"杭州六小龙"杭州宇树科技有限公司（以下简称宇树科技）创始人王兴兴、浙江强脑科技有限公司（以 下简称强脑科技）创始人韩璧丞在被委任为香港特首顾问团成员后，近日首次在香港参加了特首顾问团 会议。 7月9日至11日，香港特区行政长官李家超一连三日与特首顾问团举行午餐会议，听取特首顾问团成员就 今年《施政报告》和香港整体发展的意见。 李家超特别表示，这是新一届（第二届）特首顾问团成员的首次会议。他欢迎特首顾问团三位新成员， 包括曾担任国际货币基金组织副总裁和中国人民银行副行长的朱民博士，以及"杭州六小龙"的两家企业 ——强脑科技创始人韩璧丞及宇树科技创始人王兴兴特意赴港参与会议。他们于会上表示香港具有得天 独厚的"内联外通"及教育科研优势，地理位置优越，亦是国际金融中心，资金自由流动，十分吸引内地 企业在港上市并以香港作为"出海"的窗口。 特首顾问团就三大主题进行了广泛讨论：一是香港经济高质量与持续发展——包括如何在地缘政治变化 和经济转型之中巩固香港国际金融、航运和贸易中心的地位等；二是创新与创业——包括如何推动传统 产业升级转型，积极培育新兴产业、加快发展北部都会区，以及积极吸引资金和人才助力本港的创新科 技发 ...

Group 1: Company Performance - The company expects to achieve approximately 2.633 billion yuan in revenue for the first half of 2025, representing a year-on-year growth of about 58.17% [1] - The anticipated net profit attributable to shareholders is between 1.1 billion and 1.2 billion yuan, reflecting a year-on-year increase of 85.50% to 102.36% [1] - The growth is primarily driven by a significant increase in the shipment volume of DDR5 memory interface and module supporting chips, along with a substantial rise in sales revenue from three high-performance transport chips totaling 294 million yuan [1] Group 2: Market Position and Future Outlook - According to Frost & Sullivan, the company has become the largest memory interconnect chip supplier globally, holding a 36.8% market share in 2024 [2] - The memory interconnect chip market is projected to grow from 1.2 billion dollars in 2024 to 5 billion dollars by 2030, with a compound annual growth rate (CAGR) of 27.4% [2] - The PCIe and CXL interconnect chip market is expected to expand from 2.3 billion dollars in 2024 to 9.5 billion dollars by 2030, with a CAGR of 26.7% [2] - The chairman of the company highlighted that the rapid advancement of artificial intelligence technologies is driving a profound transformation in the industry, leading to significant growth in the AI server market [2]

2025.07. 13 本文字数：2103，阅读时长大约3分钟 作者 | 第一财经 何涛 封图 | 李家超主持特首顾问团午餐会议，图为其中一场会议。来源：香港特别行政区新闻公报 上周，香港特区行政长官李家超一连三天（7月9日至11日）与特首顾问团举行午餐会，听取顾问团 成员就今年《施政报告》和香港整体发展的意见。 第一财经记者注意到，这是新一届（第二届）特首顾问团成员的首次会议。顾问团出现三位新成员， 分别为著名经济学家朱民、浙江强脑科技有限公司创始人韩璧丞及杭州宇树科技有限公司创始人王兴 兴——强脑科技和宇树科技都是"杭州六小龙"企业。 相应地，首届特首顾问团成员中的三人不再出现，分别是李嘉诚长子、长和集团主席李泽钜，华润集 团前董事长傅育宁，以及已经去世的商汤科技创始人汤晓鸥。顾问团继续保持34人阵容不变。 特首顾问团是李家超于2023年主导成立的高层次咨询组织，分为经济高质量与持续发展，创新与创 业，以及区域与环球协作三个小组。成员均为来自政、商、学、律界的翘楚，像香港特区前财政司司 长唐英年、太古集团主席白德利、阿里巴巴创始人之一蔡崇信、诺贝尔经济学奖得主迈克尔·斯彭斯 等。公开信息显示，首届顾问团至 ...

为香港股市"打广告"，吸引更多内地企业赴港上市及借道香港"出海"，算是李家超邀请"二小龙"的"附带收获" 上周，香港特区行政长官李家超一连三天（7月9日至11日）与特首顾问团举行午餐会，听取顾问团成员就今年《施政报告》和香港整体发展的意见。 第一财经记者注意到，这是新一届（第二届）特首顾问团成员的首次会议。顾问团出现三位新成员，分别为著名经济学家朱民、浙江强脑科技有限公司创始 人韩璧丞及杭州宇树科技有限公司创始人王兴兴——强脑科技和宇树科技都是"杭州六小龙"企业。 相应地，首届特首顾问团成员中的三人不再出现，分别是李嘉诚长子、长和集团主席李泽钜，华润集团前董事长傅育宁，以及已经去世的商汤科技创始人汤 晓鸥。顾问团继续保持34人阵容不变。 特首顾问团是李家超于2023年主导成立的高层次咨询组织，分为经济高质量与持续发展，创新与创业，以及区域与环球协作三个小组。成员均为来自政、 商、学、律界的翘楚，像香港特区前财政司司长唐英年、太古集团主席白德利、阿里巴巴创始人之一蔡崇信、诺贝尔经济学奖得主迈克尔·斯彭斯等。公开 信息显示，首届顾问团至少与李家超举行了三次正式会议。 李家超主持特首顾问团午餐会议，图为其中一场会议 ...

