[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-sa-vla-closes-the-gap-between-robot-instructions-and-actions":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},2764,"sa-vla-closes-the-gap-between-robot-instructions-and-actions","SA-VLA Closes the Gap Between Robot Instructions and Actions","A state-aware tokenizer doubles manipulation success rates by teaching robots to decode commands based on their current joint positions and contact conditions.","A research team has built a tokenizer that helps robot AI models translate language instructions into precise physical movements by factoring in what the robot's body is actually doing at that moment.\n\nCurrent vision-language-action models convert robot commands into discrete tokens — compact numerical codes — then decode those codes back into continuous motor signals. The problem: existing systems map each code to a fixed action regardless of the robot's posture, grip, or contact state. SA-VLA fixes this by conditioning the decoding step on live proprioceptive data, meaning the same token can resolve to different motor outputs depending on joint angles and object positions. The system adds either a cross-attention layer or a lightweight adapter module to a standard vector-quantization pipeline, keeping the architecture mostly intact.\n\nThe gains are hard to dismiss. On 12 RoboTwin manipulation tasks, average success climbed from 0.29 to 0.56 compared to the strongest existing tokenizer. In zero-shot sim-to-real transfer across three physical tasks, success went from 0.15 to 0.33 — doubling real-world performance without any task-specific fine-tuning. That sim-to-real number matters because bridging virtual training and physical hardware is where most robot learning research quietly falls apart.\n\nThe finding reframes a long-running assumption: that richer codebooks or bigger models are the lever for better robot action fidelity. Sometimes the fix is just giving the decoder the information it was missing all along.","[\"robotics\",\"ai\",\"machine-learning\",\"research\"]","2026-06-30T04:00:00.000Z","2026-06-30T12:18:12.302Z","2026-06-30T12:18:15.262Z","published",null,[],"ai",[26,24,27,28],"robotics","machine-learning","research",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.30113",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"]