![[d4a81c85-4fbf-4135-89cb-34d421dcb716.mp3]]

DeepMind推出可本地运行的Gemini Robotics模型，提升机器人自主能力

简报：

• Google DeepMind发布了Gemini Robotics On-Device机器人AI模型，实现机器人无需依赖互联网即可本地运行和完成复杂任务，如拉拉链、叠衣服、系鞋带等精细操作。
• 该模型基于Gemini 2.0架构，具备视觉、语言、动作三位一体的多模态能力，并针对ALOHA、Franka FR3和Apollo等机器人平台做了优化。
• 本地运行特性降低了延迟与网络依赖，适用于网络不佳或安全敏感场景，并配套Gemini Robotics SDK与MuJoCo模拟器，开发者仅需50到100次演示即可训练适应新任务。
• 系统还集成安全控制与行为合规机制，当前仅向受信开发者开放，标志着机器人通用智能迈向可落地应用新阶段。

相关链接：

• Google's DeepMind unveils AI robots that work offline
• Google's New Robotics AI Model Works Offline, Making It Helpful in Real-World Environments
• Google DeepMind's optimized AI model runs directly on robots
• Google Introduces Gemini Robotics On-Device AI Model, Can Adapt to Different Types of Robots

DeepMind Unveils On-Device Gemini Robotics Model, Boosting Robot Autonomy

Brief:

• Google DeepMind has released the Gemini Robotics On-Device AI model, allowing robots to operate locally without internet reliance and perform complex tasks like zipping zippers, folding clothes, and tying shoelaces, which require delicate manipulation. 🤖
• Based on the Gemini 2.0 architecture, the model boasts a multimodal capability integrating vision, language, and action, and has been optimized for robot platforms like ALOHA, Franka FR3, and Apollo.
• The local operation feature reduces latency and network dependency, making it suitable for poor network or security-sensitive scenarios. It comes with the Gemini Robotics SDK and MuJoCo simulator, allowing developers to train for new tasks with only 50 to 100 demonstrations. ⚡
• The system also integrates safety controls and behavior compliance mechanisms. Currently, it is only accessible to trusted developers, marking a new phase where general robot intelligence moves towards practical application. ✨

Related Links:

• Google's DeepMind unveils AI robots that work offline
• Google's New Robotics AI Model Works Offline, Making It Helpful in Real-World Environments
• Google DeepMind's optimized AI model runs directly on robots
• Google Introduces Gemini Robotics On-Device AI Model, Can Adapt to Different Types of Robots

罗森东京新门店引入AI与机器人提升运营效率

简报：

• 2025年6月23日，罗森在东京都港区开设结合AI与机器人技术的“次世代门店”——罗森高轮 Gateway City 店。
• 门店配备电子看板，顾客拿起商品时可自动推荐相关商品，并显示列车运行、天气等本地资讯。
• 机器人负责部分烹饪和饮料商品的陈列工作，以提高销售效率并缓解人手短缺问题。
• 罗森计划将此技术模式推广至更多地区。

相关链接：

• 罗森在日本引入“次世代门店”：机器人能烹饪，AI电子屏会推荐商品

Lawson's New Tokyo Store Introduces AI and Robots to Boost Operational Efficiency 🤖

Brief:

• On June 23, 2025, Lawson opened its "next-generation store," Lawson Takanawa Gateway City Store, in Minato-ku, Tokyo, integrating AI and robot technologies.
• The store is equipped with digital signage that automatically recommends related products when customers pick up an item, and also displays local information such as train operation status and weather updates. 🛍️
• Robots are responsible for the display of some cooked food and beverage items to improve sales efficiency and alleviate labor shortages.
• Lawson plans to expand this technological model to more regions.

