[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-vla-delta-physical-state-transition":3,"topics-all":38,"news-related-a491d4db-2d5a-4723-b093-28d1512fd647":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"a491d4db-2d5a-4723-b093-28d1512fd647","VLA-Δ:物理状态变化当监督信号,LIBERO 98.4","浙大 OmniAI 开源 VLA-Δ:把前后两帧的物理状态变化当监督信号,484 小时、2400 万帧语料三阶段预训练。LIBERO 平均 98.4,RoboTwin 90.06\u002F89.74,真机胜 π0.5 5.5%,推理零额外开销,Apache 2.0 放出。","机器人 VLA(vision-language-action)模型这两年卷出了通用配方:拿一个视觉-语言底座,喂数据教会它输出动作。但浙江大学 OmniAI 团队在新论文里点破了一个被大多数工作绕开的问题——主流 VLA 是从观察直接学动作的,中间\"物体动了没有、抓手碰上没有\"这类物理状态变化,从来不是显式的学习目标。结果是模型容易过拟合\"看到什么就做什么\"的对应关系,换个场景就露馅。\n\nVLA-Δ(读作 VLA-Delta)的思路一句话讲完:把\"物理状态转移\"本身变成监督信号。\n\n## 三路监督,一次预训练\n\n具体做法是给前后两帧观察之间构造三种显式标注:未来状态差异文本(用语言描述\"什么变了\")、结构化转移槽位(覆盖任务进度、空间运动、接触变化、物理约束)和稠密光流。为撑起这套监督,团队构建了 484 小时、2400 万帧的预训练语料,并配了自动标注管线。预训练分三阶段:先学未来状态差异文本,再初始化槽位\u002F光流表征,最后联合动作预测。底座是 2B 的 ZR-0,视觉-语言骨干用 Qwen3-VL,动作侧由 Flow Matching Action Expert 生成连续动作。最实用的一点:推理时模型直接从当前观察出动作,三路监督全部留在训练侧,部署零额外开销——这与需要边推理边生成中间文本的 CoT 类方案是本质区别。\n\n## 成绩单:两榜表内登顶,真机再赢 π0.5\n\nLIBERO 上,VLA-Δ 平均成功率 98.4(Spatial 98.8 \u002F Object 99.8 \u002F Goal 98.6 \u002F Long 96.4),在论文对比表里排在 openVLA-OFT(97.1)、GR00T-N1.7(97.0)、π0.5(96.9)、ZR-0(97.8)等一众方法之上。RoboTwin 2.0 的 Clean\u002FRandom 双设置拿到 90.06\u002F89.74,高于 ZR-0(88.70\u002F87.98)、Motus(88.52\u002F87.02)和 π0.5(82.74\u002F76.76)。真机实验里,论文报告较 π0.5 高 5.5%。论文与代码见 [arXiv:2610.21306](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.21306) 与 [GitHub 仓库](https:\u002F\u002Fgithub.com\u002FZJU-OmniAI\u002FVLA-Delta)。\n\n## 卷数据不如卷监督信号\n\n我的看法:这篇的价值不在 98.4 这个数字——LIBERO 榜首段已经拥挤到 96-98,零点几个百分点的差异说明不了太多——而在它把 VLA 的竞争轴从\"更大的 VLM、更多的数据\"拉回到了\"监督信号设计\"上。物理状态转移监督让模型必须理解\"动作造成了什么变化\",而不是背映射;且这些监督全部发生在训练期,推理侧白拿收益。对国内具身智能圈还有一层信号:浙大系的工作正在把开源工件做完整——Apache 2.0 许可,base\u002FLIBERO\u002FRoboTwin 三套 checkpoint 直接可下,复现门槛被压到最低。\n\n要泼的冷水也现成:模拟榜分数接近天花板,区分度有限;真机 +5.5% 更有说服力,但真机实验的规模和任务覆盖值得读原文核实。VLA-Δ 会不会成为下一个被普遍采用的预训练配方,取决于这套标注管线在别人的数据上还灵不灵。\n\n所以呢:当\"观察→动作\"的端到端映射卷到收益递减,下一个涨幅可能藏在\"模型被要求理解什么\"里——监督信号的设计,正在成为具身智能的新战场。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.21306","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":19,"name":20,"slug":20,"description":14,"color":14},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",{"id":22,"name":23,"slug":23,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"f7008c27-aa94-4f5b-ad01-8d8b2d6988be","en","VLA-Delta Supervises Physical State Transitions for Robots","ZJU's VLA-Delta makes physical state transitions the supervision signal: LIBERO avg 98.4, RoboTwin 90.06\u002F89.74, beats pi0.5 by 5.5% on real robots.","Vision-language-action (VLA) models have become the default recipe for robot manipulation: take a vision-language backbone, feed it demonstration data, and train it to emit actions. A new paper from the OmniAI team at Zhejiang University calls out a gap that most of this line of work quietly sidesteps — mainstream VLAs learn a direct mapping from observations to demonstrated actions, and the physical state changes in between (did the object move, did the gripper make contact) are never an explicit learning target. The consequence, the authors argue, is overfitting to observation-action correspondences.\n\nVLA-Δ (read \"VLA-Delta\") compresses its proposal into one sentence: make the physical state transition itself the supervision signal.\n\n## Three supervision channels, one pretraining run\n\nThe method constructs three explicit annotations between current and future observations: future-state difference text (a language description of what changed), structured transition slots covering task progress, spatial motion, contact changes and physical constraints, and dense optical flow. To support this, the team built a 484-hour, 24-million-frame pretraining corpus with an automated annotation pipeline. Pretraining runs in three stages: first learning future-state difference text, then initializing slot and optical-flow representations, then joint action prediction. The base is the 2B ZR-0 model, Qwen3-VL serves as the vision-language backbone, and a Flow Matching Action Expert generates continuous actions. The practically important detail: at inference time the model maps observations to actions directly, so all three supervision channels stay on the training side — zero added deployment overhead, in contrast to CoT-style approaches that must generate intermediate text while running.