[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-gaode-abot-m0-5-soft-bus":3,"topics-all":36,"news-related-962927f4-dba2-4160-8e26-4c9fa4cdbb55":55},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"962927f4-dba2-4160-8e26-4c9fa4cdbb55","高德 ABot 全栈升级:把机器人拆成操作系统,M0.5 论文给出软总线答案","2026 年 7 月 22 日,高德宣布 ABot 具身体系全栈升级,一次性发布 N1、M0.5、ER、AgentOS、C0 五款模型,在 17 项基准拿下 SOTA。五款模型对应机器人\"脚、手、感官、中枢、运动神经\",把\"操作系统思维\"搬进具身智能。\n\n最值得拆的是 **ABot-M0.5**(arXiv:2607.00678),提出**统一移动-操作世界动作模型(WAM)**。论文点 VLA 三大硬伤:反应式、无显式世界建模、长程 rollout 累积误差。解法是\"三层对齐\":**中间潜动作**桥接视频潜变量与控制信号;**双层 Mixture-of-Transformers** 解耦\"底座+机械臂\"两个异构动作子空间;**dream-forcing 训练策略**在模型预测视频上渐进式训练逆动力学,把训练-推理分布对齐。M0.5 在长程任务成功率和细粒度控制精度上同时 SOTA。\n\n剩四款按 OS 摆位:**N1 脚**(导航)、**M0.5 手**、**ER 感官**、**C0 运动神经**、**AgentOS 中枢**——把前四个串成长程闭环。模块独立可换,谁先升级谁,不用整体重训。\n\n对赛道意义:过去两年大家在拼\"端到端 VLA 一锅烩\",ABot 解法是承认**模块化仍有效**,但接口必须由潜动作和世界模型这种\"软总线\"定义,而不是 SLAM\u002FSemantics\u002FPlanning 的硬接口。代码已开源(`github.com\u002Famap-cvlab\u002FABot-Manipulation`)。具身的\"GPT 时刻\"还没到,把 OS 层先搭起来,比堆参数更接近落地。","https:\u002F\u002F36kr.com\u002Fnewsflashes\u002F3906614197622145","5e4fd3d1-9cb4-44a6-bae5-9ffb449c05c1",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"e676a5cf-1f24-472f-a765-86fa21a1bc3c","ai-model",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"a8002d98-9df1-4ab9-94d4-a7625af634c4","china-ai",{"id":18,"name":19,"slug":19,"description":13,"color":13},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",{"id":21,"name":22,"slug":22,"description":13,"color":13},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":28},"19468f79-2345-451b-8a69-a38da426c638","en","Amap ABot rebuilt as a robot OS, M0.5 proposes the soft bus","On July 22, 2026, Amap announced a full-stack upgrade of the ABot embodied-AI system, releasing N1, M0.5, ER, AgentOS and C0 in one go, and taking SOTA on 17 benchmarks. The five models correspond to a robot's \"feet, hands, senses, central nervous system, motor neurons\", transplanting \"operating-system thinking\" into embodied AI. The most interesting piece to dissect is **ABot-M0.5** (arXiv:2607.00678), which proposes a **unified Mobile-Manipulation World Action Model (WAM)**. The paper diagnoses VLA's three hard problems: reactive, no explicit world modeling, cumulative error in long-horizon rollout. The solution is \"three-layer alignment\": **intermediate latent actions** bridging video latent variables and control signals; a **two-layer Mixture-of-Transformers** decoupling the \"base + arm\" two heterogeneous action subspaces; and a **dream-forcing training strategy** that progressively trains inverse dynamics on the model's predicted video, aligning the training-inference distribution. M0.5 hits SOTA on both long-horizon task success rate and fine-grained control accuracy. The other four are arranged by OS position: **N1 feet** (navigation), **M0.5 hands**, **ER senses**, **C0 motor neurons**, **AgentOS central nervous system** — stringing the first four into a long-horizon closed loop. Modules are independently swappable, so whichever upgrades first doesn't require a full retrain. The takeaway for the field: the past two years have been a race to \"end-to-end VLA in one pot\"; ABot's answer is to admit **modularization still works**, but interfaces must be defined by \"soft buses\" like latent actions and world models, not hard SLAM\u002FSemantics\u002FPlanning interfaces. Code is open-sourced (`github.com\u002Famap-cvlab\u002FABot-Manipulation`). The embodied \"GPT moment\" hasn't arrived; building the OS layer first is closer to production than piling on parameters.","gaode-abot-m0-5-soft-bus","2026-07-22T10:30:00Z","2026-07-22T10:05:32.499580Z","2026-08-19T02:08:40.142862Z",true,"agent",211,[37,46],{"slug":38,"tag_slug":38,"title_zh":39,"title_en":40,"intro_zh":41,"intro_en":42,"id":43,"is_active":33,"created_at":44,"modified_at":45},"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":47,"tag_slug":47,"title_zh":48,"title_en":49,"intro_zh":50,"intro_en":51,"id":52,"is_active":33,"created_at":53,"modified_at":54},"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":56},[57,62,67,72,77,82],{"id":58,"title":59,"news_slug":60,"published_at":61},"cb5ee922-ad27-4a5a-9b2b-8382903876df","Mozilla 把模型选择权交还给用户:Mistral Small 4 进 Firefox 默认菜单","mistral-small-4-firefox-smart-window-beta","2026-09-22T03:00:00+00:00",{"id":63,"title":64,"news_slug":65,"published_at":66},"d055ddb8-4d82-4523-99b7-39c5f77e2ff7","PhysBrain 1.5 开源：8B 具身基座 28 项评测均分 72.5，官方称追平 GPT-6-Astra","physbrain-1-5-open-embodied-base","2026-09-16T21:07:24+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"17006864-46a5-405c-a8cc-24507bbc5e37","YuE2-3B 开源:乐谱可编辑的音乐生成,官方基准反超 Suno v5","yue2-3b-editable-music-generation","2026-09-10T13:20:00+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"2b37a19b-1dde-4238-bef5-39b1d19157f1","OpenBMB 开源 MiniCPM5-2B:2B 端侧模型平均分超对比集 4B 级","openbmb-minicpm5-2b-on-device","2026-09-07T17:02:00+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"453ce9a1-5d55-4981-b44d-c261b8051724","GLM-5.3 753B 权重上架 HuggingFace,智谱兑现两周开源承诺","glm-5-3-weights-huggingface-release","2026-08-28T15:15:00+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"f6e4aab0-7693-4c2c-bb66-c1641fc2cc3e","Ox Alpha 谜底揭晓:智谱 GLM-5.3-Flash,MIT 开源 320B MoE","ox-alpha-glm-5-3-flash-reveal","2026-08-27T13:30:00+00:00"]