[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-google-project-genie-street-view-grounded":3,"topics-all":33,"news-related-0618843e-c631-474b-a6a3-9dd39c86582d":52},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":20,"news_slug":26,"published_at":27,"created_at":28,"modified_at":29,"is_published":30,"publish_type":31,"image_url":13,"view_count":32},"0618843e-c631-474b-a6a3-9dd39c86582d","Google Project Genie 接入 Street View：世界模型终于脚踏实地","如果能把自己熟悉的街道变成游戏世界，你会怎么玩？\n\nGoogle DeepMind 在本届 I\u002FO 上给出了一个答案：将 Project Genie 与 Google Street View 对接，让这个通用世界模型直接以真实街景为锚点，生成可交互的虚拟环境。这不是简单的贴图替换，而是让 AI 第一次能够看到真实的街道，然后用它理解世界的方式重建一个可供探索的数字孪生。\n\n支撑这个能力的是一项叫 Maps Imagery Grounding 的技术。Street View 积累了近 20 年的真实世界影像，覆盖 110 个国家、超 280 亿张图片——这个量级的数据此前从未被系统性注入世界模型的训练管道。Genie 不是简单地把这些图片当作纹理素材，而是将其作为现实锚点，学习真实空间的结构规律：路口怎么拐、阴影怎么投射、天气如何改变一条街道的氛围。在此基础上，用户可以选择风格预设（比如海洋世界或黑白电影），Genie 就会以真实地点为起点，生成一个风格化的可交互世界。\n\n世界模型一直是具身智能和自动驾驶的核心课题：机器人需要在真实部署前，在仿真环境中完成大量训练。传统的仿真环境要么依赖人工建模，成本极高；要么过于简化，与真实世界存在 sim-to-real gap——机器人在仿真中学会的技能，到真机上一执行就失效。Project Genie 的做法提供了一个新思路：用海量真实影像+生成式世界模型，构建一个介于手工仿真与纯真实数据之间的中间层。Waymo 已经率先用 Genie 来模拟极端路况下的自动驾驶决策，比如让系统学会处理难得一见的阳光直射场景，而不需要真的等那个瞬间出现。\n\n这次更新的意义不在于普通用户能去自己家门前潜水看鱼，而在于它验证了一条规模化构建物理世界仿真数据的路径。过去业界普遍认为，要训练一个能在真实物理世界中可靠运行的 AI，唯一办法是海量真实机器人数据——这让很多公司望而却步。如果 Genie + Street View 能证明真实影像驱动的世界模型可以显著缩小 sim-to-real gap，那它的影响将远超一个 I\u002FO 演示，而是会直接降低整个具身智能行业的门槛。\n\n这条路的挑战也很现实：Street View 目前仅覆盖美国，未来扩展到全球的真实影像资产需要更复杂的地理信息和更新机制。此外，如何确保生成环境的物理一致性，也是尚未解决的问题。但无论如何，Google 正在把世界模型从想象的世界拽向真实的世界——这一步，很关键。","https:\u002F\u002Fblog.google\u002Finnovation-and-ai\u002Fmodels-and-research\u002Fgoogle-deepmind\u002Fproject-genie-expands\u002F","35ce748f-48b7-4638-88ef-effa57a7e749",[10,14,17],{"id":11,"name":12,"slug":12,"description":13,"color":13},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",{"id":18,"name":19,"slug":19,"description":13,"color":13},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",[21],{"id":22,"lang":23,"title":24,"summary":25,"content":13},"6680d3d7-ee5c-469e-b1cc-76bd4cff06f7","en","Project Genie meets Street View: world models get grounded","If you could turn the streets you're familiar with into a game world, how would you play?\n\nAt this I\u002FO, Google DeepMind offered an answer: connecting Project Genie with Google Street View, allowing this general world model to generate interactive virtual environments anchored directly in real streetscapes. This isn't a simple texture swap — it's the first time AI can see real streets, then use its way of understanding the world to reconstruct a digital twin ready for exploration.\n\nThe capability is backed by a technology called Maps Imagery Grounding. Street View has accumulated nearly 20 years of real-world imagery, covering 110 countries with more than 28 billion images — a volume of data that's never been systematically injected into a world model's training pipeline. Genie doesn't simply treat these images as texture material, but as reality anchors, learning the structural rules of real space: how intersections turn, how shadows cast, how weather changes a street's atmosphere. On this basis, users can pick style presets (e.g., ocean world or black-and-white cinema), and Genie will use the real location as the starting point to generate a stylized, interactive world.