[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-coding-agents-reuse-optimized-robot-policies-for-new-tasks":10,"sections":34},{"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":29,"feedback":33,"feedback_at":22,"cost_usd":33,"total_tokens":33},6874,"coding-agents-reuse-optimized-robot-policies-for-new-tasks","Coding Agents Reuse Optimized Robot Policies for New Tasks","A study finds that letting coding agents refine robot control code before reusing it lifts success on new tasks by over 15 percentage points on average.","Robot software might get better simply by being reused, not just rewritten.\n\nResearchers tested whether coding agents that write and refine closed-loop robot control code for one task can hand that code off as a head start for a different task. In RoboCasa simulations, an agent generated policy code from a handful of demonstrations, then improved it through repeated feedback loops, pushing average success on four source tasks from 28.3% to 64.2%. Those refined programs were then archived and offered to the same agent as reference material for nine new target tasks. Across three separate runs, using the optimized source code as a starting point produced a 57.0% average success rate, versus 45.2% with no reference code at all and 41.5% with unoptimized starting code.\n\nThe gap between optimized and unoptimized references, a 15.6 percentage point average swing, suggests the payoff isn't just from having example code, but from having code that has already been debugged and tuned. That matters for anyone trying to scale robot programming beyond one-off demos: a growing library of proven task implementations could make each new robot skill cheaper to build than the last.\n\nIt's not a universal win: unoptimized starting code still beat the optimized version on two of the nine tasks, a reminder that reused code, like reused advice, doesn't always fit the new situation.","[\"robotics\",\"ai\",\"coding-agents\",\"research\"]","2026-09-18T04:00:00.000Z","2026-09-18T20:30:43.311Z","2026-09-18T20:30:55.200Z","published",null,[],"ai",[26,24,27,28],"robotics","coding-agents","research",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.19906",0,{"sections":35},[36,39,43,48,53,57,61,66,70,75,80,85,90,95],{"name":37,"slug":24,"count":38,"latest_published_at":18},"AI",4031,{"name":40,"slug":41,"count":42,"latest_published_at":18},"Security","security",654,{"name":44,"slug":45,"count":46,"latest_published_at":47},"Policy","policy",338,"2026-09-11T04:00:00.000Z",{"name":49,"slug":50,"count":51,"latest_published_at":52},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":54,"slug":55,"count":56,"latest_published_at":18},"Hardware","hardware",155,{"name":58,"slug":59,"count":60,"latest_published_at":18},"Science","science",121,{"name":62,"slug":63,"count":64,"latest_published_at":65},"Consumer Tech","consumer-tech",99,"2026-09-09T17:27:33.000Z",{"name":67,"slug":68,"count":69,"latest_published_at":18},"Dev Tools","dev-tools",78,{"name":71,"slug":72,"count":73,"latest_published_at":74},"Software","software",75,"2026-09-10T20:41:21.000Z",{"name":76,"slug":77,"count":78,"latest_published_at":79},"Startups","startups",55,"2026-09-09T23:14:29.000Z",{"name":81,"slug":82,"count":83,"latest_published_at":84},"Gaming","gaming",43,"2026-09-10T12:18:06.000Z",{"name":86,"slug":87,"count":88,"latest_published_at":89},"General","general",41,"2026-09-08T01:57:23.000Z",{"name":91,"slug":92,"count":93,"latest_published_at":94},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":96,"slug":97,"count":98,"latest_published_at":99},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]