Related Links:

• Lawson Introduces "Next-Generation Store" in Japan: Robots Can Cook, AI Digital Screens Recommend Products

微软与OpenAI因AGI技术访问权发生合同争议

简报：

• 据《The Information》报道，微软与OpenAI在与通用人工智能（AGI）相关的合同条款上存在分歧。
• 现有协议规定，一旦OpenAI实现AGI，微软对该技术的访问将失效。
• 微软希望OpenAI删除这一限制条款，但OpenAI目前拒绝了该请求，双方僵持不下。

相关链接：

• 36氪快讯：微软与OpenAI在AGI合同条款上存分歧
• TechCrunch报道
• MSN报道

Microsoft and OpenAI in Contract Dispute Over AGI Technology Access 💥

Brief:

• According to The Information, Microsoft and OpenAI are at odds over contract terms related to Artificial General Intelligence (AGI).
• The existing agreement stipulates that Microsoft's access to the technology would cease once OpenAI achieves AGI.
• Microsoft wants OpenAI to remove this restrictive clause, but OpenAI has currently rejected the request, leading to a stalemate between the two parties. 😬

Relevant Links:

• 36Kr News Flash: Microsoft and OpenAI Disagree on AGI Contract Terms
• TechCrunch report
• MSN report

OpenAI推出搜索连接器，提升ChatGPT企业生产力

简报：

• OpenAI正式上线搜索连接器（Search Connectors）功能，允许ChatGPT与Google Drive、Dropbox、SharePoint、Box等外部云存储服务无缝集成。
• 此功能现正逐步向欧盟以外的Pro与Team用户开放，用户可直接在ChatGPT内搜索并处理各类文档、表格和演示文稿，提高信息检索与工作协作效率。
• 搜索连接器支持企业级权限管理，未来将支持更多平台如Microsoft SharePoint和Slack，并计划与即将发布的Operator代理集成。
• 该功能暂未在欧盟等地区上线，因数据隐私等法规限制，OpenAI正持续推进合规与全球扩展。

相关链接：

• AIBase：OpenAI搜索连接器上线

OpenAI Launches Search Connectors, Boosting ChatGPT Enterprise Productivity 🚀

Brief:

• OpenAI has officially launched its Search Connectors feature, enabling seamless integration between ChatGPT and external cloud storage services like Google Drive, Dropbox, SharePoint, and Box.
• This feature is currently being rolled out incrementally to Pro and Team users outside the EU, allowing them to directly search and process various documents, spreadsheets, and presentations within ChatGPT, thereby enhancing information retrieval and collaborative efficiency. 📈
• Search Connectors support enterprise-level permission management. Future plans include support for more platforms such as Microsoft SharePoint and Slack, and integration with the upcoming Operator agents.
• The feature is not yet available in regions like the EU due to data privacy and other regulatory restrictions. OpenAI is actively working towards compliance and global expansion. 🌍

Related Links:

• AIBase: OpenAI Search Connectors Go Live

Anthropic宣布今秋将在东京开设首个亚太办事处

简报：

• 美国AI初创公司Anthropic于2025年6月24日宣布，将于今年秋季在东京设立其亚太地区首个办事处。
• Anthropic成立于2020年，专注于开发以人为本的安全AI技术，并与全球多家科技企业有广泛合作。
• 新办事处将助力Anthropic拓展亚太市场、加强与本地企业和科研机构的合作，并促进AI人才吸引与技术创新。
• 此举标志着Anthropic全球化战略的重要推进，也为日本和亚太地区的人工智能产业注入新动力。

相关链接：

• Anthropic宣布今秋将在东京开设亚太首办

Anthropic Announces First Asia-Pacific Office to Open in Tokyo This Fall

Brief:

• US AI startup Anthropic announced on June 24, 2025, that it will establish its first Asia-Pacific office in Tokyo this autumn.
• Founded in 2020, Anthropic focuses on developing human-centered, safe AI technologies and collaborates extensively with numerous tech companies worldwide.
• The new office will help Anthropic expand into the Asia-Pacific market, strengthen collaborations with local businesses and research institutions, and promote AI talent attraction and technological innovation. 🚀
• This move marks a significant advancement in Anthropic's globalization strategy and injects new momentum into the AI industry in Japan and the Asia-Pacific region. 🌏✨

