{"@context":"https://schema.org","@type":"NewsArticle","generatedAt":"2026-07-23T07:21:26.498Z","headline":"SeededGrasp：面向复杂场景与多机械臂的语言引导抓取框架","description":"研究者提出 SeededGrasp，一种数据高效的模块化框架，利用 VLM 零样本预测抓取种子点，再交由轻量级流匹配模型生成具体抓取姿态，将高层语义推理与底层几何执行解耦。该方法无需端到端训练即可支持多机械臂，在 610 个杂乱场景、334 个物体、3 种夹爪的合成数据集（含 256 万次抓取）上训练，仿真成功率达 72%，真实实验成功率达 78%。项目已开源代码与数据。","url":"https://www.aioga.com/news/cmrwwy3d8038crobh6zlfdrxo/","mainEntityOfPage":"https://www.aioga.com/news/cmrwwy3d8038crobh6zlfdrxo/","datePublished":"2026-07-22T00:00:00.000Z","dateModified":"2026-07-22T00:00:00.000Z","inLanguage":"zh-CN","publisher":{"@type":"NewsMediaOrganization","name":"Aioga","url":"https://www.aioga.com"},"citation":["https://arxiv.org/abs/2607.20207","https://aihot.virxact.com/items/cmrwwy3d8038crobh6zlfdrxo"],"canonicalUrl":"https://www.aioga.com/news/cmrwwy3d8038crobh6zlfdrxo/","directAnswer":{"@type":"Answer","text":"Aioga 编辑摘要：研究者提出 SeededGrasp，一种数据高效的模块化框架，利用 VLM 零样本预测抓取种子点，再交由轻量级流匹配模型生成具体抓取姿态，将高层语义推理与底层几何执行解耦。 Aioga 将其归入「论文研究」方向，重点关注它对真实使用和行业竞争的影响。","url":"https://www.aioga.com/news/cmrwwy3d8038crobh6zlfdrxo/","dateCreated":"2026-07-22T00:00:00.000Z","author":{"@type":"Organization","@id":"https://www.aioga.com/authors/aioga-editorial/#editorial-team","name":"Aioga Editorial Team","url":"https://www.aioga.com/authors/aioga-editorial/"}},"evidence":[{"@type":"CreativeWork","name":"arXiv source article","url":"https://arxiv.org/abs/2607.20207","datePublished":"2026-07-22T00:00:00.000Z","provider":{"@type":"Organization","name":"arXiv","url":"https://arxiv.org/abs/2607.20207"}},{"@type":"CreativeWork","name":"AIHot archive record","url":"https://aihot.virxact.com/items/cmrwwy3d8038crobh6zlfdrxo","datePublished":"2026-07-22T00:00:00.000Z","provider":{"@type":"Organization","name":"AIHot","url":"https://aihot.virxact.com/items/cmrwwy3d8038crobh6zlfdrxo"}}],"aggregationSource":"HuggingFace Daily Papers（社区热门论文）","originalPublisher":{"name":"arXiv","url":"https://arxiv.org/abs/2607.20207"},"article":{"id":"cmrwwy3d8038crobh6zlfdrxo","slug":"cmrwwy3d8038crobh6zlfdrxo","url":"https://www.aioga.com/news/cmrwwy3d8038crobh6zlfdrxo/","title":"SeededGrasp：面向复杂场景与多机械臂的语言引导抓取框架","title_en":"SeededGrasp： Language-Guided Grasping in Complex Scenes with Multiple Embodiments","summary":"研究者提出 SeededGrasp，一种数据高效的模块化框架，利用 VLM 零样本预测抓取种子点，再交由轻量级流匹配模型生成具体抓取姿态，将高层语义推理与底层几何执行解耦。该方法无需端到端训练即可支持多机械臂，在 610 个杂乱场景、334 个物体、3 种夹爪的合成数据集（含 256 万次抓取）上训练，仿真成功率达 72%，真实实验成功率达 78%。项目已开源代码与数据。","source":"HuggingFace Daily Papers（社区热门论文）","sourceUrl":"https://arxiv.org/abs/2607.20207","aiHotUrl":"https://aihot.virxact.com/items/cmrwwy3d8038crobh6zlfdrxo","publishedAt":"2026-07-22T00:00:00.000Z","category":"论文研究","score":53,"selected":false,"articleBody":["arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.","Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs ：https://info.arxiv.org/labs/index.html."],"articleImages":[{"sourceUrl":"https://arxiv.org/static/base/1.0.1/images/funders/simons-foundation.png","alt":"Simons Foundation","afterParagraph":1,"url":"/media/articles/cmrwwy3d8038crobh6zlfdrxo/e2d7f38d62f5ca91.png"},{"sourceUrl":"https://arxiv.org/static/base/1.0.1/images/funders/simons-foundation-international.png","alt":"Simons Foundation International","afterParagraph":1,"url":"/media/articles/cmrwwy3d8038crobh6zlfdrxo/1d56e29c5557cbdc.png"}],"mediaStatus":"ok","articleBodyZh":["arXivLabs 是一个框架，允许合作者直接在我们的网站上开发和分享新的 arXiv 功能。","有一个可以为 arXiv 社区增加价值的项目想法吗？