[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-nvidia-halos-for-robotics-fsi-igx-thor":3,"news-related-f453f1d8-c41b-4fdf-9eba-54a7f6222d74":36},{"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},"f453f1d8-c41b-4fdf-9eba-54a7f6222d74","英伟达把自动驾驶十年的安全账本搬进机器人：Halos for Robotics 全栈落地工厂","NVIDIA 把过去十年在自动驾驶上攒下的安全家底，整建制搬进了机器人领域。6 月 22 日发布的 **Halos for Robotics** 是一套全栈功能安全系统，号称业界首个为工业机器人、人形机器人和 AMR 设计的端到端安全栈，覆盖从底层算力、上层软件到第三方认证的完整链条。\n\n底层硬件是 **IGX Thor**：一颗工业级 AI 计算模块，最高 2070 FP4 TFLOPs、14 核 Neoverse CPU、128 GB 显存。它把一块独立的 Safety Island (FSI) 与主算力物理隔离，提供 IEC 61508 SIL 3 能力——安全监控可以独立运行、不被主域故障牵连。配套的 Holoscan Sensor Bridge 把多模态传感器以安全通道送进算力域。\n\n中间层是 **Halos OS**：把 DRIVE AGX 上跑了多年的安全软件栈移植到机器人，包含 Halos Core 基础安全 OS 和 Outside-In Safety Blueprint 等应用模板。Agility 的 Digit 是首个客户，已将它集成进自家的安全人类检测系统，工厂客户包括亚马逊、GXO、Schaeffler、丰田加拿大。\n\n最高层是 **Halos AI Systems Inspection Lab**——一个 ANAB 认证的 ISO\u002FIEC 17025 检查机构，成员已从 AV 时代的 16 家扩展到 43 家（23 家机器人、Boston Dynamics 也在其中）。第三方认证的时间和成本被显著压缩。\n\n值得一提的是，**NVIDIA 同时是 IEC 61508 和 ISO\u002FIEC TS 22440 的制定参与方**。在制定规则的同时卖实现，这种\"标准 + 芯片 + 系统\"的一体化打法，是过去十年汽车、航空安全行业的典型路径，现在被搬到了机器人身上。对于人形机器人能否真正走出笼子、与工人并肩干活这件事，这可能比又一项 VLA 模型更关键。","https:\u002F\u002Fdeveloper.nvidia.com\u002Fblog\u002Finside-nvidia-halos-for-robotics-a-full-stack-functional-safety-system-for-physical-ai","474eef8c-e0c3-46cf-adee-c089558220f9",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"1fcfaaf2-67de-43d3-9e35-5784852fec60","ai-safety",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},"e0d31e94-ce47-4c8f-831c-d3d2926d42f3","hardware",{"id":21,"name":22,"slug":22,"description":13,"color":13},"8dac812d-3839-4abe-a855-5f56ec9515fd","nvidia",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"cf50d1e5-0236-4e5c-b6fc-8b62231ddc2f","en","NVIDIA Halos for Robotics lands a decade of AV safety","NVIDIA released Halos for Robotics, a full-stack functional-safety system that brings a decade of autonomous-driving safety engineering into robotics. The system is designed for industrial robots, and the first deployment is in Audi's manufacturing plants.\n\nThe technical details: Halos for Robotics has four components — (1) a \"safety-certified\" robot foundation model (trained with safety constraints built-in); (2) a \"safety monitor\" (a separate model that watches the primary model's actions and intervenes if necessary); (3) a \"safety case\" documentation framework (formal proof that the system meets ISO 13849 and IEC 61508 safety standards); (4) a \"deployment toolchain\" (CI\u002FCD for safety-certified models).\n\nThe \"decade of AV experience\" highlight: Halos is built on NVIDIA's decade of experience in autonomous-driving safety (the original DRIVE platform). The key insight: AV safety is fundamentally about \"perception + planning + control\" with safety constraints, and the same architecture applies to robotics. The \"safety monitor\" pattern is directly borrowed from the AV world.\n\nThe Audi deployment: the first Halos for Robotics deployment is in Audi's body-shop robots. The robots perform welding, painting, and assembly tasks, with a safety monitor watching every action. The monitor can stop the robot within 5ms if it detects an unsafe action (e.g., the robot is about to collide with a human worker). The system is certified to ISO 13849 PL d.\n\nThe bigger takeaway: \"safety as a first-class concern\" is arriving in robotics. The robotics industry has historically treated safety as an add-on (e.g., physical cages, light curtains), but with AI-powered robots, software-level safety is becoming essential. Halos for Robotics is the first comprehensive \"software safety\" system for industrial robots, and the Audi deployment is a real-world validation.","nvidia-halos-for-robotics-fsi-igx-thor","2026-06-23T10:00:00Z","2026-06-23T10:17:34.260778Z","2026-08-19T02:08:40.142862Z",true,"agent",81,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"66d66fa2-364e-4fa2-a9b2-e69e6f86dc8c","Vera Rubin 平台登陆 ISC 2026：144 张 GPU + 100% 液冷，把 TOP500 算力压进科研机柜","nvidia-vera-rubin-isc2026-144-gpu","2026-06-22T20:00:00+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"f28c994f-1922-4abc-ae57-0e53c44ff31c","太空算力被热力学打回原形：IEEE 量化「轨道数据中心」十年内难落地","orbital-data-center-ieee-thermo-decade","2026-06-14T06:10:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"6242571f-c635-4794-b19a-ba08eac4f6d1","NVIDIA 发布全球首款面向 Agent 时代的 CPU：Vera 已送抵 Anthropic、OpenAI、SpaceXAI","nvidia-vera-cpu-agent-anthropic-openai","2026-05-20T01:30:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"b51af942-496b-4b58-95fb-37980d12a743","逆向工程发现:微软画图本地生成的 AI 图像,像素里埋着服务器下发的水印 GUID","mspaint-invisible-watermark-guid","2026-08-26T13:00:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"fb1cbe25-8b85-41ec-b619-9a27b405ec34","AMD 收购 Taalas:把 AI 模型权重「刻进硅片」的推理新打法","amd-acquires-taalas-inference-chip","2026-08-19T01:00:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"45375854-7739-4dd1-bc6a-30db4474652a","Taalas HC2:把单片参数拉到 200 亿,「模型刻进硅片」的第二章","taalas-hc2-20b-mxfp4-50-chips-1t-amd","2026-08-19T00:00:00+00:00"]