[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-q2rl-trains-robot-arms-to-near-perfection-in-two-hours":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},2943,"q2rl-trains-robot-arms-to-near-perfection-in-two-hours","Q2RL Trains Robot Arms to Near-Perfection in Two Hours","A new algorithm extracts Q-values from behavior cloning policies to let robots keep improving on the job, without forgetting what they already learned.","Robots that learn from human demonstrations can now keep getting better after those demonstrations end — without unlearning everything they already know.\n\nResearchers have released Q2RL, an algorithm that bridges the gap between imitation learning and reinforcement learning. The approach works in two steps. First, it pulls a Q-function — a way to estimate how good any given action is — directly out of a behavior cloning policy, using only a handful of real-world interactions to calibrate it. Then, during online training, a gating mechanism decides action-by-action whether to follow the imitation policy or the RL policy, based on which one scores higher. The result is that the robot keeps the good habits it learned from demonstrations while still hunting for improvements.\n\nThe distribution mismatch problem — where online RL training causes a robot to overwrite hard-won skills with new, untested ones — has been a persistent headache for the field. Q2RL sidesteps it without requiring a separate dataset of corrective demonstrations or elaborate regularization schemes. In tests on manipulation benchmarks D4RL and robomimic, it beat existing offline-to-online baselines on both success rate and time to convergence.\n\nThe practical numbers are notable: the system learned contact-rich tasks like pipe assembly and kitting in one to two hours of on-robot interaction, hitting success rates as high as 100% and improving up to 3.75 times over the base behavior cloning policy. That kind of wall-clock efficiency matters because most competing approaches are either too slow or too fragile to run directly on hardware. The caveat is that \"up to 100%\" covers a range — not every task hit the ceiling — and the benchmarks used are well-worn enough that skepticism about real-world transfer is warranted.","[\"robotics\",\"reinforcement-learning\",\"ai\",\"research\"]","2026-06-30T04:00:00.000Z","2026-06-30T15:29:57.829Z","2026-06-30T15:30:00.720Z","published",null,[],"ai",[26,27,24,28],"robotics","reinforcement-learning","research",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.05172",0,{"sections":35},[36,40,45,50,55,60,65,70,75,80,85,89,94,99],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",2590,"2026-07-16T04:00:00.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":44},"Security","security",294,"2026-07-15T19:59:48.000Z",{"name":46,"slug":47,"count":48,"latest_published_at":49},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":51,"slug":52,"count":53,"latest_published_at":54},"Policy","policy",158,"2026-07-16T00:02:48.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":59},"Hardware","hardware",122,"2026-07-14T19:46:26.000Z",{"name":61,"slug":62,"count":63,"latest_published_at":64},"Consumer Tech","consumer-tech",93,"2026-07-13T13:20:48.000Z",{"name":66,"slug":67,"count":68,"latest_published_at":69},"Software","software",70,"2026-07-13T19:52:25.000Z",{"name":71,"slug":72,"count":73,"latest_published_at":74},"Science","science",66,"2026-07-10T10:29:37.000Z",{"name":76,"slug":77,"count":78,"latest_published_at":79},"Dev Tools","dev-tools",59,"2026-07-07T04:00:00.000Z",{"name":81,"slug":82,"count":83,"latest_published_at":84},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":86,"slug":87,"count":83,"latest_published_at":88},"Startups","startups","2026-06-29T20:55:50.000Z",{"name":90,"slug":91,"count":92,"latest_published_at":93},"General","general",29,"2026-07-10T22:28:58.000Z",{"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"]