了解更多关于 arXivLabs 的信息：https://info.arxiv.org/labs/index.html。"],"translationStatus":"translated","bodyOrigin":"source-page","editorial":{"summary":"Aioga 编辑摘要：研究者提出 SeededGrasp，一种数据高效的模块化框架，利用 VLM 零样本预测抓取种子点，再交由轻量级流匹配模型生成具体抓取姿态，将高层语义推理与底层几何执行解耦。 Aioga 将其归入「论文研究」方向，重点关注它对真实使用和行业竞争的影响。","background":"背景分析：模型与研究类动态需要结合能力边界、开放方式、成本、可用性和真实任务表现判断，单项指标领先不等于已经形成稳定采用。","viewpoint":"Aioga 判断：这条动态更适合作为行业观察信号，当前信息足以建立线索，但不足以推导长期结论。","implications":"影响分析：对相关团队而言，短期应先核对来源、可用范围和实际成本，再判断是否值得接入或跟进。","nextStep":"后续观察：继续观察官方文档、实际可用性、价格变化、开发者反馈和竞品回应。","evidenceRefs":["title","summary","articleBody"],"confidence":"medium","status":"published","aiGenerated":false,"autoApproved":true,"generatedBy":"rule-safe-fallback","generatedAt":"2026-07-23T07:30:40.463Z","sourceHash":"720a5db673ea4ca4","validation":{"passed":true,"mode":"rule-safe-fallback","checks":["schema","length","source-attribution","no-html"]}},"tags":["论文研究","HuggingFace Daily Papers（社区热门论文）"],"translations":{"zh-CN":{"title":"SeededGrasp：面向复杂场景与多机械臂的语言引导抓取框架","summary":"研究者提出 SeededGrasp，一种数据高效的模块化框架，利用 VLM 零样本预测抓取种子点，再交由轻量级流匹配模型生成具体抓取姿态，将高层语义推理与底层几何执行解耦。该方法无需端到端训练即可支持多机械臂，在 610 个杂乱场景、334 个物体、3 种夹爪的合成数据集（含 256 万次抓取）上训练，仿真成功率达 72%，真实实验成功率达 78%。项目已开源代码与数据。","category":"论文研究","source":"arXiv","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"SeededGrasp：面向复杂场景与多机械臂的语言引导抓取框架 - Aioga AI资讯","description":"研究者提出 SeededGrasp，一种数据高效的模块化框架，利用 VLM 零样本预测抓取种子点，再交由轻量级流匹配模型生成具体抓取姿态，将高层语义推理与底层几何执行解耦。该方法无需端到端训练即可支持多机械臂，在 610 个杂乱场景、334 个物体、3 种夹爪的合成数据集（含 256 万次抓取）上训练，仿真成功率达 72%，真实实验成功率达 78%。项目已开...","url":"https://www.aioga.com/news/cmrwwy3d8038crobh6zlfdrxo/"},"en":{"title":"SeededGrasp: A Language-Guided Grasping Framework for Complex Scenes and Multiple Robotic Arms","summary":"Researchers proposed SeededGrasp, a data-efficient modular framework that leverages VLMs for zero-shot prediction of grasp seed points, which are then passed to a lightweight flow-matching model to generate specific grasp poses, decoupling high-level semantic reasoning from low-level geometric execution. This method does not require end-to-end training and can support multiple robotic arms. It was trained on a synthetic dataset of 610 cluttered scenes, 334 objects, and 3 types of grippers (including 2.56 million grasps), achieving a simulation success rate of 72% and a real-world experimental success rate of 78%. The project has open-sourced its code and data.","category":"Research","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"SeededGrasp: A Language-Guided Grasping Framework for Complex Scenes and Multiple Robotic Arms - Aioga AI News","description":"Researchers proposed SeededGrasp, a data-efficient modular framework that leverages VLMs for zero-shot prediction of grasp seed points, which are then passed to a lightweight flow-...","url":"https://www.aioga.com/en/news/cmrwwy3d8038crobh6zlfdrxo/","contentTranslated":true,"sourceHash":"2073365ff5b0c4e9","translatedAt":"2026-07-23T03:42:13.274Z"},"ja":{"title":"SeededGrasp：複雑なシナリオと複数のロボットアーム向けの言語誘導型把持フレームワーク","summary":"研究者は SeededGrasp を提案しました。これはデータ効率の高いモジュラー型フレームワークで、VLM によるゼロショットでの把持シード点予測を利用し、その後軽量なフローマッチングモデルにより具体的な把持姿勢を生成し、高次の意味推論と低次の幾何学的実行を切り離します。この方法はエンドツーエンドの訓練を必要とせず、複数のロボットアームに対応可能です。610 の乱雑なシーン、334 の物体、3 種類のグリッパーを含む合成データセット（256 万回の把持を含む）で訓練したところ、シミュレーションの成功率は 72%、実験環境での成功率は 78% でした。