⚡️ 假期充电系列继续 今天为大家整理 2025 年 CoRL 期间举办的首届 "Making Sense of Data in Robotics" Workshop，一起探究： 在机器人技术飞速发展的今天，人 们常常把目光聚焦在算法与模型上，是否忽视了真正决定"能否走出实验室、实现大规模落地"的底层变量——数据。 数据不仅是训练基础模型的燃料，更是 支撑策略泛化、稳定运行与安全可控 的地基。没有高质量、场景匹配的数据，再先进的模型也只能停留在论文 与Demo里。 此次Workshop正是一次针对这一"被低估的核心要素"的集体深思。会议聚焦于数据构成、数据筛选与数据可解释性三大命题，试图回答机器人行业最 迫切的问题： 1. 机器人真正需要什么样的数据？ 2. 如何从海量原始信息中提炼出能提升策略表现的数据？ 3. 又该如何理解数据对机器人决策与行为的实际影响？ 锦秋基金（公众号：锦秋集，ID：jqcapital）认为， 这场 Workshop 的价值不只是学术交流，而是揭示了实体智能走向产业化过程中的"关键一 环"。 无论是 Joseph Lim 团队提出的"任务拆解 + 模块复用"式数据高效利用，还是 Ke ...

Investment Rating - The report does not explicitly state an investment rating for the industry or specific companies involved in the Physical AI sector [4]. Core Insights - The collaboration between NVIDIA and Alibaba Cloud outlines a three-in-one implementation roadmap for Physical AI, integrating cloud-based training, virtual simulation, and edge deployment, which is expected to enhance automation across various industries [1][13]. - The effectiveness of the Cosmos/simulation technology relies heavily on multi-level calibration and robust data lineage management to minimize Sim2Real gaps, which are critical for achieving real-world success [2][14]. - A disciplined pilot cadence is recommended to avoid the "great demo, hard deployment" trap, emphasizing a structured four-gate process for engineering rollout [3][15]. - Optimizing inference economics and clarifying the roles of cloud and edge computing are essential for scaling applications in the Physical AI sector [3][16]. - Governance, organization, and supply chain resilience are identified as foundational elements for the successful implementation of Physical AI technologies [3][17]. Summary by Sections Event Overview - On September 25, 2025, NVIDIA and Alibaba Cloud presented a roadmap for Physical AI at the Apsara Conference, focusing on the integration of cloud training, virtual simulation, and edge deployment [1][13]. Technical Implementation - The proposed framework utilizes the Omniverse simulation platform and Cosmos world model, aiming to reduce reliance on real-world data and facilitate automation in manufacturing and logistics [1][13]. - A three-layer calibration mechanism is essential for ensuring data accuracy and effectiveness in simulation technologies [2][14]. Engineering and Deployment - A structured approach to deployment is recommended, involving a four-gate process to manage risks effectively [3][15]. - Key performance indicators (KPIs) should be established at various levels to monitor progress and ensure alignment between simulation and real-world applications [2][15]. Economic and Organizational Considerations - The report emphasizes the importance of optimizing costs and defining clear roles for cloud and edge computing to enhance operational efficiency [3][16]. - Building a resilient supply chain and governance framework is crucial for the long-term success of Physical AI technologies [3][17].

以下文章来源于暗涌Waves ，作者暗涌 暗涌Waves . 钱的流向，人的沉浮。36氪旗下投资报道账号。 AI能把人与人连接起来吗？ 文 ｜ 施嘉翔 编辑 ｜ 陈之琰 来源｜ 暗涌Waves（ID:waves36kr） 封面来源 ｜ IC Photo "暗涌Waves"获悉，AI初创公司指数引力已于年初完成pre-A轮融资， 由红杉中国种子基金和阿尔法公社联合领投。 于北川 ：看起来跨度大，但对我们来说是同一条路径。我们最初做AI influencer marketing，因为这是最集中、最紧迫的"找人"场 景：品牌要找达人合作，需求清晰、频率高、痛点深。 做着做着我们发现，本质上我们解决的不是"达人营销"，而是"找对的人"——无论是达人、客户、专家还是合作伙伴。我们就把 能力泛化，让AI能处理所有商业关系场景。 指数引力创始人于北川是抖音早期核心成员，历经抖音从几千万用户到6个亿的日活全过程，负责抖音早期社交关系的构建。 2022年年初，于北川选择创业。这场创业的开始是一家海外电商公司。于北川卖过穿戴甲、吸尘器，还有仅售20美元的无人机。 2023年5月，他们因TikTok店铺关闭导致资金链断裂，不得不停 ...

