[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-a-new-robot-ai-model-separates-knowing-from-doing":10,"sections":45},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":35,"tags":36,"sources":40,"feedback":44,"feedback_at":22,"cost_usd":44,"total_tokens":44},11361,"a-new-robot-ai-model-separates-knowing-from-doing","A New Robot AI Model Separates Knowing From Doing","Researchers built a robot AI model that separates conceptual knowledge from motor skills, letting it handle objects it never saw in training.","A new robot learning model claims to let robots handle objects they have never seen before, without collecting more training data.\n\nA paper posted to arXiv proposes w2VLA, a new design for vision-language-action models, the systems that let robots turn camera input and instructions into physical movement. The authors argue that today's dominant approach, fine-tuning billion-parameter vision-language models first on large robot datasets and then on task-specific demonstrations, tangles two different kinds of knowledge together: knowing what an object is, and knowing how to manipulate it. That tangling is why these models perform well on objects they trained on but get brittle fast when an object's shape, position, or label shifts even slightly. w2VLA instead feeds visual, spatial, and skill information into the robot's action pipeline through separate, interpretable channels rather than one large, opaque transformer, and the authors report this produces what they call unprecedented zero-shot transfer to objects the model never saw during training.\n\nIf that holds up outside the lab, it addresses one of robotics' most expensive constraints: every new object currently requires fresh demonstration data, because models cannot cleanly separate \"this is a mug\" from \"this is how I pick up a mug-shaped thing.\" A model that generalizes to dissimilar, unseen objects without new demos would shrink that data-collection tax substantially, which matters more for getting robots into messy real-world settings than any single manipulation benchmark score.\n\nThe abstract does not share benchmark numbers or head-to-head comparisons, so \"unprecedented\" remains the authors' own characterization until the full results get independent scrutiny.","[\"robotics\",\"vision-language-action models\",\"ai research\"]","2026-10-12T04:00:00.000Z","2026-10-12T10:17:33.600Z","2026-10-12T10:17:38.791Z","published",null,[24,30],{"id":25,"reviewer":26,"round":27,"reason":28,"status":29},"editor-r1","editor",1,"Fix the technical description: the source says w²VLA modulates the robot's state sequence with visual\u002Fspatial\u002Fskill information, not the 'action sequence' as the draft states — state and action are distinct inputs\u002Foutputs in VLA models and conflating them misstates how the method works.","resolved",{"id":31,"reviewer":32,"round":33,"reason":34,"status":29},"publisher-r2","publisher",2,"The closing line quotes \"unprecedented zero-shot transfer\" as if debunking a specific claim, but that phrase never appears anywhere earlier in the article, suggesting cut or leftover editing content that needs to be fixed before publishing.","ai",[37,38,39],"robotics","vision-language-action models","ai research",[41],{"name":42,"url":43},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.21496",0,{"sections":46},[47,51,56,61,66,71,76,81,86,91,96,101,106,111],{"name":48,"slug":35,"count":49,"latest_published_at":50},"AI",6967,"2026-10-12T15:20:00.000Z",{"name":52,"slug":53,"count":54,"latest_published_at":55},"Security","security",954,"2026-10-12T14:20:37.000Z",{"name":57,"slug":58,"count":59,"latest_published_at":60},"Policy","policy",496,"2026-10-11T20:06:43.000Z",{"name":62,"slug":63,"count":64,"latest_published_at":65},"Deals","deals",492,"2026-10-12T14:55:53.000Z",{"name":67,"slug":68,"count":69,"latest_published_at":70},"Hardware","hardware",237,"2026-10-12T13:24:45.000Z",{"name":72,"slug":73,"count":74,"latest_published_at":75},"Science","science",196,"2026-10-11T15:34:40.000Z",{"name":77,"slug":78,"count":79,"latest_published_at":80},"Consumer Tech","consumer-tech",192,"2026-10-12T14:07:13.000Z",{"name":82,"slug":83,"count":84,"latest_published_at":85},"Startups","startups",121,"2026-10-12T13:22:34.000Z",{"name":87,"slug":88,"count":89,"latest_published_at":90},"Software","software",115,"2026-10-11T07:34:45.000Z",{"name":92,"slug":93,"count":94,"latest_published_at":95},"Dev Tools","dev-tools",111,"2026-10-12T11:22:24.000Z",{"name":97,"slug":98,"count":99,"latest_published_at":100},"General","general",71,"2026-10-12T16:10:40.000Z",{"name":102,"slug":103,"count":104,"latest_published_at":105},"Gaming","gaming",59,"2026-10-09T11:43:43.000Z",{"name":107,"slug":108,"count":109,"latest_published_at":110},"Reviews","reviews",35,"2026-10-12T14:00:00.000Z",{"name":112,"slug":113,"count":114,"latest_published_at":115},"How-To","how-to",8,"2026-10-05T09:00:00.000Z"]