プロジェクトはコードとデータをオープンソース化しています。","category":"論文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"SeededGrasp：複雑なシナリオと複数のロボットアーム向けの言語誘導型把持フレームワーク - Aioga AIニュース","description":"研究者は SeededGrasp を提案しました。これはデータ効率の高いモジュラー型フレームワークで、VLM によるゼロショットでの把持シード点予測を利用し、その後軽量なフローマッチングモデルにより具体的な把持姿勢を生成し、高次の意味推論と低次の幾何学的実行を切り離します。この方法はエンドツーエンドの訓練を必要とせず、複数のロボットアームに対応可能です。61...","url":"https://www.aioga.com/ja/news/cmrwwy3d8038crobh6zlfdrxo/","contentTranslated":true,"sourceHash":"2073365ff5b0c4e9","translatedAt":"2026-07-23T03:42:24.800Z"},"ko":{"title":"SeededGrasp: 복잡한 장면과 다중 로봇 팔을 위한 언어 기반 잡기 프레임워크","summary":"연구자들은 SeededGrasp를 제안했는데, 이는 데이터 효율적인 모듈형 프레임워크로, VLM을 이용해 제로샷으로 그립 시드 포인트를 예측하고, 이어서 경량 스트림 매칭 모델에게 구체적인 그립 자세를 생성하도록 하며, 고수준 의미 추론과 저수준 기하학적 실행을 분리합니다. 이 방법은 엔드투엔드 학습이 필요 없으며 다중 로봇 팔을 지원할 수 있습니다. 610개의 혼잡한 장면, 334개의 물체, 3종류의 그립 퍼를 포함한 합성 데이터셋(총 256만 번의 그립)을 학습한 결과, 시뮬레이션 성공률은 72%, 실제 실험 성공률은 78%에 달했습니다. 프로젝트는 코드와 데이터를 오픈소스로 공개했습니다.","category":"연구","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"SeededGrasp: 복잡한 장면과 다중 로봇 팔을 위한 언어 기반 잡기 프레임워크 - Aioga AI 뉴스","description":"연구자들은 SeededGrasp를 제안했는데, 이는 데이터 효율적인 모듈형 프레임워크로, VLM을 이용해 제로샷으로 그립 시드 포인트를 예측하고, 이어서 경량 스트림 매칭 모델에게 구체적인 그립 자세를 생성하도록 하며, 고수준 의미 추론과 저수준 기하학적 실행을 분리합니다. 이 방법은 엔드투엔드 학습이 필요 없으며 다중...","url":"https://www.aioga.com/ko/news/cmrwwy3d8038crobh6zlfdrxo/","contentTranslated":true,"sourceHash":"2073365ff5b0c4e9","translatedAt":"2026-07-23T03:43:18.122Z"},"es":{"title":"SeededGrasp: Marco de agarre guiado por lenguaje para escenarios complejos y múltiples brazos robóticos","summary":"Los investigadores propusieron SeededGrasp, un marco modular eficiente en cuanto a datos, que utiliza la predicción de puntos semilla mediante VLM en cero disparos, y luego un modelo ligero de emparejamiento de flujo genera posturas de agarre específicas, desacoplando la inferencia semántica de alto nivel de la ejecución geométrica de bajo nivel. Este método no requiere entrenamiento de extremo a extremo y puede soportar múltiples brazos robóticos. Se entrenó en un conjunto de datos sintético con 610 escenarios desordenados, 334 objetos y 3 tipos de pinzas (incluyendo 2,56 millones de agarres), logrando una tasa de éxito en simulación del 72% y una tasa de éxito en experimentos reales del 78%. El proyecto ha abierto el código y los datos.","category":"Investigación","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"SeededGrasp: Marco de agarre guiado por lenguaje para escenarios complejos y múltiples brazos robóticos - Aioga Noticias de IA","description":"Los investigadores propusieron SeededGrasp, un marco modular eficiente en cuanto a datos, que utiliza la predicción de puntos semilla mediante VLM en cero disparos, y luego un mode...","url":"https://www.aioga.com/es/news/cmrwwy3d8038crobh6zlfdrxo/","contentTranslated":true,"sourceHash":"2073365ff5b0c4e9","translatedAt":"2026-07-23T03:43:06.608Z"},"fr":{"title":"SeededGrasp : Cadre de préhension guidée par langage pour des scénarios complexes et plusieurs bras robotiques","summary":"Les chercheurs ont proposé SeededGrasp, un cadre modulaire efficace en termes de données, utilisant la prédiction de points de saisie par VLM en zéro-shot, puis confiant à un modèle léger de correspondance de flux la génération de postures de saisie spécifiques, séparant ainsi le raisonnement sémantique de haut niveau de l'exécution géométrique de bas niveau. Cette méthode peut supporter plusieurs bras robotiques sans entraînement de bout en bout. Elle a été entraînée sur un ensemble de