出品｜虎嗅科技组 作者｜李一飞 编辑｜陈伊凡 头图｜Clay领英主页 "AI 原生 100" 是虎嗅科技组推出针对 AI 原生创新栏目，这是本系列的第「 21 」篇文章。 18个月，估值飙到 100 亿美元，到账 6.35 亿美元现金，年经常性收入逼近 1 亿美元——放在任何时 代都是"火箭"，在 AI 赛道也属罕见。 即便是在快速发展的AI创业时代，也很少见。 9 月，全球知名互联网投资公司Greenoaks Capital 又添一把火：领投 3.5 亿美元，让 Sierra 正式跻 身"百亿美金俱乐部"。 这家由前 Salesforce 联席 CEO Bret Taylor 与前谷歌高管 Clay Bavor 联手创办的 AI 客服公司，只 做一件事：用生成式 AI 替企业"包办"客户体验。成立伊始，它就按下快进键：产品上线、拿下大客 户、数据反哺模型、体验更优，飞轮越转越快。 "需求爆了。"嘉和资本 CEO 袁子恒一句话点破玄机，因为美国客服是人力"黑洞"，工资高、流动 大；恰好大模型最擅长多轮对话，企业换 AI 立竿见影。如今语音 AI 又成熟，电话端节省的人力可 量化、可计算尤其是随着AI语音技术的 ...

Core Insights - Sierra, an AI customer service company, achieved a valuation of $10 billion in just 18 months, with $635 million in cash and an annual recurring revenue nearing $100 million, marking it as a rare success in the AI sector [2][4][12] - The company focuses on enhancing customer experience through generative AI, addressing the high costs and turnover associated with human customer service [4][9][10] - Sierra's founders, Bret Taylor and Clay Bavor, leverage their extensive backgrounds in tech to drive the company's rapid growth and innovation [11][12] Company Overview - Sierra was co-founded by former Salesforce co-CEO Bret Taylor and ex-Google executive Clay Bavor, who aimed to revolutionize customer service by using AI to understand and fulfill customer needs rather than merely executing commands [7][11] - The company has rapidly acquired major clients, including WeightWatchers and Sonos, and has expanded its customer base to hundreds across various industries such as finance, consumer goods, and healthcare [13][14] Business Model - Sierra targets medium to large enterprises, focusing on high-value contracts with an average starting price of $150,000, which allows for deep integration and customization of AI services [19][17] - The company employs an outcome-based pricing model, where clients pay for successful resolutions of customer issues rather than usage, aligning Sierra's incentives with those of its clients [32] Technology and Innovation - Sierra does not develop its own large language models but integrates various leading models into its platform, allowing flexibility for clients to choose based on their needs [23] - The company has implemented a governance mechanism to ensure data security and compliance, which includes automatic detection and encryption of personal information [26] Market Trends - The AI customer service industry is projected to continue expanding rapidly, with increasing demand for self-service solutions and intelligent customer engagement [33] - Sierra faces competition from various players in the market, including Intercom, Kore.ai, and Genesys, each offering unique features and services [33] Challenges and Future Outlook - The AI customer service sector is not without risks, including issues related to model reliability, data privacy, and evolving customer expectations [34] - Sierra's success will depend on its ability to navigate these challenges while continuing to innovate and expand its client base [34]