Related Links:

• Anthropic to Open First Asia-Pacific Office in Tokyo This Fall

SuperDec创新方法提升3D重建与机器人应用效率

简报：

• 近日ETH Zurich、斯坦福大学和微软团队推出SuperDec方法，通过超四面体几何模型实现3D场景的高效、紧凑分解，有效提升3D重建表现。
• SuperDec结合了实例分割技术与基于Transformer的神经网络，可将复杂场景的点云数据解析为一组超四面体，显著优化了场景表达和处理效率。
• 该方法在ShapeNet、ScanNet++等数据集上展示出优异的泛化能力，实现无需微调即可扩展至复杂场景，其在物体抓取和路径规划等机器人应用中也表现出强大潜力。
• SuperDec能够与扩散模型结合，为图像生成和空间控制等内容生产场景带来双重几何与语义控制的新手段。

相关链接：

• SuperDec项目主页
• 新闻源

SuperDec: Innovative Method Boosts Efficiency in 3D Reconstruction and Robotics Applications

Briefing:

• Recently, teams from ETH Zurich, Stanford University, and Microsoft introduced the SuperDec method. This approach achieves efficient and compact decomposition of 3D scenes using hypertetrahedral geometric models, significantly enhancing 3D reconstruction performance. 🚀
• SuperDec combines instance segmentation techniques with Transformer-based neural networks to parse point cloud data from complex scenes into a set of hypertetrahedra, significantly optimizing scene representation and processing efficiency. ✨
• The method demonstrates excellent generalization capabilities on datasets like ShapeNet and ScanNet++, allowing for extension to complex scenes without fine-tuning. It also shows strong potential in robotics applications such as object grasping and path planning. 🤖
• SuperDec can be integrated with diffusion models, offering a novel means of dual geometric and semantic control for content creation scenarios, including image generation and spatial control.

Related Links:

• SuperDec Project Page
• News Source

谷歌发布基于 Gemini 2.5 Pro 的 Gemini CLI 工具，开发者可免费体验

简报：

• 2025年6月25日，谷歌宣布推出全新AI命令行工具 Gemini CLI，基于自家 Gemini 2.5 Pro 模型，允许开发者在终端界面实现AI问答、内容生成、代码补全等功能。
• Gemini CLI 支持高达100万个token的上下文窗口，并与 Gemini Code Assist 及 VS Code 深度集成，同时兼容谷歌搜索和插件、脚本扩展。
• 目前该工具处于预览状态，开发者可凭谷歌账号获得免费许可证，每日免费调用达1000次，每分钟支持60次请求，谷歌未说明未来是否收费。

相关链接：

• IT之家：谷歌推出 Gemin CLI（命令行界面），基于 Gemini 2.5 Pro AI 模型
• Aibase 新闻 — 谷歌发布 Gemini CLI 工具

Google Releases Gemini CLI Tool Based on Gemini 2.5 Pro, Free for Developers to Experience! ✨

Briefing:

• On June 25, 2025, Google announced the launch of its new AI command-line interface (CLI) tool, Gemini CLI. Powered by its proprietary Gemini 2.5 Pro model, it enables developers to perform AI Q&A, content generation, code completion, and other functions directly within the terminal.
• Gemini CLI supports a context window of up to 1 million tokens 🤯, deeply integrates with Gemini Code Assist and VS Code, and is also compatible with Google Search, as well as plugin and script extensions.
• The tool is currently in preview. Developers can obtain a free license using their Google account, allowing up to 1,000 free daily calls and supporting 60 requests per minute. Google has not yet specified whether it will charge for the service in the future. 🚀

Related Links:

• IT之家: Google Launches Gemini CLI (Command-Line Interface), Based on Gemini 2.5 Pro AI Model
• Aibase News — Google Releases Gemini CLI Tool