[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-robots-that-rebuild-their-failures-in-sim-to-get-better":10,"sections":35},{"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":24,"tags":25,"sources":30,"feedback":34,"feedback_at":22,"cost_usd":34,"total_tokens":34},8737,"robots-that-rebuild-their-failures-in-sim-to-get-better","Robots That Rebuild Their Failures in Sim to Get Better","A new framework named F4R turns robot failures into simulated retraining, lifting out-of-distribution success by nearly 19 points over a baseline.","Robots that fail at a task usually just fail at it again. A new framework aims to make failure the thing that makes them better.\n\nThe system, called F4R, watches a robot's failed attempts, automatically figures out what went wrong, and rebuilds that exact failure as a simulated tabletop scene - same objects, same physical setup. The robot's policy is then retrained in that simulation through a mix of failure-conditioned training and targeted reinforcement learning, then redeployed in the real world. Any new failures loop back into another round of reconstruction and training. Across four manipulation tasks, the resulting policy hit 93.75% success on familiar tasks and 90% on tasks it hadn't seen, beating a budget-matched baseline by 18.75 percentage points on the unfamiliar ones.\n\nThe bottleneck for real-world robots has always been the same: you can't collect enough real demonstrations of every way a task can go sideways, and trying to is slow, expensive, and occasionally unsafe. F4R's answer is to turn the robot's own mistakes into training material, simulated rather than re-recorded, without adding a single new real-world demonstration. That's the more useful number here than the accuracy figures - the same data budget produced a policy that generalizes noticeably further.\n\nFour tabletop tasks is a solid proof of concept, not proof this holds up on a warehouse floor. The real test is whether rebuilding the failure in simulation still works once failures get messier than a dropped block.","[\"robotics\",\"ai-research\",\"reinforcement-learning\",\"simulation\"]","2026-09-30T04:00:00.000Z","2026-09-30T22:47:39.915Z","2026-09-30T22:47:45.464Z","published",null,[],"ai",[26,27,28,29],"robotics","ai-research","reinforcement-learning","simulation",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.35575",0,{"sections":36},[37,41,45,49,54,59,63,68,73,77,82,87,92,97],{"name":38,"slug":24,"count":39,"latest_published_at":40},"AI",5568,"2026-10-01T04:00:00.000Z",{"name":42,"slug":43,"count":44,"latest_published_at":40},"Security","security",815,{"name":46,"slug":47,"count":48,"latest_published_at":40},"Policy","policy",430,{"name":50,"slug":51,"count":52,"latest_published_at":53},"Deals","deals",298,"2026-09-30T21:00:26.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":58},"Hardware","hardware",196,"2026-09-30T13:00:00.000Z",{"name":60,"slug":61,"count":62,"latest_published_at":40},"Science","science",163,{"name":64,"slug":65,"count":66,"latest_published_at":67},"Consumer Tech","consumer-tech",149,"2026-09-30T22:57:11.000Z",{"name":69,"slug":70,"count":71,"latest_published_at":72},"Dev Tools","dev-tools",93,"2026-10-01T02:30:48.000Z",{"name":74,"slug":75,"count":71,"latest_published_at":76},"Software","software","2026-09-30T21:41:11.000Z",{"name":78,"slug":79,"count":80,"latest_published_at":81},"Startups","startups",84,"2026-09-30T20:39:09.000Z",{"name":83,"slug":84,"count":85,"latest_published_at":86},"Gaming","gaming",51,"2026-09-30T16:24:30.000Z",{"name":88,"slug":89,"count":90,"latest_published_at":91},"General","general",50,"2026-09-30T21:37:54.000Z",{"name":93,"slug":94,"count":95,"latest_published_at":96},"Reviews","reviews",31,"2026-09-28T14:31:34.000Z",{"name":98,"slug":99,"count":100,"latest_published_at":101},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]