Core Viewpoint - The article emphasizes the importance of spatial reasoning in embodied AI and robotics, highlighting the limitations of existing 3D vision-language benchmarks and the need for a new standard that effectively evaluates spatial reasoning capabilities [3][4][5]. Group 1: Background and Limitations - Spatial reasoning is essential for intelligent agents to navigate and interact in real environments, requiring an understanding of 3D spatial layouts and context [3]. - Current 3D vision-language benchmarks fail to capture and assess spatial reasoning effectively, leading to models relying on semantic shortcuts rather than true spatial understanding [4]. - Three main limitations of existing benchmarks are identified: over-reliance on explicit queries, limited and shallow reasoning coverage, and template-driven or simplistic spatial queries [4]. Group 2: SURPRISE3D Dataset - SURPRISE3D is introduced as a new benchmark that combines linguistic intricacy with geometric complexity, featuring over 900 richly annotated indoor environments and more than 200,000 query-object mask pairs [5][6]. - The dataset's queries are designed to be implicit, ambiguous, and semantically lightweight, compelling models to rely on reasoning rather than recognition [5]. - Empirical evaluations show that even the most advanced existing 3D foundational models struggle on this dataset, indicating a significant innovation space for improving spatial reasoning capabilities [5][6]. Group 3: Query Types and Annotation Process - The dataset includes complex spatial queries that require various types of reasoning, such as narrative perspective, parametric perspective, relative position, and absolute distance [11][12]. - The annotation process involves dual workflows focusing on spatial reasoning and common-sense/human intention reasoning, ensuring a rich and complementary set of queries [16][18]. - Quality control measures include human verification and a multi-stage review process to ensure high-quality annotations [21][22]. Group 4: Experimental Results and Insights - Baseline models were evaluated for their effectiveness in spatial reasoning tasks, revealing that overall spatial reasoning capabilities are weaker than knowledge reasoning capabilities [26]. - After fine-tuning on the SURPRISE3D dataset, all models showed significant improvements in reasoning abilities, particularly in spatial reasoning, with average performance enhancements of approximately three times [28]. - The findings suggest that current methods have substantial room for improvement in spatial reasoning, highlighting important directions for future research [29].

Core Viewpoint - The article discusses the introduction of the Fast-in-Slow (FiS-VLA) model, a novel dual-system visual-language-action model that integrates high-frequency response and complex reasoning in robotic control, showcasing significant advancements in control frequency and task success rates [5][29]. Group 1: Model Overview - FiS-VLA combines a fast execution module with a pre-trained visual-language model (VLM), achieving a control frequency of up to 117.7Hz, which is significantly higher than existing mainstream solutions [5][25]. - The model employs a dual-system architecture inspired by Kahneman's dual-system theory, where System 1 focuses on rapid, intuitive decision-making, while System 2 handles slower, deeper reasoning [9][14]. Group 2: Architecture and Design - The architecture of FiS-VLA includes a visual encoder, a lightweight 3D tokenizer, and a large language model (LLaMA2-7B), with the last few layers of the transformer repurposed for the execution module [13]. - The model utilizes heterogeneous input modalities, with System 2 processing 2D images and language instructions, while System 1 requires real-time sensory inputs, including 2D images and 3D point cloud data [15]. Group 3: Performance and Testing - In simulation tests, FiS-VLA achieved an average success rate of 69% across various tasks, outperforming other models like CogACT and π0 [18]. - Real-world testing on robotic platforms showed success rates of 68% and 74% for different tasks, demonstrating superior performance in high-precision control scenarios [20]. - The model exhibited robust generalization capabilities, with a smaller accuracy decline when faced with unseen objects and varying environmental conditions compared to baseline models [23]. Group 4: Training and Optimization - FiS-VLA employs a dual-system collaborative training strategy, enhancing System 1's action generation through diffusion modeling while retaining System 2's reasoning capabilities [16]. - Ablation studies indicated that the optimal performance of System 1 occurs when sharing two transformer layers, and the best operational frequency ratio between the two systems is 1:4 [25]. Group 5: Future Prospects - The authors suggest that future enhancements could include dynamic adjustments to the shared structure and collaborative frequency strategies, which would further improve the model's adaptability and robustness in practical applications [29].