\n\nWorld models have always been a core topic for embodied intelligence and autonomous driving: robots need to complete extensive training in simulation before real-world deployment. Traditional simulation environments either rely on manual modeling with high cost, or are over-simplified with a sim-to-real gap — skills learned in simulation often fail when executed on real hardware. Project Genie's approach offers a new direction: using massive real imagery + generative world models to build an intermediate layer between manual simulation and pure real data. Waymo has already taken the lead in using Genie to simulate autonomous-driving decisions in extreme road conditions — for example, training the system to handle rare direct-sunlight scenarios, without actually waiting for that moment to appear.\n\nThe significance of this update isn't that ordinary users can go diving in front of their own homes, but that it validates a path to scale physical-world simulation data. The industry used to broadly believe that the only way to train an AI that could reliably run in the real physical world was massive real-robot data — a barrier that deterred many companies. If Genie + Street View can prove that world models driven by real imagery can significantly close the sim-to-real gap, the impact will extend far beyond an I\u002FO demo, directly lowering the bar for the entire embodied-intelligence industry.\n\nThe challenges along this road are real: Street View currently only covers the US, and expanding real-imagery assets globally will require more complex geographic-information and update mechanisms. Additionally, ensuring physical consistency in generated environments remains unsolved. But regardless, Google is pulling world models from the imagined world toward the real world — this step matters.","google-project-genie-street-view-grounded","2026-05-19T19:00:00Z","2026-05-19T19:07:21.874300Z","2026-08-19T02:08:40.142862Z",true,"agent",159,[34,43],{"slug":35,"tag_slug":35,"title_zh":36,"title_en":37,"intro_zh":38,"intro_en":39,"id":40,"is_active":30,"created_at":41,"modified_at":42},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":44,"tag_slug":44,"title_zh":45,"title_en":46,"intro_zh":47,"intro_en":48,"id":49,"is_active":30,"created_at":50,"modified_at":51},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":53},[54,59,64,69,74,79],{"id":55,"title":56,"news_slug":57,"published_at":58},"2657cbe0-7743-43f2-9332-ee18b84b1229","Directing the World: 中国电信 TeleAI 把自回归视频世界模型推到\"组合控制\"","teleai-directing-the-world","2026-07-01T10:30:00+00:00",{"id":60,"title":61,"news_slug":62,"published_at":63},"e4776508-8e3b-4eba-a804-ff4ee7e8a76d","「Holo-World」用一张图控制相机、物体和天气：视频世界模型首次把\"环境状态\"做成独立控制轴","holo-world-camera-object-weather-control","2026-06-21T16:00:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"800de722-720c-4fd6-bc58-c09398927fd9","Kairos 把世界模型做成「Native Stack」：混合时序注意力 + 误差上界，给 Physical AI 一个长程一致底座","kairos-hybrid-temporal-attention-physical-ai","2026-06-18T22:30:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"0b9b2439-b5f8-413e-8e62-58621123cc2f","Continuous Audio Thinking：把「思考」搬进音频 LLM，零解码成本补齐声学信息损失","coat-continuous-audio-thinking-latent","2026-06-18T06:00:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"efca492c-b9b1-4d72-bf3b-32f08fc0f515","世界模型崛起：AI 从数字世界走向物理世界的关键一步","mit-world-models-physical-ai-lecun-fei-fei","2026-04-27T13:05:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"33c8038a-f1dc-4b64-b6b7-604baf43f729","CESA 数据:85.8% 日本游戏开发者已把生成式 AI 写进工作流","cesa-2026-japan-game-dev-genai-85pct","2026-09-20T00:00:00+00:00"]