données synthétiques comprenant 610 scènes désordonnées, 334 objets et 3 types de pinces (totalisant 2,56 millions de saisies), avec un taux de réussite en simulation de 72 % et un taux de réussite en expérimentation réelle de 78 %. Le code et les données du projet ont été rendus open source.","category":"Recherche","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"SeededGrasp : Cadre de préhension guidée par langage pour des scénarios complexes et plusieurs bras robotiques - Aioga Actualités IA","description":"Les chercheurs ont proposé SeededGrasp, un cadre modulaire efficace en termes de données, utilisant la prédiction de points de saisie par VLM en zéro-shot, puis confiant à un modèl...","url":"https://www.aioga.com/fr/news/cmrwwy3d8038crobh6zlfdrxo/","contentTranslated":true,"sourceHash":"2073365ff5b0c4e9","translatedAt":"2026-07-23T03:44:03.133Z"},"de":{"title":"SeededGrasp: Ein sprachgesteuertes Greifrahmenwerk für komplexe Szenarien und mehrere Roboterarme","summary":"Die Forscher schlagen SeededGrasp vor, einen daten-effizienten modularen Rahmen, der VLM zur Null-Proben-Vorhersage von Greifpunkten nutzt und diese dann an ein leichtgewichtiges Fluss-Matching-Modell übergibt, um konkrete Greifhaltungen zu erzeugen, wodurch hochrangiges semantisches Schlussfolgern von der niedrigstufigen geometrischen Ausführung entkoppelt wird. Diese Methode benötigt kein End-to-End-Training und unterstützt mehrere Roboterarme. Sie wurde an einem synthetischen Datensatz mit 610 unordentlichen Szenen, 334 Objekten und 3 Greifarmen (einschließlich 2,56 Millionen Greifversuchen) trainiert, mit einer Simulations-Erfolgsrate von 72 % und einer realen Experiment-Erfolgsrate von 78 %. Der Projektcode und die Daten wurden bereits veröffentlicht.","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"SeededGrasp: Ein sprachgesteuertes Greifrahmenwerk für komplexe Szenarien und mehrere Roboterarme - Aioga KI-News","description":"Die Forscher schlagen SeededGrasp vor, einen daten-effizienten modularen Rahmen, der VLM zur Null-Proben-Vorhersage von Greifpunkten nutzt und diese dann an ein leichtgewichtiges F...","url":"https://www.aioga.com/de/news/cmrwwy3d8038crobh6zlfdrxo/","contentTranslated":true,"sourceHash":"2073365ff5b0c4e9","translatedAt":"2026-07-23T03:44:01.390Z"},"pt-BR":{"title":"SeededGrasp: Estrutura de captura guiada por linguagem para cenários complexos e múltiplos braços robóticos","summary":"Os pesquisadores propuseram o SeededGrasp, uma estrutura modular eficiente em dados, que utiliza VLM para prever pontos de captura sem exemplos e, em seguida, passa a um modelo leve de correspondência de fluxo para gerar posturas de captura específicas, desacoplando o raciocínio semântico de alto nível da execução geométrica de baixo nível. Este método não requer treinamento de ponta a ponta e pode suportar múltiplos braços robóticos. Ele foi treinado em um conjunto de dados sintético com 610 cenários desordenados, 334 objetos e 3 tipos de garras (incluindo 2,56 milhões de capturas), alcançando uma taxa de sucesso de simulação de 72% e uma taxa de sucesso em experimentos reais de 78%. O projeto já disponibilizou o código e os dados como open source.","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"SeededGrasp: Estrutura de captura guiada por linguagem para cenários complexos e múltiplos braços robóticos - Aioga Notícias de IA","description":"Os pesquisadores propuseram o