{"id":10028,"date":"2026-07-08T19:32:00","date_gmt":"2026-07-08T11:32:00","guid":{"rendered":"https:\/\/infernews.com\/blog\/structure-aware-data-selection-for-imitation-learning-with-vla-models\/"},"modified":"2026-07-08T19:32:00","modified_gmt":"2026-07-08T11:32:00","slug":"structure-aware-data-selection-for-imitation-learning-with-vla-models","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/structure-aware-data-selection-for-imitation-learning-with-vla-models\/","title":{"rendered":"SIEVE \u9ede\u6a23\u63c0\u51fa\u66f4\u503c\u9322\u7684\u6a5f\u68b0\u4eba\u793a\u7bc4\u6578\u64da"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/07\/pasted-1be68d90e24d.jpg\" alt=\"Repository image for ChangtiWu\/SIEVE\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u6a5f\u68b0\u4eba\u793a\u7bc4\u8cc7\u6599\u6700\u5e38\u898b\u7684\u554f\u984c\uff0c\u4e0d\u662f\u6578\u91cf\u4e0d\u5920\uff0c\u800c\u662f\u91cd\u8907\u3001\u5608\u96dc\uff0c\u751a\u81f3\u9577\u6bb5\u4efb\u52d9\u5176\u5be6\u53ea\u662f\u4e0d\u65b7\u91cd\u6f14\u76f8\u4f3c\u52d5\u4f5c\u3002SIEVE \u5c6c\u65bc\u4e00\u500b\u9762\u5411 imitation learning \u7684\u8cc7\u6599\u7be9\u9078\u5de5\u5177\uff0c\u540c\u6642\u5e36\u6709\u7814\u7a76\u65b9\u6cd5\u6027\u8cea\uff0c\u91cd\u9ede\u4e0d\u662f\u9010\u689d trajectory \u7c97\u7565\u8a55\u5206\uff0c\u4e5f\u4e0d\u662f\u53ea\u770b state-action\uff0c\u800c\u662f\u628a\u9577\u4efb\u52d9\u62c6\u6210\u53ef\u91cd\u7528\u7684 visuo-motor primitives \u8207 transition interfaces\uff0c\u518d\u6c7a\u5b9a\u54ea\u4e9b episode \u66f4\u503c\u5f97\u7559\u4e0b\u4f86\u8a13\u7df4 VLA \u6a21\u578b\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5b83\u6279\u8a55\u7684\u820a\u7bc4\u5f0f\u76f8\u7576\u660e\u78ba\uff1a\u73fe\u6709 data selection \u65b9\u6cd5\u591a\u6578\u53ea\u5728 trajectory level \u6216 state-action level \u505a\u5224\u65b7\uff0c\u56e0\u800c\u5ffd\u7565\u9577\u6642\u5e8f\u884c\u70ba\u5167\u90e8\u53ef\u91cd\u7528\u7684\u7d50\u69cb\u3002SIEVE \u7684\u505a\u6cd5\u662f\u5148\u7528 end-effector pose \u8207\u63a7\u5236\u8a0a\u865f\u505a segmentation\uff0c\u518d\u62bd\u53d6 V-JEPA \u7279\u5fb5\u3001\u7528 PCA \u58d3\u5230\u9810\u8a2d 256 \u7dad\u3001\u4ee5 MiniBatchKMeans \u627e\u51fa primitive pattern\uff0c\u4e4b\u5f8c\u6309 cluster-sequence pattern \u505a\u5169\u968e\u6bb5 episode selection\uff0c\u6700\u5f8c\u53ef\u4ee5\u532f\u51fa\u56de LeRobot \u683c\u5f0f\uff0c\u65b9\u4fbf\u76f4\u63a5\u63a5\u56de\u539f\u672c\u8a13\u7df4\u6d41\u7a0b\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u9019\u7a2e\u53d6\u5411\u7684\u597d\u8655\uff0c\u5728\u65bc\u5b83\u4e0d\u662f\u55ae\u7d14\u6311\u300c\u6700\u597d\u300d\u6216\u300c\u6700\u4e7e\u6de8\u300d\u7684\u793a\u7bc4\uff0c\u800c\u662f\u512a\u5148\u4fdd\u7559\u7d50\u69cb\u8986\u84cb\u5ea6\u8207\u53ef\u91cd\u7528\u6027\u3002\u8ad6\u6587\u63d0\u4f9b\u7684\u8a0a\u606f\u4ea6\u76f8\u7576\u76f4\u63a5\uff1aSIEVE \u5728\u591a\u500b datasets\u3001benchmarks \u8207 VLA models \u4e0a\uff0c\u90fd\u6bd4\u7af6\u722d\u6027\u7684 baseline \u66f4\u7a69\u5b9a\uff0c\u751a\u81f3\u5728\u53ea\u7528 50% demonstrations \u548c 50% training steps \u7684\u60c5\u6cc1\u4e0b\uff0c\u8868\u73fe\u53ef\u4ee5\u8d85\u904e full-data