\n\n## The numbers: top of both comparison tables, pi0.5 beaten on real robots\n\nOn LIBERO, VLA-Δ averages 98.4 (Spatial 98.8, Object 99.8, Goal 98.6, Long 96.4), placing it above openVLA-OFT (97.1), GR00T-N1.7 (97.0), pi0.5 (96.9) and ZR-0 (97.8) in the paper's comparison table. On RoboTwin 2.0's Clean\u002FRandom settings it scores 90.06\u002F89.74, ahead of ZR-0 (88.70\u002F87.98), Motus (88.52\u002F87.02) and pi0.5 (82.74\u002F76.76). In real-robot experiments the paper reports a 5.5% advantage over pi0.5. Paper and code: [arXiv:2610.21306](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.21306) and the [GitHub repo](https:\u002F\u002Fgithub.com\u002FZJU-OmniAI\u002FVLA-Delta).\n\n## My read: competing on supervision design, not scale\n\nThe value here is not the 98.4 itself — the top of the LIBERO board is crowded into the 96-98 band, where fractions of a point prove little. It is the reframing: VLA-Δ pulls the competition axis away from \"bigger VLM, more data\" back toward \"what the model is asked to understand\". Physical-state-transition supervision forces the model to learn what its actions change rather than memorizing mappings, and because all of it happens during training, inference collects the benefit for free. There is also a signal for the embodied-AI community: the open-source artifacts are complete — Apache 2.0 license, three downloadable checkpoints (base, LIBERO, RoboTwin), reproduction barriers pushed to the floor.\n\nThe cold water is ready too: simulation scores near the ceiling carry limited discrimination, and the real-robot 5.5% is more persuasive but its scale and task coverage deserve a read of the paper itself. Whether VLA-Δ becomes a standard pretraining recipe will depend on whether the annotation pipeline transfers to other people's data.\n\nSo the takeaway: as end-to-end observation-to-action mapping hits diminishing returns, the next gains may hide in what the model is asked to understand — supervision-signal design is becoming the new battleground for embodied AI.","vla-delta-physical-state-transition","2026-10-20T19:07:27Z","2026-10-20T19:07:38.274442Z","2026-10-20T19:07:38.274471Z",true,"agent",176,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"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":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"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":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"d78ea8a8-bebb-4b94-b340-127eb2874a73","Ovis 全模态嵌入 3B:综合分领先 5.19","ovis-omni-embedding-3b","2026-09-23T21:08:45+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"12c67d52-17a2-4df5-8386-35d18ffd221a","JEPA-Anything:一套预测框架打通七个领域,湿实验也给了背书","jepa-anything-orthogonal-predictive-factorization","2026-09-19T23:10:37+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"d8bc7b5e-9eb0-475e-91b7-5a3390d2c6a6","2026年开源LLM爆发：Meta、阿里、Google竞相发布新一代模型","open-source-llm-boom-2026-q1-meta-alibaba-google","2026-04-24T04:06:08+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"ace7a537-e2de-448f-811f-a028357aed42","FocusGUI:给全模态GUI智能体装上注意力滤镜","focusgui-omni-modal-gui-agents","2026-10-21T19:05:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"55663822-d49d-4c7e-a7c7-ba9770657c25","清华DBS动态分频:音源分离SI-SDR提升29%","dynamic-band-splitting-audio-separation","2026-10-21T15:02:40+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"b44fb160-9027-44cd-b355-a66869004abc","Nemotron 3.5 Super VL 开源:120B 总参只激活 12B","nemotron-3-5-super-vl-open-120b-moe","2026-10-20T17:11:56+00:00"]