and scientific rigor [5][9]. Group 3: Hardware and Software Configuration - The Atlas robot has 78 degrees of freedom (DoF) for extensive movement and flexibility, while the Atlas MTS focuses on pure manipulation tasks with 29 DoF [9]. - The remote operation system uses HDR stereo cameras for situational awareness and integrates with a model predictive controller (MPC) to ensure precise operations while maintaining balance [9][10]. Group 4: LBM Technology and Simulation - The LBM architecture is based on TRI's LBM, utilizing a diffusion transformer with 450 million parameters to predict actions based on sensory input and language prompts [11]. - Simulation technology plays a key role in development, allowing for efficient training and evaluation while sharing data pipelines and reducing costs [11]. Group 5: Enhanced Capabilities - Through LBM training, Atlas has surpassed traditional robotic limitations, enabling it to autonomously respond to unexpected situations and perform complex long-range tasks [12][14]. - The robot can execute a variety of tasks, from simple pick-and-place actions to complex operations like manipulating a 22-pound (9.9 kg) tire, showcasing the advantages of "learning by demonstration" [16]. Group 6: Future Directions - Boston Dynamics aims to expand its "data flywheel" to improve throughput, quality, task diversity, and difficulty while exploring new algorithmic concepts [19].

Core Viewpoint - WIRobotics has made significant advancements in humanoid robotics with the launch of ALLEX, a versatile humanoid robot capable of human-like responses to physical stimuli, marking a breakthrough in force perception and interaction technology [1][3][4]. Group 1: ALLEX Robot Features - ALLEX features 15 joints and a precision of 0.3 mm, allowing it to perform complex tasks and respond to physical stimuli without the need for tactile sensors [4][6]. - The robot's high-degree-of-freedom compliant hand can detect forces as low as 100 grams, showcasing its human-like adaptability and control performance [6][8]. - WIRobotics aims to establish ALLEX as a modular platform, facilitating its application in various sectors such as service, manufacturing, and home assistance [10]. Group 2: Company Background and Products - WIRobotics, founded in 2021, also specializes in exoskeleton technology, with its lightweight wearable robot WIM gaining market traction since its launch in April 2024, achieving sales of approximately 500 units within eight months [12][14]. - The company has secured significant funding, including a 13 billion KRW Series A round in March 2024, and has raised a total of $3.11 million through four funding rounds [14]. Group 3: Competitive Landscape - The humanoid robotics market in South Korea is competitive, with notable products like AMBIDEX, RB-Y1, and Atlas gaining attention for their commercial applications [15][17]. - AMBIDEX, developed by NAVER, is utilized in Starbucks for precise ingredient dispensing and operational assistance, while RB-Y1, created by Rainbow Robotics, excels in industrial manufacturing and logistics [17][19]. - Atlas, developed by Boston Dynamics, is still in training but shows promise in performing tasks in confined spaces, contributing to the automotive assembly process [19][21].

Group 1 - Robocore Technology, a Hong Kong robotics startup, recently secured a $10 million investment from Foxconn Technology Group, marking the first tranche of its Series D funding, with options for two additional $10 million investments in 2026 and 2027, giving Foxconn a 6.6% stake in Robocore [2] - Robocore was founded in 2018 by engineer Roy Lim and initially served as the exclusive distributor of Temi wheeled robots in Hong Kong, acquiring the Israeli company last year [2] - Foxconn is reportedly in talks with Nvidia to deploy humanoid robots in a Houston factory for the production of AI servers, and has previously collaborated with Shenzhen-based UBTECH Robotics to deploy humanoid robots in its factories [2] Group 2 - Boston Dynamics has trained a new large behavior model (LBMs) for its Atlas humanoid robot, enabling it to perform complex tasks requiring long-term planning, utilizing its unique capabilities such as walking, precise foot placement, and avoiding self-collision [5] - The LBM model converts inputs like images, proprioception, and language prompts into action commands for the Atlas robot, operating at a frequency of 30Hz [5] Group 3 - UBTECH Robotics has led the establishment of two national technical standards for humanoid robots, focusing on positioning and navigation, and human-robot interaction, aiming to support the development and standardization of humanoid robot technology [6] - The company has previously spearheaded the first batch of national standards for humanoid robot operational requirements and plans to continue its involvement in industry standardization efforts [6] Group 4 - Shengshi Technology announced plans to establish a wholly-owned subsidiary, Shenzhen Feifeite Cultural Innovation Co., Ltd., with an investment of 10 million yuan, aiming to combine AI and humanoid robot technology with innovative products for emotional interaction [9] - The company has acquired core assets from Aldebaran Robotics, focusing on educational and commercial robots to diversify its business [9] Group 5 - Grebo signed a cooperation agreement with Zhejiang Zhiding Robotics at the inaugural partner conference of Zhiyuan Robotics, marking a significant step in expanding its presence in the embodied intelligence sector [12] - The collaboration aims to enhance the global market reach of commercial cleaning robots, leveraging Grebo's overseas channel resources to increase product market share [12] - This partnership is seen as a key component of Grebo's strategic layout in the embodied intelligence field, promoting resource complementarity and collaborative development [12]