training\u3002\u7576\u7136\uff0c\u9019\u4e5f\u4ee3\u8868\u5b83\u8f03\u9069\u5408\u5df2\u6709\u4e00\u5b9a\u898f\u6a21\u793a\u7bc4\u8cc7\u6599\u3001\u4e26\u4e14\u9858\u610f\u5148\u8dd1\u4e00\u8f2a\u96e2\u7dda\u6574\u7406\u6d41\u7a0b\u7684\u5718\u968a\uff0c\u800c\u4e0d\u662f\u8ffd\u6c42\u5373\u63d2\u5373\u7528\u7684\u5c0f\u578b\u8173\u672c\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4ee5 LeRobot v2 \u8cc7\u6599\u6839\u76ee\u9304\u4f5c\u8f38\u5165\uff0c\u652f\u63f4\u55ae\u4e00\u6216\u591a\u500b dataset<\/li>\n<li>\u6d41\u7a0b\u7531 segmentation\u3001feature extraction\u3001dimensionality reduction\u3001clustering\u3001selection \u7d44\u6210<\/li>\n<li>\u7279\u5fb5\u62bd\u53d6\u4f9d\u8cf4 V-JEPA\uff0c\u8f38\u51fa\u4e2d\u9593\u7d50\u679c\u5230 Zarr\uff0c\u518d\u532f\u51fa\u9078\u4e2d\u7684 LeRobot episodes<\/li>\n<li>\u6838\u5fc3\u5dee\u7570\u662f\u6309 reusable structure \u63c0\u6578\u64da\uff0c\u4e0d\u662f\u53ea\u6309\u6574\u689d trajectory \u6216\u9010\u6b65 state-action \u6253\u5206<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u90e8\u7f72\u7406\u89e3\u4e0a\uff0c\u9019\u500b\u9805\u76ee\u66f4\u50cf\u4e00\u689d\u53ef\u91cd\u8907\u57f7\u884c\u7684\u96e2\u7dda\u8cc7\u6599\u8655\u7406 pipeline\uff0c\u800c\u4e0d\u662f\u4e00\u500b\u76f4\u63a5\u63d0\u4f9b\u63a8\u7406\u670d\u52d9\u7684\u5957\u4ef6\u3002\u9069\u5408\u7528\u4f86\u6574\u7406\u5927\u578b robot demonstration corpus\u3001\u70ba VLA imitation learning \u6e1b\u5c11\u5197\u9918\u8a13\u7df4\u6a23\u672c\uff1b\u76f8\u95dc\u6a21\u578b\u8207\u6280\u8853\u8108\u7d61\u5305\u62ec Vision-Language-Action (VLA) models\u3001V-JEPA\uff0c\u4ee5\u53ca\u8f38\u51fa\u7aef\u76f8\u5bb9\u7684 LeRobot\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/github.com\/ChangtiWu\/SIEVE\" rel=\"noopener noreferrer\" target=\"_blank\"><strong>GitHub<\/strong><\/a> \u00b7 <a href=\"https:\/\/arxiv.org\/pdf\/2607.06442\" rel=\"noopener noreferrer\" target=\"_blank\"><strong>Paper<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u793a\u7bc4\u6578\u64da\u6108\u591a\uff0c\u672a\u5fc5\u4ee3\u8868 VLA \u8a13\u7df4\u6548\u679c\u6108\u597d\u3002SIEVE \u60f3\u505a\u7684\uff0c\u662f\u5148\u62c6\u958b\u52d5\u4f5c\u7d50\u69cb\uff0c\u518d\u6311\u51fa\u66f4\u503c\u5f97\u4fdd\u7559\u7684\u7247\u6bb5\u8207\u8ecc\u8de1\u3002<\/p>\n","protected":false},"author":8,"featured_media":10027,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","footnotes":""},"categories":[133,163,165,119,120,76,127,149,184,189,199,204],"tags":[],"class_list":["post-10028","post","type-post","status-publish","format-standard","hentry","category-133","category-163","category-165","category-119","category-120","category-76","category-127","category-149","category-robotic","category-189","category-dataset-","category-visionlanguageaction"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/10028","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=10028"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/10028\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/10027"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=10028"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=10028"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=10028"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}