{"id":9794,"date":"2026-06-30T04:05:47","date_gmt":"2026-06-29T20:05:47","guid":{"rendered":"https:\/\/infernews.com\/blog\/qwen-robotmanip\/"},"modified":"2026-06-30T04:05:47","modified_gmt":"2026-06-29T20:05:47","slug":"qwen-robotmanip","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/qwen-robotmanip\/","title":{"rendered":"Qwen-RobotManip \u5982\u4f55\u628a\u6a5f\u68b0\u81c2\u8a13\u7df4\u63a8\u5411\u901a\u7528\u5316"},"content":{"rendered":"\n
\"Og<\/figure>\n\n\n\n

\u9019\u662f\u4e00\u500b\u6a5f\u68b0\u4eba\u64cd\u4f5c\u6a21\u578b\uff0c\u540d\u70ba Qwen-RobotManip<\/strong>\uff0c\u5c6c\u65bc\u5efa\u57fa\u65bc Qwen-VL \u7684 Vision-Language-Action foundation model\u3002\u5b83\u4e3b\u8981\u8655\u7406\u6a5f\u68b0\u81c2\u64cd\u4f5c\u8cc7\u6599\u5206\u6563\u3001\u6602\u8cb4\u800c\u4e14\u96e3\u4ee5\u7d71\u4e00\u8a13\u7df4\u7684\u554f\u984c\uff0c\u76ee\u6a19\u662f\u8b93\u6a21\u578b\u5728\u672a\u898b\u904e\u7684\u4efb\u52d9\u3001\u5834\u666f\u8207\u6a5f\u68b0\u5e73\u53f0\u4e0a\u4ecd\u80fd\u4fdd\u6301\u53ef\u7528\u8868\u73fe\u3002<\/p>\n\n\n\n

\u5b83\u7684\u6838\u5fc3\u505a\u6cd5\uff0c\u662f\u628a\u64cd\u4f5c\u5b78\u7fd2\u4e2d\u7684\u8868\u5fb5\u3001\u52d5\u4f5c\u8207\u884c\u70ba\u4e09\u500b\u5c64\u9762\u653e\u9032\u540c\u4e00\u5957 alignment framework\u3002\u7814\u7a76\u5718\u968a\u540c\u6642\u5efa\u7acb human-to-robot synthesis pipeline\uff0c\u5c07\u7b2c\u4e00\u8eab\u624b\u90e8\u793a\u7bc4\u5f71\u7247\u8f49\u6210 15 \u500b\u5e73\u53f0\u53ef\u7528\u7684 robot trajectories\uff0c\u518d\u914d\u5408\u591a\u4f86\u6e90\u8cc7\u6599\u6574\u7406\u6d41\u7a0b\uff0c\u6574\u5408\u771f\u5be6\u6a5f\u68b0\u4eba\u3001\u5408\u6210\u8cc7\u6599\u8207\u4eba\u985e\u793a\u7bc4\u5f71\u7247\uff0c\u5f62\u6210\u7d04 38,100 \u5c0f\u6642 pretraining corpus\u3002<\/p>\n\n\n\n

\u548c\u5e38\u898b\u53ea\u96c6\u4e2d\u55ae\u4e00\u6a5f\u68b0\u5e73\u53f0\u3001\u55ae\u4e00\u8cc7\u6599\u4f86\u6e90\uff0c\u6216\u504f\u91cd\u5206\u4f48\u5167\u8868\u73fe\u7684\u505a\u6cd5\u76f8\u6bd4\uff0cQwen-RobotManip \u66f4\u8457\u91cd genuine generalization\u3002\u8a55\u4f30\u4e0a\u4ea6\u6c92\u6709\u505c\u7559\u5728\u4e00\u822c benchmark\uff0c\u800c\u662f\u52a0\u5165\u591a\u500b OOD \u8a2d\u5b9a\uff0c\u5305\u62ec RoboCasa365\u3001LIBERO-Plus\u3001EBench\u3001RoboTwin-Clean2Rand\u3001RoboTwin-IF \u8207 RoboTwin-XE\uff0c\u7528\u4f86\u6aa2\u67e5\u6307\u4ee4\u8ddf\u96a8\u3001\u64fe\u52d5\u7a69\u5065\u6027\u3001\u932f\u8aa4\u6062\u5fa9\uff0c\u4ee5\u53ca cross-embodiment knowledge transfer\u3002<\/p>\n\n\n\n

