{"id":7008,"date":"2025-07-01T06:05:00","date_gmt":"2025-06-30T21:05:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=7008"},"modified":"2025-06-28T22:10:03","modified_gmt":"2025-06-28T13:10:03","slug":"distilling-on-device-language-models-for-robot-planning-with-minimal-human-intervention","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=7008","title":{"rendered":"Distilling On-device Language Models for Robot Planning with Minimal Human Intervention\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Distilling On-device Language Models for Robot Planning with Minimal Human Intervention\u00a0<\/strong>[117.9]<br>PRISM\u306f\u3001SLM(Small Language Model)\u5bfe\u5fdc\u30ed\u30dc\u30c3\u30c8\u30d7\u30e9\u30f3\u30ca\u30fc\u3092\u84b8\u7559\u3059\u308b\u305f\u3081\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3067\u3042\u308b\u3002 PRISM\u30923\u3064\u306eLCM\u5bfe\u5fdc\u30d7\u30e9\u30f3\u30ca\u30fc\u306b\u9069\u7528\u3057\u3001\u30de\u30c3\u30d4\u30f3\u30b0\u3001\u63a2\u7d22\u3001\u64cd\u4f5c\u3001\u5bb6\u4e8b\u652f\u63f4\u3092\u884c\u3046\u3002 GPT-4o \u306e 10-20% \u304b\u3089 93% \u4ee5\u4e0a\u307e\u3067, PRISM \u306f Llama-3.2-3B \u306e\u6027\u80fd\u3092\u5411\u4e0a\u3059\u308b\u3053\u3068\u3092\u793a\u3057\u305f\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2506.17486v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2506.17486v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Fri, 20 Jun 2025 21:44:27 GMT)<\/li>\n\n\n\n<li>robot planning\u3092\u5bfe\u8c61\u3068\u3057\u305f\u300cGiven a source LLM-enabled planner, PRISM synthesizes tasks and environments, elicits plans from the LLM-enabled planner in these synthesized environments, and then uses the resulting data to train an SLM-enabled planner that serves as a drop-in replacement for the source model.\u300d\u3068\u3044\u3046\u84b8\u7559\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306e\u63d0\u6848\u3002\u76f4\u89b3\u7684\u306b\u3082\u6709\u52b9\u305d\u3046\u3060\u304c\u5b9f\u969b\u6709\u671b\u306a\u7d50\u679c\u3002<\/li>\n\n\n\n<li>\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u30b5\u30a4\u30c8\u306f<a href=\"https:\/\/zacravichandran.github.io\/PRISM\/\">PRISM<\/a><\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[342,390,635,636],"class_list":["post-7008","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-robot","tag-synthetic-data","tag-635","tag-636"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7008","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=7008"}],"version-history":[{"count":1,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7008\/revisions"}],"predecessor-version":[{"id":7009,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7008\/revisions\/7009"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=7008"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=7008"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=7008"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}