老黄又投了一家具身智能公司！ Dyna Robotics，1年前刚成立，现在对外官宣了1.2亿美元 （折合人民币约8.6亿） A轮融资，新晋股东中，老黄治下的英伟达赫然在列。 众所周知，老黄已经明确下一波硬科技浪潮属于具身智能、属于物理AI……所以英伟达的投资押注，也已经在遍地播种了。 Dyna Robotics不是第一家英伟达投资的具身智能机器人公司。 但 全华班创业团队 ——三个创始人都是华人的具身智能机器人创业公司，似乎还是第一家。 Dyna Robotics有什么独特之处？ Jay 发自 凹非寺 量子位 | 公众号 QbitAI Dyna Robotics登场，老黄押注 就在最近，Dyna对外官宣了 1.2亿美元 A轮融资，投后估值6亿美元。更早之前的种子轮，大概获得了2000万美元融资。 有意思的是，这轮早期融资中，挤满了巨头产业投资部，包括英伟达、亚马逊和Salesforce。 Dyna披露，他们希望能利用这笔资金进一步完善其AI模型并部署更多机器人。 Dyna成立于2024年，目前公司只有大概30名员工，总部位于美国加利福尼亚州红木城，但他们在上海长宁区也设有分部，公司的中文名叫 达纳灵动 。 ...

Core Insights - Chinese autonomous driving companies are increasingly entering international markets, shifting from technology importers to exporters, and becoming essential partners in global collaborations [1][2][3] Group 1: International Expansion - Hesai Technology signed a laser radar order worth over $40 million with a leading US Robotaxi company [1] - Momenta plans to start L4 autonomous Robotaxi testing in Munich, Germany, in 2026, having established deep partnerships with over 20 global automakers [2] - Companies like Baidu and Xiaoma Zhixing are also expanding their Robotaxi services internationally, with plans to launch in various regions by 2025-2026 [2][3] Group 2: Technological Advancements - Chinese companies are leveraging complex road environments to develop superior algorithms, enhancing their problem-solving capabilities [4] - Momenta's "data flywheel" approach allows for continuous training and optimization of its algorithms using data from over 400,000 vehicles [4] - The shift from high-precision maps to "mapless" solutions is gaining traction, with Chinese firms leading this technological transition [5] Group 3: Cost Reduction and Commercial Viability - The cost of manufacturing Robotaxis has decreased by 80% over the past five years, making them more competitive globally [6] - Companies like Hesai Technology are producing high-performance laser radars at significantly lower costs, enabling larger fleet deployments [6] - The total cost of Xiaoma Zhixing's seventh-generation autonomous driving suite has decreased by 70%, with substantial reductions in key components [6] Group 4: Market Dynamics and Future Outlook - The capital market's focus is shifting from technology feasibility to commercialization timelines and cash flow expectations [7] - A potential wave of mergers and acquisitions may occur as companies with specific technological expertise seek partnerships with larger firms [7] - Collaborations between tech companies and ride-hailing platforms like Uber are expected to accelerate profitability in the autonomous driving sector [7][8]

Group 1 - Chinese autonomous driving companies are increasingly entering overseas markets, with significant contracts being signed, such as Hesai Technology's $40 million lidar order and Junsheng Electronics' 15 billion yuan automotive intelligence project [1][2] - Momenta has partnered with Uber to conduct L4 autonomous driving Robotaxi tests in Munich by 2026, showcasing the shift from technology import to export in the Chinese autonomous driving sector [2][3] - The capital landscape is evolving, with companies like Hello Chuxing securing strategic financing to support their Robotaxi business, indicating a shift from pure investment to collaboration [3][4] Group 2 - The competitive edge of Chinese companies lies in their ability to produce cost-effective solutions, with the manufacturing costs of Robotaxi decreasing by 80% over the past five years [6][7] - The advancements in algorithms, particularly Momenta's data-driven approach, allow for rapid iteration and optimization, leveraging the complex driving conditions in China [4][5] - The market is witnessing a shift in focus from technological feasibility to commercialization timelines and cash flow expectations, leading to potential mergers and acquisitions in the sector [7][8]