\u91cd\u9ede\u53ef\u6574\u7406\u70ba\uff1a
\n– \u5efa\u57fa\u65bc Qwen-VL<\/strong>\uff0c\u9762\u5411 robotic manipulation \u7684\u901a\u7528\u57fa\u790e\u6a21\u578b
\n– \u4ee5 unified alignment framework \u6574\u5408 heterogeneous manipulation data
\n– \u4f7f\u7528 human-to-robot synthesis pipeline\uff0c\u8986\u84cb 15 \u500b\u6a5f\u68b0\u5e73\u53f0
\n– \u53ea\u4f9d\u9760 open-source robotic manipulation datasets \u8207 human demonstration videos\uff0c\u672a\u63d0\u53ca\u79c1\u6709\u8cc7\u6599\u6536\u96c6
\n– \u5728\u591a\u500b OOD \u8a55\u6e2c\u4e2d\u512a\u65bc\u904e\u5f80 state-of-the-art models\uff0c\u5305\u62ec \u03c00.5\uff0c\u4e26\u5728 RoboChallenge \u6392\u540d\u7b2c\u4e00<\/p>\n\n\n\n

\u9019\u500b\u9805\u76ee\u8f03\u9069\u5408\u95dc\u6ce8 robotic manipulation\u3001VLA\u3001\u8de8\u6a5f\u68b0\u5e73\u53f0\u9077\u79fb\u8207\u6a5f\u68b0\u4eba\u8cc7\u6599\u64f4\u5c55\u6d41\u7a0b\u7684\u4eba\u95b1\u8b80\u3002\u73fe\u6709\u8cc7\u6599\u986f\u793a\uff0c\u5b83\u4e0d\u55ae\u662f\u518d\u52a0\u5927\u8a13\u7df4\u898f\u6a21\uff0c\u800c\u662f\u5148\u89e3\u6c7a\u8cc7\u6599\u5c0d\u9f4a\u554f\u984c\uff0c\u4ee4\u64f4\u5145\u898f\u6a21\u4e4b\u5f8c\u7684\u8a13\u7df4\u4fe1\u865f\u4e0d\u6703\u4e92\u76f8\u885d\u7a81\uff0c\u9019\u4e5f\u662f\u5b83\u80fd\u5728\u771f\u5be6\u6a5f\u68b0\u5e73\u53f0\u9a57\u8b49\u6cdb\u5316\u80fd\u529b\u7684\u95dc\u9375\u3002<\/p>\n\n\n\n

\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 Paper<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

\u9019\u662f\u4e00\u500b\u9762\u5411\u6a5f\u68b0\u4eba\u64cd\u4f5c\u7684 Vision-Language-Action \u57fa\u790e\u6a21\u578b\u3002\u5b83\u5617\u8a66\u7528\u7d71\u4e00\u5c0d\u9f4a\u8207\u5927\u898f\u6a21\u8cc7\u6599\u8a13\u7df4\uff0c\u63d0\u5347\u8de8\u5834\u666f\u8207\u8de8\u6a5f\u68b0\u5e73\u53f0\u6cdb\u5316\u80fd\u529b\u3002<\/p>\n","protected":false},"author":8,"featured_media":9793,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","footnotes":""},"categories":[133,185,119,76,127,149,184,197,204],"tags":[],"class_list":["post-9794","post","type-post","status-publish","format-standard","hentry","category-133","category-qwen","category-119","category-76","category-127","category-149","category-robotic","category-framework","category-visionlanguageaction"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9794","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/comments?post=9794"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9794\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/9793"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=9794"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=9794"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=9794"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}