SeededGrasp, uma estrutura modular eficiente em dados, que utiliza VLM para prever pontos de captura sem exemplos e, em seguida, passa a um modelo lev...","url":"https://www.aioga.com/pt-BR/news/cmrwwy3d8038crobh6zlfdrxo/","contentTranslated":true,"sourceHash":"2073365ff5b0c4e9","translatedAt":"2026-07-23T03:44:50.262Z"},"ru":{"title":"SeededGrasp: Языковая направляемая система захвата для сложных сценариев и множественных роботизированных рук","summary":"Исследователи предложили SeededGrasp, высокоэффективную по данным модульную структуру, использующую VLM для нулевого образца предсказания точек захвата, а затем передающую их легковесной модели сопоставления потоков для генерации конкретной позы захвата, отделяя высокоуровневое семантическое рассуждение от низкоуровневого геометрического исполнения. Этот метод не требует обучения от конца до конца и может поддерживать несколько манипуляторов. Он был обучен на синтетическом наборе данных из 610 беспорядочных сцен, 334 объектов и 3 типов захватов (с включением 2,56 млн захватов), успешно демонстрируя в симуляции 72% успешных случаев и 78% в реальных экспериментах. Проект уже опубликовал исходный код и данные.","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"SeededGrasp: Языковая направляемая система захвата для сложных сценариев и множественных роботизированных рук - Aioga Новости ИИ","description":"Исследователи предложили SeededGrasp, высокоэффективную по данным модульную структуру, использующую VLM для нулевого образца предсказания точек захвата, а затем передающую их легко...","url":"https://www.aioga.com/ru/news/cmrwwy3d8038crobh6zlfdrxo/","contentTranslated":true,"sourceHash":"2073365ff5b0c4e9","translatedAt":"2026-07-23T03:44:53.077Z"},"ar":{"title":"SeededGrasp: إطار عمل للإمساك موجه باللغة للمشاهد المعقدة والذراع الروبوتية المتعددة","summary":"اقترح الباحثون إطار عمل معياري عالي الكفاءة للبيانات يُسمى SeededGrasp، يستخدم نموذج اللغة البصري (VLM) للتنبؤ بنقاط القبض بدون عينات، ثم يتم تسليمها إلى نموذج مطابقة التدفق خفيف الوزن لتوليد الوضعيات الدقيقة للقبض، ما يفصل بين الاستنتاج الدلالي العالي المستوى والتنفيذ الهندسي الأساسي. هذه الطريقة لا تتطلب تدريب شامل من البداية للنهاية ويمكن أن تدعم العديد من الأذرع الآلية. تم التدريب على مجموعة بيانات تركيبية تضم 610 مشاهد فوضوية، 334 جسمًا، و3 أنواع من المخالب (بما في ذلك 2.56 مليون عملية قبض)، وبلغ معدل النجاح في المحاكاة 72٪، بينما وصل معدل النجاح في التجارب الواقعية إلى 78٪. تم نشر الشفرة والبيانات كمشروع مفتوح المصدر.","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"SeededGrasp: إطار عمل للإمساك موجه باللغة للمشاهد المعقدة والذراع الروبوتية المتعددة - Aioga أخبار الذكاء الاصطناعي","description":"اقترح الباحثون إطار عمل معياري عالي الكفاءة للبيانات يُسمى SeededGrasp، يستخدم نموذج اللغة البصري (VLM) للتنبؤ بنقاط القبض بدون عينات، ثم يتم تسليمها إلى نموذج مطابقة التدفق خفيف ا...","url":"https://www.aioga.com/ar/news/cmrwwy3d8038crobh6zlfdrxo/","contentTranslated":true,"sourceHash":"2073365ff5b0c4e9","translatedAt":"2026-07-23T03:45:38.736Z"},"hi":{"title":"SeededGrasp: जटिल परिदृश्यों और बहु-यांत्रिक भुजाओं के लिए भाषा-निर्देशित पकड़ फ्रेमवर्क","summary":"शोधकर्ताओं ने SeededGrasp पेश किया, एक डेटा-कुशल मॉड्यूलर फ्रेमवर्क, जो VLM ज़ीरो-शॉट के माध्यम से पकड़ने के बीज बिंदुओं की भविष्यवाणी करता है, और फिर हल्के बहाव मिलान मॉडल को विशिष्ट पकड़ने की मुद्रा उत्पन्न करने देता है, जिससे उच्च-स्तरीय सेमांटिक तर्क और निचले-स्तरीय ज्यामितीय निष्पादन अलग हो जाते