编辑丨王多鱼 排版丨水成文 生物医学领域的技术创新直接促进了生活质量的提高和健康寿命的延长。从历史上看，药物研发、外科技术、对生物通路的理解、成像技术以及其他领域的进步 推动了这一进程。如今，随着 人工智能 （AI） 的最新进展，生物医学领域即将迎来一个新的发展阶段。从技术角度来看，现代人工智能的进步得益于几个关键架 构创新，包括 Transformer 架构 、生成对抗网络和 diffusion 模 型，这些共同推动了越来越复杂的 生成式人工智能系统 （ generative AI ） 的开发。 2025 年 7 月 10 日，哈佛大学医学院、Scripps 研究所的研究人员在 Cell 期刊发表了题为： The generative era of medical AI 的综述论文。 大语言模型 （LLM） 、多模态人工智能、医疗实践的变革以及多尺度医疗预测具有变革性的潜力。这篇综述旨在总结过去三年中这些看似呈指数级增长的进展， 探讨这些新技术的背景、实施情况、影响以及一些持续存在的挑战。 多尺度医疗预测 ：AI 算法可用于医疗预测，基于各种动态输入来预测未来的医疗事件。这些算法可应用于多个层面，从分子层面 ...

点击下方 卡片 ，关注" 自动驾驶之心 "公众号 戳我-> 领取 自动驾驶近15个 方向 学习 路线 MCAM：面向自车层面驾驶视频理解的多模态因果分析模型 重庆大学&国防科技大ICCV25中稿的工作，本文提出 MCAM 模型，通过 DSDAG 因果图建模自车状态动 态演化，在BDD-X数据集上将驾驶行为描述任务BLEU-4提升至 35.7%，推理任务BLEU-4提升至 9.1%，显 著优于DriveGPT4等基线模型。 主要贡献： 算法框架： 实验结果： 论文标题：MCAM: Multimodal Causal Analysis Model for Ego-Vehicle-Level Driving Video Understanding 论文链接：https://arxiv.org/abs/2507.06072 代码：https://github.com/SixCorePeach/MCAM 1. 提出驾驶状态有向无环图（DSDAG），用于建模动态驾驶交互和状态转换，为因果分析模块（CAM） 提供结构化理论基础。 2. 提出多模态因果分析模型（MCAM），这是首个针对 ego-vehicle 级驾驶视频理解 ...

说到机器人导航，技术路线已经逐渐从早期传统的建图定位导航，到后期基于大模型方案的导航演变。而 基于大模型方案的导航又分为视觉语言导航和目标导航！ 如果说一句话说明这两个任务的区别，视觉语言导航是""听懂指令走对路"，目标导航是""看懂世界自己找 路"。 视觉语言导航是什么？ 点击下方 卡片 ，关注" 具身智能 之心 "公众号 与传统视觉语言导航（VLN）依赖显式指令不同，目标驱动导航系统需要实现从"听懂指令走对路"到"看懂 世界自己找路"的跃迁：当人类下达"去厨房拿可乐"的指令时，机器人需自主完成语义解析（识别厨房空间 特征与可乐视觉属性）、环境建模（构建家居场景的空间拓扑）以及动态决策（避开移动的人类或宠 物），这背后凝聚着计算机视觉、强化学习与3D语义理解的交叉突破。 商业落地与需求怎么样？ 视觉语言导航本质上是个指令跟随的任务。任务囊括了三个方面，理解语⾔指令、感知周围环境，规划运 动策略。一般来说，VLN机器人系统主要由视觉语言编码器，环境历史信息表征，以及动作策略三个模块 构成。 机器人从环境中获取语⾔指令和每⼀步的视觉观测，首先需要同时视觉语⾔编码器从中压缩出有效信息。 采用怎样的编码器，视觉和语 ...

Group 1 - The commercialization of humanoid robots is accelerating, with significant developments reported in the industry [1][2] - Chery and AiMOGA's collaboration on the Moja robot is set to launch in late September, targeting both dealers and individual consumers [1][2] - The recent procurement orders from China Mobile for humanoid robots worth 120 million yuan indicate a growing interest in deploying these robots in marketing scenarios [1][4] Group 2 - Chery has a long history in robotics, evolving from industrial robots to humanoid robots, with significant milestones including the launch of the CheryGPT language model [2][3] - The Moja robot, designed for automotive 4S stores, aims to enhance customer interaction by providing vehicle information and sales guidance [2][3] - The humanoid robots are designed with a focus on realism, featuring detailed human-like attributes [2] Group 3 - The humanoid robot market is divided into production robots for factories and service robots for customer interaction, with the latter requiring advanced emotional intelligence [3] - The effectiveness of humanoid robots in sales environments remains uncertain, as they may not yet possess the necessary skills to fully engage customers [3][5] - The recent large-scale orders for humanoid robots exceed industry expectations, suggesting a shift towards integrating these robots into specific marketing applications [5] Group 4 - The procurement project by China Mobile includes a total budget of 124.05 million yuan, with significant portions allocated for both full-size and small-size humanoid robots [4][5] - Other telecom companies, such as China Telecom, are also exploring humanoid robots for various applications, indicating a broader trend in the industry [5] Group 5 - The evolution of humanoid robots is expected to progress through three stages: developing leading technology, creating relevant content for specific industries, and ensuring a smooth user experience [6]