{"id":9780,"date":"2026-06-30T03:32:55","date_gmt":"2026-06-29T19:32:55","guid":{"rendered":"https:\/\/infernews.com\/blog\/vla\/"},"modified":"2026-06-30T03:32:55","modified_gmt":"2026-06-29T19:32:55","slug":"vla","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/vla\/","title":{"rendered":"\u5fae\u8edf\u7528\u6b98\u5dee\u5f37\u5316\u5b78\u7fd2\u88dc\u5f37\u6a5f\u68b0\u4eba VLA"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/06\/image-17-1024x434-1.png\" alt=\"Og image\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u9019\u662f Microsoft Research \u7684 <strong>Object-Centric Residual RL<\/strong> \u65b9\u6cd5\u3002\u5b83\u7684\u7528\u9014\u662f\u70ba <strong>Vision-Language-Action\uff08VLA\uff09<\/strong> \u6a21\u578b\u52a0\u5165\u4e00\u5c64\u4fee\u6b63\u7b56\u7565\uff0c\u6e1b\u5c11\u6a5f\u68b0\u4eba\u5728\u771f\u5be6\u64cd\u63a7\u4e2d\u56e0\u7d30\u5c0f\u8aa4\u5dee\u7d2f\u7a4d\u800c\u5931\u624b\u7684\u60c5\u6cc1\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u6838\u5fc3\u505a\u6cd5\u662f\u628a\u4e00\u500b\u5df2\u51cd\u7d50\u7684 base VLA \u4fdd\u7559\u4e0d\u8b8a\uff0c\u518d\u984d\u5916\u758a\u52a0\u4e00\u500b\u8f15\u91cf residual policy\u3002\u9019\u500b residual policy \u5b8c\u5168\u5728\u6a21\u64ec\u74b0\u5883\u8a13\u7df4\uff0c\u8f38\u5165\u4e0d\u662f\u539f\u59cb\u5f71\u50cf\uff0c\u800c\u662f object-centric state\u3001proprioception\uff0c\u4ee5\u53ca base VLA action\uff0c\u85c9\u6b64\u907f\u958b\u5e38\u898b\u7684 visual sim-to-real gap\uff0c\u4ee4\u7b56\u7565\u53ef\u76f4\u63a5 zero-shot \u8f49\u79fb\u5230\u771f\u5be6\u6a5f\u68b0\u4eba\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u9019\u500b\u65b9\u6cd5\u8655\u7406\u7684\u662f imitation learning \u5e38\u898b\u7684\u8106\u5f31\u6027\uff1a\u793a\u7bc4\u8cc7\u6599\u672a\u8986\u84cb\u5230\u7684\u72c0\u614b\u4e00\u65e6\u51fa\u73fe\uff0c\u8aa4\u5dee\u6703\u4e00\u8def\u653e\u5927\u3002\u8207\u76f4\u63a5\u5fae\u8abf VLA\u3001\u6216\u5728\u771f\u5be6\u4e16\u754c\u518d\u505a reinforcement learning \u76f8\u6bd4\uff0c\u9019\u500b\u65b9\u6cd5\u7684\u5dee\u7570\u5728\u65bc\u53ea\u5b78\u300c\u4fee\u6b63\u91cf\u300d\uff0c\u800c\u4e14\u8a13\u7df4\u653e\u5728\u6a21\u64ec\u4e2d\u5b8c\u6210\uff0c\u6210\u672c\u8207\u98a8\u96aa\u90fd\u8f03\u4f4e\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4fdd\u7559 frozen base VLA\uff0c\u4e0d\u7528\u91cd\u8a13\u6574\u500b\u6a21\u578b<\/li>\n<li>residual policy \u4ee5\u7269\u4ef6\u4e2d\u5fc3\u72c0\u614b\u70ba\u4e3b\uff0c\u6e1b\u5c11\u4f9d\u8cf4\u5f71\u50cf\u5c0d\u9f4a<\/li>\n<li>\u76ee\u6a19\u662f zero-shot sim-to-real enhancement<\/li>\n<li>\u9069\u5408\u9700\u8981\u7cbe\u6e96\u6293\u53d6\u3001\u653e\u7f6e\u6216\u63a5\u89f8\u64cd\u4f5c\u7684\u6a5f\u68b0\u4eba\u5de5\u4f5c\u6d41<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u9019\u985e\u65b9\u6cd5\u8f03\u9069\u5408\u95dc\u6ce8 <strong>VLA<\/strong>\u3001<strong>Robotic<\/strong> \u8207 sim-to-real transfer \u7684\u7814\u7a76\u8005\u548c\u958b\u767c\u8005\u3002\u73fe\u6709\u8cc7\u6599\u6e05\u695a\u4ea4\u4ee3\u65b9\u6cd5\u65b9\u5411\u8207\u554f\u984c\u8a2d\u5b9a\uff0c\u4f46\u672a\u63d0\u4f9b\u5b8c\u6574\u5b89\u88dd\u6d41\u7a0b\u6216\u64cd\u4f5c\u6b65\u9a5f\uff1b\u8a55\u4f30\u90e8\u5206\u53ef\u78ba\u8a8d\u4f5c\u8005\u4ee5\u771f\u5be6\u6a5f\u68b0\u4eba\u6210\u529f\u7387\u6539\u5584\u4f5c\u70ba\u91cd\u9ede\uff0c\u66f4\u591a\u6578\u503c\u7d30\u7bc0\u4ecd\u9700\u53c3\u8003\u539f\u59cb\u8ad6\u6587\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/articles\/object-centric-residual-rl\/\" rel=\"noopener noreferrer\" target=\"_blank\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/arxiv.org\/pdf\/2606.18953\" rel=\"noopener noreferrer\" target=\"_blank\"><strong>Paper<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u9019\u662f\u4e00\u500b\u63d0\u5347\u6a5f\u68b0\u4eba\u64cd\u63a7\u7a69\u5b9a\u6027\u7684\u7814\u7a76\u65b9\u6cd5\u3002\u5b83\u7528\u6a21\u64ec\u8a13\u7df4\u7684\u6b98\u5dee\u7b56\u7565\uff0c\u96f6\u6a23\u672c\u88dc\u5f37\u771f\u5be6\u4e16\u754c\u7684 VLA \u8868\u73fe\u3002<\/p>\n","protected":false},"author":8,"featured_media":9779,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","footnotes":""},"categories":[173,116,30,162,119,170,146,149,184,197,204],"tags":[],"class_list":["post-9780","post","type-post","status-publish","format-standard","hentry","category-173","category-agentic","category-image","category-ai-productions","category-119","category-170","category-146","category-149","category-robotic","category-framework","category-visionlanguageaction"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9780","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=9780"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9780\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/9779"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=9780"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=9780"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=9780"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}