Group 1 - Google launched the Pixel 10 series with four models, featuring the Tensor G5 chip and Gemini Nano model, emphasizing deep AI integration as a hallmark characteristic [1] - The new models include various AI functionalities such as Gemini Live voice assistant, Voice Translate for real-time speech translation, Nano Banana photo editor, and Camera Coach for photography guidance [1] - Pro Res Zoom supports up to 100x smart zoom, and Magic Cue intelligently extracts content from Gmail and calendar, marking the end of the traditional smartphone era according to Google [1] Group 2 - DeepSeek officially released the V3.1 model, utilizing a hybrid reasoning architecture that significantly enhances both thinking efficiency and agent capabilities [2] - The new model shows notable improvements in programming agent assessments and search agent evaluations, while reducing output tokens by 20%-50% without compromising performance [2] - The model is fully open-source, employing UE8M0 FP8 Scale parameter precision, with API upgrades supporting Anthropic API format and extending context to 128K [2] Group 3 - ByteDance's Seed team open-sourced three models: Seed-OSS-36B-Base (with and without synthetic data) and Seed-OSS-36B-Instruct [3] - The models were trained on 12 trillion tokens and are licensed under Apache-2.0, supporting a 512K ultra-long context window and flexible reasoning budget control [3] - The Instruct version achieved new state-of-the-art records in various open-source benchmark tests, particularly in MMLU-Pro, MATH, and AIME24 [3] Group 4 - The University of Hong Kong and Kuaishou's Keling team introduced Context as Memory technology, achieving long-term scene memory retention in video generation, comparable to Google's Genie 3 and released earlier [4] - This innovative technology uses historical generated context as "memory" and designs a memory retrieval mechanism based on camera trajectory, significantly enhancing computational efficiency [4] - Research indicates that video generation models can implicitly learn 3D priors without explicit 3D modeling, maintaining static scene memory within seconds [4] Group 5 - Baidu released the MuseSteamer video model 2.0, utilizing integrated Chinese audio-video generation technology to address the unnatural dialogue issue in AI video generation [5] - The new model offers four versions (turbo, pro, lite, and voiced), accurately matching Chinese lip movements, supporting emotional expression and dialects, and enabling static photos to speak [6] - This technology synchronizes sound and visuals during conception, eliminating the need for post-production matching, and employs a "multi-modal latent space planner" to significantly reduce video production costs and complexity [6] Group 6 - Tencent's Yuanbao integrated Tencent Video functionality, allowing users to view videos directly from search results during conversations with Yuanbao [7] - Users can search for films by title, receive personalized recommendations based on scene descriptions, and retrieve films they can't remember by vague memories [7] - In addition to searching and recommending, Yuanbao can engage users in discussions about film creation backgrounds, plot meanings, and genre styles, with direct links to watch related works [7] Group 7 - Boston Dynamics showcased a new video of the Atlas humanoid robot, demonstrating evolution based on the latest large behavior models (LBMs) for precise control in multi-tasking and language-driven operations [8] - The system consists of four components: collecting embodied behavior data through remote control, processing labeled data, training a unified neural network policy model, and evaluating the policy model through testing tasks [8] - The Atlas robot can now smoothly perform "repair station" tasks, including complex movement operations, dexterous grasping, and secondary gripping, intelligently responding to unexpected situations, advancing general AI robotics [8] Group 8 - OpenAI researchers stated that GPT-5's behavior design intentionally addresses "flattery issues," aiming to balance interactivity with healthy assistant attributes, with significant improvements in creative writing and programming capabilities [9] - As evaluation benchmarks become saturated, the future differentiation of models will primarily depend on actual use cases, with the team designing internal assessments based on real-world needs [9] - OpenAI's agent development strategy has evolved from ChatGPT to Deep Research and more complete functional agents, aiming to build systems capable of asynchronous task execution and maintaining cross-platform memory over time [9] Group 9 - Index Ventures' investment director emphasized that founder traits are more important than market size, as exceptional founders can expand small markets, as demonstrated by Adyen and Figma [10] - There are notable differences between American and European founders: American founders tend to have more global ambitions and fundraising capabilities, while European founders are more pragmatic but often limited by market fragmentation and insufficient capital [10] - For Europe to produce global AI giants, three core issues must be addressed: increasing capital density, accelerating market integration, and improving talent systems to retain top researchers and entrepreneurs [10]

