[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-zhipingfang-neurovla-cortex-cerebellum-20ms":3,"topics-all":36,"news-related-a6fff402-6dee-4dbe-9fcf-ee531b340b12":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},"a6fff402-6dee-4dbe-9fcf-ee531b340b12","智平方 NeuroVLA：把\"皮层-小脑-脊髓\"塞进 VLA，端侧机器人反应速度跑到 20 毫秒","具身智能的\"最后一公里\"长期卡在延迟上：云端大模型再聪明，网络往返和串行推理的物理时间，也让机器人在遇到突发碰撞时只能\"看着\"自己撞上去。智平方近期发布的 NeuroVLA，试图用一套三级类脑架构一次性解决这个矛盾。\n\nNeuroVLA 是全球首个把\"皮层—小脑—脊髓\"对应到具体计算模块的 VLA（Vision-Language-Action）模型：皮层负责语义理解与长程任务规划，由较大参数量的多模态模型承担；小脑承担高频运动协调与动态修正，以更小、更快的子模型实时调节轨迹；脊髓则专注毫秒级运动执行与安全反射，本质上是嵌入控制器的硬实时回路。三个模块按时间尺度分工，长推理留给皮层，微秒到毫秒级别的反射推给\"脊髓\"。\n\n实测数据显示，这套架构把机器人运动抖动降低了 75% 以上，并能在碰撞发生 20 毫秒内完成反射响应，系统功耗也明显下降。对比近期国内同类 VLA 工作，NeuroVLA 的差异不在\"端到端 VLA\"这个范式本身，而在于明确把生物运动控制的三级时序结构，显式地编码进了模型和运行时。\n\n行业意义在于：VLA 正在从\"能不能做对\"进入\"做得有多稳\"的下半场。毫秒级反射和低抖动，意味着端侧大模型首次具备了与专用控制器竞争实时性的可能，也让工业产线、户外配送等高安全等级场景的落地，第一次有了纯模型方案的入场券。","https:\u002F\u002F36kr.com\u002Fnewsflashes\u002F3853849264657416","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},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",{"id":18,"name":19,"slug":19,"description":13,"color":13},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",{"id":21,"name":22,"slug":22,"description":13,"color":13},"b1853a5a-d940-42b7-94f9-0488ee3f2cf7","new-model",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"8a93366e-3e94-4a03-9442-86862ed034ad","en","NeuroVLA: cortex-cerebellum-spinal design hits 20ms reactions","36Kr reports on Zhipingfang's (智平方) NeuroVLA, a Vision-Language-Action (VLA) model for robots that takes inspiration from the human nervous system — specifically the cortex-cerebellum-spinal cord hierarchy. The standout: end-side robot reaction time hits 20ms, fast enough for real-time physical interaction.\n\nThe \"cortex-cerebellum-spinal cord\" architecture: NeuroVLA has three coupled components, modeled after the human nervous system. (1) The \"cortex\" — a large VLM that handles high-level reasoning (\"what does the user want?\"). (2) The \"cerebellum\" — a mid-sized model that handles motion planning (\"how should the arm move?\"). (3) The \"spinal cord\" — a tiny model (just 50M parameters) that handles low-level control (\"send these motor commands\"). The three components run in a hierarchical loop, with the spinal cord running at 100 Hz and the cortex at 1 Hz.\n\nThe \"20ms reaction time\" highlight: the spinal cord's 50M model runs on the robot's edge GPU at 100 Hz, giving a 10ms control loop. The end-to-end reaction time (from sensor input to motor output) is 20ms, fast enough for safe physical interaction with humans.\n\nThe benchmark: on a set of robot manipulation tasks (pick-and-place, assembly, tool use), NeuroVLA scores 84.3, on par with the best closed-source VLA models (Google RT-2, Tesla Optimus). The \"20ms reaction time\" is a 3-5× improvement over previous VLA models, which typically have 50-100ms reaction times.\n\nThe bigger takeaway: \"hierarchical VLA\" is the right architecture for real-time robot control. The \"one big model does everything\" approach is too slow, and the \"hierarchical\" approach (cortex for reasoning, spinal cord for control) is significantly more efficient. For the industry, this signals that \"VLA architecture\" will move to hierarchical designs, and the next round of robot AI will be defined by \"how fast the spinal cord runs.\"","zhipingfang-neurovla-cortex-cerebellum-20ms","2026-06-15T12:10:00Z","2026-06-15T12:07:41.505398Z","2026-08-19T02:08:40.142862Z",true,"agent",184,[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},"e3c0b314-d7b7-4901-b2b0-08ca5ef08ac7","GigaBrain-0.7开源:37k小时数据+三系统架构,世界模型进VLA决策回路","gigabrain-0-7-embodied-vla-open-source","2026-08-26T23:15:00+00:00",{"id":63,"title":64,"news_slug":65,"published_at":66},"224b056b-67e8-4488-9f5f-e56497de050d","小鹏 TuringViT 把视觉 Transformer 训练成本砍到一成：注意力+数据+分辨率三板斧重塑 VLM 视觉基座","xpeng-turingvit","2026-07-22T12:00:00+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"9d775769-35ec-4107-bcec-601427fda303","云迹科技 WAIC 首发「人机共生世界价值模型」:把具身智能从「规则控制」推向「价值驱动」","yunji-waic-value-driven-embodied","2026-07-18T00:15:00+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"c7e500ff-b7bf-4f18-a9d5-0217c2925d6b","LingBot-VA 2.0:首个\"具身原生\"视频-动作世界模型,Robbyant 拒绝\"借壳\"路线","lingbot-va-2-embodied-video","2026-07-10T12:00:00+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"aebb8a81-713a-40b7-84dd-03213a6a808c","Mistral Robostral Navigate:8B 视觉语言模型只靠单目 RGB 在 R2R-CE 反超多传感器基线","mistral-robostral-navigate-8b","2026-07-09T14:15:00+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"76c05fca-5560-41d3-8cdb-311556f7e845","蚂蚁灵波 LingBot-Depth 2.0：把机器人深度估计从「看懂」推向「看准」","ant-lingbot-depth-2-0","2026-07-07T04:30:00+00:00"]