हैं। इस विधि को एंड-टू-एंड प्रशिक्षण की आवश्यकता नहीं है और यह कई रोबोटिक भुजाओं का समर्थन कर सकती है। यह 610 अस्त-व्यस्त दृश्य, 334 वस्तुएं, और 3 प्रकार के ग्रैपर वाले सिंथेटिक डेटासेट (256 मिलियन पकड़ों सहित) पर प्रशिक्षण प्राप्त हुआ, जिसमें सिमुलेशन सफलता दर 72% और वास्तविक प्रयोग में सफलता दर 78% थी। प्रोजेक्ट के कोड और डेटा को ओपन-सोर्स किया गया है।","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"SeededGrasp: जटिल परिदृश्यों और बहु-यांत्रिक भुजाओं के लिए भाषा-निर्देशित पकड़ फ्रेमवर्क - Aioga AI समाचार","description":"शोधकर्ताओं ने SeededGrasp पेश किया, एक डेटा-कुशल मॉड्यूलर फ्रेमवर्क, जो VLM ज़ीरो-शॉट के माध्यम से पकड़ने के बीज बिंदुओं की भविष्यवाणी करता है, और फिर हल्के बहाव मिलान मॉडल को विशि...","url":"https://www.aioga.com/hi/news/cmrwwy3d8038crobh6zlfdrxo/","contentTranslated":true,"sourceHash":"2073365ff5b0c4e9","translatedAt":"2026-07-23T03:45:43.747Z"},"it":{"title":"SeededGrasp: un framework di presa guidata dal linguaggio per scenari complessi e bracci robotici multipli","summary":"I ricercatori hanno proposto SeededGrasp, un framework modulare ad alta efficienza dei dati, che utilizza VLM per prevedere punti di presa in modalità zero-shot, quindi li passa a un modello leggero di corrispondenza dei flussi per generare pose di presa specifiche, separando il ragionamento semantico di alto livello dall'esecuzione geometrica di basso livello. Questo metodo supporta più bracci robotici senza necessità di addestramento end-to-end. È stato addestrato su un dataset sintetico con 610 scenari disordinati, 334 oggetti e 3 tipi di pinze (contenente 2,56 milioni di prese), ottenendo un tasso di successo del 72% in simulazione e del 78% in esperimenti reali. Il progetto ha rilasciato il codice e i dati come open source.","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"SeededGrasp: un framework di presa guidata dal linguaggio per scenari complessi e bracci robotici multipli - Aioga Notizie IA","description":"I ricercatori hanno proposto SeededGrasp, un framework modulare ad alta efficienza dei dati, che utilizza VLM per prevedere punti di presa in modalità zero-shot, quindi li passa a...","url":"https://www.aioga.com/it/news/cmrwwy3d8038crobh6zlfdrxo/","contentTranslated":true,"sourceHash":"2073365ff5b0c4e9","translatedAt":"2026-07-23T03:46:26.488Z"},"nl":{"title":"SeededGrasp: Een taalgestuurd grijpraamwerk voor complexe scenario's en meerdere robotarmen","summary":"Onderzoekers hebben SeededGrasp voorgesteld, een data-efficiënt modulair kader dat VLM gebruikt voor zero-shot voorspelling van grijpbare punten, die vervolgens door een lichtgewicht flow-matchingmodel worden omgezet in specifieke grijpposes, waardoor hoge-niveau semantische redenering wordt gescheiden van lage-niveau geometrische uitvoering. Deze methode vereist geen end-to-end training en kan meerdere robotarmen ondersteunen. Het is getraind op een synthetische dataset met 610 rommelige scènes, 334 objecten en 3 grijpers (inclusief 2,56 miljoen grijppogingen), met een simulatieresultaat van 72% en een succesrate in echte experimenten van 78%. Het project heeft de code en data open-source beschikbaar gesteld.","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"SeededGrasp: Een taalgestuurd grijpraamwerk voor complexe scenario's en meerdere robotarmen - Aioga AI-nieuws","description":"Onderzoekers hebben SeededGrasp voorgesteld, een data-efficiënt modulair kader dat VLM gebruikt voor zero-shot