Core Insights - The humanoid robot market is on the brink of explosive growth, with major tech companies vying for ecosystem dominance, as evidenced by recent open-source model releases from Apple, Meta, Google, and Huawei [1][2] - China's government support for the humanoid robot industry has reached unprecedented levels, with investment funds totaling approximately 187 billion RMB [1][6] - Morgan Stanley projects that the global humanoid robot market could reach annual revenues of $5 trillion by 2050, with cumulative adoption expected to hit 1 billion units [1][8][9] Group 1: Market Dynamics - Major tech companies are accelerating their open-source strategies in the robotics field to bind developers to their ecosystems [2][3] - The competition for ecosystem control is intensifying, with companies diversifying AI investments into hardware [3] - The "Humanoid Robot 100" index has risen by 14.4% this year, outperforming the S&P 500 by approximately 11 percentage points, highlighting strong market interest [1] Group 2: Government Support and Investment - The Chinese government has established substantial industry funds, including a 10 billion RMB fund in Wuhan, to support the humanoid robot value chain [6][4] - Recent financing activities in China's humanoid robot sector reached a record 25 transactions in May, indicating robust investor interest [6] Group 3: Technological Advancements - Major companies have released significant AI models and platforms aimed at enhancing humanoid robot capabilities, such as Meta's V-JEPA 2 and Google's Gemini Robotics On-Device [5] - The integration of AI into physical robotics is seen as a critical factor for success, with data acquisition being a key challenge [7] Group 4: Commercialization and Future Projections - Commercial deployment of humanoid robots is accelerating, with companies like Foxconn and Amazon planning to implement robots in manufacturing and logistics [8] - By 2036, approximately 23.7 million humanoid robots are expected to be adopted, with the average price in high-income countries projected to drop from $200,000 in 2024 to $50,000 by 2040 [9]

Market News - US stock indices collectively declined, with the Dow Jones down 0.95%, Nasdaq down 0.87%, and S&P 500 down 0.77%. Major tech stocks mostly fell, with Meta down over 2%, Tesla and Intel down over 1%, while Netflix saw a slight increase [1] - International oil prices surged, with WTI crude oil up 3.27% at $59.00 per barrel and Brent crude oil up 2.99% at $62.03 per barrel. Gold prices also rose significantly, with spot gold up 2.95% at $3431.11 per ounce [1] Industry Insights - Hyundai announced the deployment of Boston Dynamics' Atlas robots in its new factory in Georgia, aiming to produce 300,000 electric and hybrid vehicles annually. This marks a significant breakthrough in manufacturing automation [2] - Huawei is set to unveil its HarmonyOS PC on May 8, which is the first domestically produced computer restructured from the operating system kernel. The success of HarmonyOS PC will depend on building a robust ecosystem [3] - The global DRAM market is expected to grow by 42.5% year-on-year to $26.729 billion in Q1 2025, driven by AI server demand and recovery in PC and mobile sectors. SK Hynix has overtaken Samsung to become the market leader with a 36.7% share [4] - The launch of the "Benyuan Tianji 4.0" quantum computing measurement and control system by Benyuan Quantum Computing Technology marks a significant step towards the industrialization of quantum computing in China, with expectations of reaching a market size of hundreds of billions by 2025-2030 [5][6]