[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-a-lightweight-memory-trick-could-make-robots-less-forgetful":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},7018,"a-lightweight-memory-trick-could-make-robots-less-forgetful","A Lightweight Memory Trick Could Make Robots Less Forgetful","Researchers built a compact memory token that lets robots skip expensive AI reasoning at runtime while still recalling what matters from past actions.","Researchers have found a way to give robots a working memory without forcing them to reason out loud at every step.\n\nA new paper describes \"workspace tokens,\" a compact memory format for robots that need to recall past actions to finish long manipulation tasks, since feeding a controller a robot's entire raw history of observations introduces noise and hurts performance. The existing fix, having a large vision-language model decide in real time which past details matter, works but is slow and expensive to run during live operation. The new approach moves that judgment call to training time: a VLM sorts out what's relevant once, offline, and distills it into a small workspace token the robot can consult instantly at runtime, no VLM needed in the loop. Tested in both simulation and on physical hardware, robots using workspace tokens matched or beat the costlier VLM-in-the-loop approach while skipping those live queries entirely.\n\nThat gap, between reasoning that's smart enough to work and reasoning that's cheap enough to run on a robot in real time, has been one of the practical limits on deploying capable models outside of demos. This result suggests the fix isn't necessarily a smaller or faster VLM, but skipping it at runtime altogether by front-loading the expensive thinking into training. It's the same distill-once, run-cheap trade AI researchers already use for compressing large models, applied here to a robot's memory of its own history.\n\nThe more striking claim, that the compressed memory outperformed the expensive version it was distilled from, is the kind of result that deserves a second lab's confirmation before it becomes conventional wisdom.","[\"robotics\",\"ai\",\"machine-learning\",\"research\"]","2026-09-18T04:00:00.000Z","2026-09-19T07:30:29.016Z","2026-09-19T07:30:40.915Z","published",null,[],"ai",[26,24,27,28],"robotics","machine-learning","research",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.20820",0,{"sections":35},[36,40,45,50,55,60,65,70,74,79,84,89,94,99],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",4114,"2026-09-19T18:33:46.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":44},"Security","security",673,"2026-09-19T17:30:00.000Z",{"name":46,"slug":47,"count":48,"latest_published_at":49},"Policy","policy",345,"2026-09-19T19:22:20.000Z",{"name":51,"slug":52,"count":53,"latest_published_at":54},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":59},"Hardware","hardware",156,"2026-09-19T11:00:00.000Z",{"name":61,"slug":62,"count":63,"latest_published_at":64},"Science","science",129,"2026-09-19T19:45:00.000Z",{"name":66,"slug":67,"count":68,"latest_published_at":69},"Consumer Tech","consumer-tech",99,"2026-09-09T17:27:33.000Z",{"name":71,"slug":72,"count":73,"latest_published_at":18},"Dev Tools","dev-tools",78,{"name":75,"slug":76,"count":77,"latest_published_at":78},"Software","software",75,"2026-09-10T20:41:21.000Z",{"name":80,"slug":81,"count":82,"latest_published_at":83},"Startups","startups",55,"2026-09-09T23:14:29.000Z",{"name":85,"slug":86,"count":87,"latest_published_at":88},"Gaming","gaming",43,"2026-09-10T12:18:06.000Z",{"name":90,"slug":91,"count":92,"latest_published_at":93},"General","general",42,"2026-09-18T22:35:10.000Z",{"name":95,"slug":96,"count":97,"latest_published_at":98},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":100,"slug":101,"count":102,"latest_published_at":103},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]