voorspelling van grijpbare punten, die vervolgens door een lichtgewic...","url":"https://www.aioga.com/nl/news/cmrwwy3d8038crobh6zlfdrxo/","contentTranslated":true,"sourceHash":"2073365ff5b0c4e9","translatedAt":"2026-07-23T03:46:32.047Z"},"tr":{"title":"SeededGrasp: Karmaşık Senaryolar ve Çoklu Robot Kolları için Dil Yoluyla Yönlendirilen Kavrama Çerçevesi","summary":"Araştırmacılar, VLM sıfır örnek tahmini ile yakalama tohum noktalarını kullanan ve ardından hafif akış eşleştirme modeli tarafından belirli yakalama duruşlarını üreten, yüksek veri verimliliğine sahip modüler bir çerçeve olan SeededGrasp'ı öneriyor; bu, üst düzey anlamsal çıkarımı alt düzey geometrik yürütmeden ayırır. Bu yöntem uçtan uca eğitim gerektirmeden çoklu robot kollarını destekleyebilir, 610 dağınık sahne, 334 nesne ve 3 tür pençe içeren sentetik veri kümesinde (2,56 milyon yakalama dahil) eğitildiğinde, simülasyon başarı oranı %72 ve gerçek deneylerde başarı oranı %78'dir. Projenin kodu ve verileri zaten açık kaynak haline getirilmiştir.","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"SeededGrasp: Karmaşık Senaryolar ve Çoklu Robot Kolları için Dil Yoluyla Yönlendirilen Kavrama Çerçevesi - Aioga AI Haberleri","description":"Araştırmacılar, VLM sıfır örnek tahmini ile yakalama tohum noktalarını kullanan ve ardından hafif akış eşleştirme modeli tarafından belirli yakalama duruşlarını üreten, yüksek veri...","url":"https://www.aioga.com/tr/news/cmrwwy3d8038crobh6zlfdrxo/","contentTranslated":true,"sourceHash":"2073365ff5b0c4e9","translatedAt":"2026-07-23T03:47:20.063Z"},"vi":{"title":"SeededGrasp: Khung điều khiển nắm bắt dựa trên ngôn ngữ hướng tới các cảnh phức tạp và nhiều cánh tay cơ học","summary":"Các nhà nghiên cứu đề xuất SeededGrasp, một khung mô-đun hiệu quả về dữ liệu, sử dụng VLM để dự đoán không mẫu các điểm hạt giống, sau đó giao cho mô hình khớp luồng nhẹ để tạo ra tư thế cầm nắm cụ thể, tách rời suy luận ngữ nghĩa cấp cao và thực thi hình học cấp thấp. Phương pháp này không cần đào tạo end-to-end vẫn có thể hỗ trợ nhiều cánh tay robot, được huấn luyện trên bộ dữ liệu tổng hợp gồm 610 cảnh hỗn loạn, 334 vật thể, 3 loại càng kẹp (bao gồm 2,56 triệu lần cầm nắm), tỷ lệ thành công mô phỏng đạt 72%, tỷ lệ thành công trong thực nghiệm thực tế đạt 78%. Dự án đã công khai mã và dữ liệu.","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"SeededGrasp: Khung điều khiển nắm bắt dựa trên ngôn ngữ hướng tới các cảnh phức tạp và nhiều cánh tay cơ học - Tin tức AI Aioga","description":"Các nhà nghiên cứu đề xuất SeededGrasp, một khung mô-đun hiệu quả về dữ liệu, sử dụng VLM để dự đoán không mẫu các điểm hạt giống, sau đó giao cho mô hình khớp luồng nhẹ để tạo ra...","url":"https://www.aioga.com/vi/news/cmrwwy3d8038crobh6zlfdrxo/","contentTranslated":true,"sourceHash":"2073365ff5b0c4e9","translatedAt":"2026-07-23T03:47:23.215Z"},"id":{"title":"SeededGrasp: Kerangka Penangkapan yang Dipandu Bahasa untuk Skenario Kompleks dan Multi-lengan Robot","summary":"Para peneliti mengajukan SeededGrasp, sebuah kerangka modular yang efisien dalam penggunaan data, yang memanfaatkan VLM untuk prediksi titik benih tanpa contoh, kemudian diserahkan ke model pencocokan aliran ringan untuk menghasilkan pose penangkapan yang spesifik, sehingga memisahkan penalaran semantik tingkat tinggi dari eksekusi geometris tingkat rendah. Metode ini tidak memerlukan pelatihan end-to-end dan dapat mendukung multi-lengan