[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-actioncodec-tokenizer-pushes-robot-ai-to-new-accuracy-highs":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},11114,"actioncodec-tokenizer-pushes-robot-ai-to-new-accuracy-highs","ActionCodec Tokenizer Pushes Robot AI to New Accuracy Highs","A tokenizer redesign grounded in information theory let a small vision-language-action model hit 97.4% success without robotics pretraining.","Researchers found a better way to translate robot actions into tokens a language model can read, and it closes much of the gap that huge robotics pretraining datasets were supposed to fix.\n\nA new paper describes ActionCodec, a tokenizer built for vision-language-action (VLA) models, the systems that let a language model control a robot by predicting actions the same way it predicts words. The authors argue that most prior tokenizers were judged only on how well they could reconstruct an action, not on how well they helped the model actually learn. ActionCodec instead follows four design rules pulled from information theory: keep more overlap between tokens over time, cut redundant vocabulary, boost the shared signal between action tokens and the language\u002Fvision tokens, and make tokens less dependent on each other. Fine-tuned into a 2.2-billion-parameter SmolVLM2 model with no robotics-specific pretraining, it hit a 95.5% success rate on the LIBERO benchmark, rising to 97.4% with some added architectural tweaks - a new best for VLA models that skip robotics pretraining entirely.\n\nThe field has mostly chased better results by throwing more robot demonstration data at pretraining, following the lead of large data-collection efforts. This result suggests how you encode an action into tokens can matter as much as how much robot data a model has seen beforehand, which is a cheaper lever to pull than building ever-bigger pretraining pipelines.\n\nTokenization is the kind of plumbing work that rarely gets headlines, but if these four design rules hold up outside LIBERO's simulated kitchens, they could save a lot of labs a lot of expensive data collection.","[\"robotics\",\"vla-models\",\"tokenization\",\"ai-research\"]","2026-10-09T04:00:00.000Z","2026-10-10T06:46:27.352Z","2026-10-10T06:46:32.465Z","published",null,[],"ai",[26,27,28,29],"robotics","vla-models","tokenization","ai-research",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2602.15397",0,{"sections":36},[37,41,45,50,55,59,63,68,73,78,82,87,92,97],{"name":38,"slug":24,"count":39,"latest_published_at":40},"AI",6834,"2026-10-09T11:52:17.000Z",{"name":42,"slug":43,"count":44,"latest_published_at":18},"Security","security",937,{"name":46,"slug":47,"count":48,"latest_published_at":49},"Policy","policy",487,"2026-10-09T11:39:54.000Z",{"name":51,"slug":52,"count":53,"latest_published_at":54},"Deals","deals",483,"2026-10-09T11:20:39.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":18},"Hardware","hardware",232,{"name":60,"slug":61,"count":62,"latest_published_at":18},"Science","science",194,{"name":64,"slug":65,"count":66,"latest_published_at":67},"Consumer Tech","consumer-tech",181,"2026-10-08T23:26:35.000Z",{"name":69,"slug":70,"count":71,"latest_published_at":72},"Startups","startups",117,"2026-10-08T16:45:00.000Z",{"name":74,"slug":75,"count":76,"latest_published_at":77},"Software","software",114,"2026-10-08T17:57:01.000Z",{"name":79,"slug":80,"count":81,"latest_published_at":18},"Dev Tools","dev-tools",106,{"name":83,"slug":84,"count":85,"latest_published_at":86},"General","general",66,"2026-10-09T04:46:11.000Z",{"name":88,"slug":89,"count":90,"latest_published_at":91},"Gaming","gaming",59,"2026-10-09T11:43:43.000Z",{"name":93,"slug":94,"count":95,"latest_published_at":96},"Reviews","reviews",34,"2026-10-08T14:00:22.000Z",{"name":98,"slug":99,"count":100,"latest_published_at":101},"How-To","how-to",8,"2026-10-05T09:00:00.000Z"]