Core Viewpoint - The Hong Kong government is promoting AI development, which has positively impacted the stock performance of SenseTime, leading to a significant increase in its market valuation [1][3]. Financial Performance - SenseTime reported a revenue of 2.4 billion yuan for the first half of 2025, representing a 36% year-on-year growth, exceeding market expectations [3]. - The company's generative AI revenue surged by 73%, accounting for 77% of total revenue, indicating a successful strategic shift [3]. - Gross margin remained at 39%, with adjusted net losses narrowing by 50% and cash reserves improving to 13.2 billion yuan [3]. - Accounts receivable management showed a 96% increase in cash collection, with accounts receivable turnover days reduced by 49% [3]. Business Strategy - SenseTime's "one foundation, two wings" strategy has enabled its products to penetrate various industries, with over 3 million users for its enterprise service series [4]. - The company has established a commercial closed loop from "model" to "application" to "scenario," making generative AI a quantifiable and scalable business source [4]. Technological Advancements - SenseTime's AI infrastructure has achieved a total computing power of over 25,000 PetaFLOPS, with a 95% efficiency in heterogeneous training [7]. - The latest model, V6.5, has improved pre-training throughput by over 20% and inference throughput by 35%, significantly enhancing performance and cost-effectiveness [8]. - The automated data preparation process has reduced costs and accelerated model optimization, creating a competitive data barrier [8]. Market Outlook - Goldman Sachs has upgraded SenseTime's rating to "buy" with a target price of 2.72 HKD, indicating a potential upside of approximately 30% [10]. - The firm anticipates that generative AI revenue will account for 91% of total revenue by 2030, marking a significant shift towards industry leadership [10]. - Multiple brokerages have expressed optimism about SenseTime's growth potential, citing its strong market position and innovative capabilities [11]. Stock Performance - SenseTime's stock has risen nearly 80% year-to-date, with a peak increase of over 110%, reflecting strong market interest and trading activity [12].

Core Insights - The article highlights a significant turning point in the Chinese AI application market, with ByteDance's Doubao surpassing DeepSeek to become the leading native AI application with 157.42 million monthly active users, reflecting a 6.6% month-over-month growth [2] Group 1: Product Philosophy - The first principle of Doubao's product is to address user needs, focusing on the essential value of AI rather than technical competition, which differentiates it from other AI products [4] - Doubao's approach is more lifestyle and entertainment-oriented, making it accessible for a broader audience, including students and professionals, who use it for various purposes such as companionship and learning social skills [4] Group 2: Business Strategy - Doubao's success is attributed to ByteDance's "great effort leads to miracles" philosophy, leveraging massive traffic, professional data, and algorithmic advantages [6] - The application has benefited from significant advertising investments, with over 625 million ads placed in 2024, amounting to 1.5 billion yuan, enhancing its visibility and user acquisition [6] Group 3: Data and Model Development - Doubao is not a single text model but part of a family of models that includes text, video generation, and real-time voice capabilities, supported by ByteDance's extensive product ecosystem [7][8] - The introduction of the UltraMem sparse model architecture has improved inference speed by 2-6 times and reduced costs by up to 83%, enhancing Doubao's efficiency [7][8] Group 4: User Growth and Data Intelligence - The user base of Doubao has rapidly increased, with daily new users rising from 200,000 to 900,000 between May and July 2024, leading to over 160 million users by November 2024, providing vast data for model optimization [9] - The essence of AI products lies in data intelligence, creating a positive feedback loop where more users generate more data, making the AI smarter and attracting even more users [9] Group 5: Key Takeaways - Doubao's success illustrates the importance of returning to user scenarios, emphasizing that users prioritize practical solutions over technical specifications [10][11] - The core of data intelligence is the data flywheel effect, rather than isolated models, highlighting the need for AI applications to lower usage barriers for widespread adoption [11] - The story of Doubao serves as a reminder that the most fundamental answers often hold the most power in the evolving AI landscape [11]