robot. Dilatih pada dataset sintetis dengan 610 skenario berantakan, 334 objek, dan 3 jenis gripper (termasuk 2,56 juta percobaan penangkapan), tingkat keberhasilan simulasi mencapai 72%, dan tingkat keberhasilan eksperimen nyata mencapai 78%. Kode dan data proyek telah dibuka.","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"SeededGrasp: Kerangka Penangkapan yang Dipandu Bahasa untuk Skenario Kompleks dan Multi-lengan Robot - Berita AI Aioga","description":"Para peneliti mengajukan SeededGrasp, sebuah kerangka modular yang efisien dalam penggunaan data, yang memanfaatkan VLM untuk prediksi titik benih tanpa contoh, kemudian diserahkan...","url":"https://www.aioga.com/id/news/cmrwwy3d8038crobh6zlfdrxo/","contentTranslated":true,"sourceHash":"2073365ff5b0c4e9","translatedAt":"2026-07-23T03:48:03.647Z"},"th":{"title":"SeededGrasp: กรอบการจับโดยใช้ภาษาที่มุ่งเน้นไปที่สถานการณ์ซับซ้อนและแขนกลหลายตัว","summary":"นักวิจัยได้เสนอ SeededGrasp ซึ่งเป็นกรอบงานแบบโมดูลาร์ที่ใช้ข้อมูลอย่างมีประสิทธิภาพ ใช้ VLM ในการทำนายจุดเมล็ดการจับแบบศูนย์ตัวอย่าง จากนั้นส่งต่อไปยังโมเดลจับคู่แบบไหลที่มีน้ำหนักเบาเพื่อสร้างท่าทางการจับที่ชัดเจน ทำให้การใช้เหตุผลเชิงความหมายระดับสูงและการดำเนินการทางเรขาคณิตระดับต่ำถูกแยกออกจากกัน วิธีการนี้ไม่ต้องการการฝึกแบบ end-to-end ก็สามารถรองรับแขนกลหลายตัวได้ โดยฝึกบนชุดข้อมูลสังเคราะห์ที่มี 610 ฉากสับสนวุ่นวาย, 334 วัตถุ, และ 3 ประเภทกรงเล็บ (รวมการจับ 2.56 ล้านครั้ง) อัตราความสำเร็จในการจำลองอยู่ที่ 72% และอัตราความสำเร็จในการทดลองจริงอยู่ที่ 78% โครงการได้เปิดเผยรหัสและข้อมูลแล้ว","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"SeededGrasp: กรอบการจับโดยใช้ภาษาที่มุ่งเน้นไปที่สถานการณ์ซับซ้อนและแขนกลหลายตัว - ข่าว AI Aioga","description":"นักวิจัยได้เสนอ SeededGrasp ซึ่งเป็นกรอบงานแบบโมดูลาร์ที่ใช้ข้อมูลอย่างมีประสิทธิภาพ ใช้ VLM ในการทำนายจุดเมล็ดการจับแบบศูนย์ตัวอย่าง จากนั้นส่งต่อไปยังโมเดลจับคู่แบบไหลที่มีน้ำหนั...","url":"https://www.aioga.com/th/news/cmrwwy3d8038crobh6zlfdrxo/","contentTranslated":true,"sourceHash":"2073365ff5b0c4e9","translatedAt":"2026-07-23T03:48:16.559Z"},"pl":{"title":"SeededGrasp: Ramy chwytania sterowane językowo dla złożonych scen i wielu ramion robotycznych","summary":"Badacze zaproponowali SeededGrasp, modułową ramę efektywną pod względem danych, która wykorzystuje VLM do przewidywania punktów chwytania w trybie zero-shot, a następnie przekazuje je do lekkiego modelu dopasowania przepływu w celu wygenerowania konkretnych postaw chwytania, rozdzielając wysokopoziomowe wnioskowanie semantyczne od wykonania geometrycznego na niskim poziomie. Metoda ta nie wymaga treningu end-to-end, aby wspierać wiele robotów, była trenowana na syntetycznym zestawie danych obejmującym 610 chaotycznych scen, 334 obiekty i 3 rodzaje chwytaków (łącznie 2,56 miliona prób chwytania), osiągając w symulacji skuteczność 72% oraz 78% w eksperymentach rzeczywistych. Projekt udostępnił już kod i dane.","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"SeededGrasp: Ramy chwytania sterowane językowo dla złożonych scen i wielu ramion robotycznych - Aioga Wiadomości AI","description":"Badacze zaproponowali SeededGrasp, modułową ramę efektywną pod względem danych, która wykorzystuje VLM do przewidywania punktów chwytania w trybie zero-shot, a następnie przekazuje...","url":"https://www.aioga.com/pl/news/cmrwwy3d8038crobh6zlfdrxo/","contentTranslated":true,"sourceHash":"2073365ff5b0c4e9","translatedAt":"2026-07-23T03:49:03.643Z"}}}}