{"id":12967,"date":"2026-10-06T08:25:45","date_gmt":"2026-10-06T00:25:45","guid":{"rendered":"https:\/\/infernews.com\/blog\/proar\/"},"modified":"2026-10-06T08:25:45","modified_gmt":"2026-10-06T00:25:45","slug":"proar","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/proar\/","title":{"rendered":"ProAR \u8b93\u5f71\u7247\u6a21\u578b\u5148\u60f3\u76ee\u6a19\u518d\u751f\u6210"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/10\/proar-teaser.jpg\" alt=\"ProAR overview\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u7576\u5f71\u7247\u6a21\u578b\u53ea\u6309\u904e\u53bb\u756b\u9762\u9010\u6b65\u9810\u6e2c\uff0c\u756b\u9762\u5373\u4f7f\u5c40\u90e8\u5408\u7406\uff0c\u4e5f\u672a\u5fc5\u80fd\u5b8c\u6210\u6307\u5b9a\u4efb\u52d9\u3002ProAR \u5c6c\u65bc\u81ea\u56de\u6b78\u5f71\u7247\u6a21\u578b\u9805\u76ee\uff0c\u900f\u904e prospective reasoning \u9810\u6e2c\u672a\u4f86\u8996\u89ba\u72c0\u614b\uff0c\u8655\u7406\u9577\u7a0b\u76ee\u6a19\u8207\u4e2d\u9593\u8f49\u63db\u5bb9\u6613\u5931\u63a7\u7684\u554f\u984c\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">ProAR \u4ee5 Wan2.2-TI2V-5B \u4f5c\u70ba\u5171\u4eab backbone\uff0c\u52a0\u5165\u5169\u689d\u4e92\u88dc\u6a5f\u5236\uff1aoutcome guidance \u6703\u4e00\u540c\u9810\u6e2c\u672a\u4f86 goal frame\uff0c\u8b93\u76ee\u524d chunk \u53c3\u8003\u76ee\u6a19\u4fe1\u5ff5\uff1btransition guidance \u5247\u5728\u8a13\u7df4\u671f\u9593\u5229\u7528 predictor\uff0c\u4ee4\u76ee\u524d hidden states \u5c0d\u9f4a\u4e7e\u6de8\u7684 next-chunk representation\u3002\u524d\u8005\u504f\u5411\u9577\u8ddd\u96e2\u898f\u5283\uff0c\u5f8c\u8005\u88dc\u8db3\u77ed\u8ddd\u96e2\u7684\u72c0\u614b\u929c\u63a5\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u2022 Outcome guidance \u5e6b\u52a9\u751f\u6210\u7d50\u679c\u7b26\u5408\u4efb\u52d9\u76ee\u6a19<br \/>\n\u2022 Transition guidance \u9f13\u52f5\u4e2d\u9593\u6b65\u9a5f\u9075\u5b88\u6642\u9593\u53ca\u7269\u7406\u908f\u8f2f<br \/>\n\u2022 \u63a8\u7406\u6642\u4e0d\u9700\u8981\u771f\u5be6\u672a\u4f86\u756b\u9762\uff0c\u8a13\u7df4\u7528 predictor \u6703\u79fb\u9664<br \/>\n\u2022 \u9805\u76ee\u63d0\u4f9b\u6a21\u578b checkpoint\u3001ProAR-test \u6578\u64da\u96c6\u53ca VBVR-10Tasks \u5be6\u9a57\u914d\u7f6e<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u8981\u6e2c\u8a66\u9805\u76ee\uff0c\u9700\u8981 Python 3.10\u3001PyTorch 2.8.0\uff0c\u4e26\u628a Wan2.2-TI2V-5B \u6b0a\u91cd\u3001VAE\u3001T5 \u7de8\u78bc\u5668\u53ca tokenizer \u653e\u5230\u6307\u5b9a\u672c\u5730\u8def\u5f91\uff1bflash-attn \u5c6c\u65bc\u53ef\u9078\u7684\u52a0\u901f\u5143\u4ef6\u3002\u9019\u5957\u6d41\u7a0b\u8f03\u9069\u5408\u7814\u7a76\u8996\u89ba\u63a8\u7406\u3001\u7269\u7406\u63a8\u6f14\u53ca embodied intelligence \u7684\u5718\u968a\uff0c\u672a\u5fc5\u662f\u5373\u88dd\u5373\u7528\u7684\u5f71\u7247\u5275\u4f5c\u5de5\u5177\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u63d0\u4f9b\u7684\u8cc7\u6599\u78ba\u8a8d\u4e86 AR \u8207 ProAR checkpoint\uff0c\u4ee5\u53ca VBVR-10Tasks \u548c\u6e2c\u8a66\u6578\u64da\u96c6\uff0c\u4f46\u672a\u5217\u51fa\u8db3\u5920\u7684\u6578\u503c\u7d50\u679c\uff0c\u96e3\u4ee5\u55ae\u6191\u5132\u5b58\u5eab\u5167\u5bb9\u5224\u65b7\u63d0\u5347\u5e45\u5ea6\u3002\u63a8\u7406\u6642\u9593\u6703\u589e\u52a0\uff0c\u63db\u4f86\u6a21\u578b\u5728\u751f\u6210\u524d\u9810\u6e2c\u76ee\u6a19\u53ca\u672a\u4f86\u8f49\u63db\uff1b\u53d6\u6368\u662f\u5426\u5408\u7406\uff0c\u8981\u8996\u4e4e\u4efb\u52d9\u662f\u5426\u91cd\u8996\u6b65\u9a5f\u6709\u6548\u6027\u591a\u65bc\u751f\u6210\u901f\u5ea6\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/luka-group.github.io\/ProAR\/\" rel=\"noopener noreferrer\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/github.com\/luka-group\/ProAR\" rel=\"noopener noreferrer\"><strong>GitHub<\/strong><\/a> \u00b7 <a href=\"https:\/\/huggingface.co\/LinghuiShen\/ProAR\" rel=\"noopener noreferrer\"><strong>\u6a21\u578b<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>ProAR \u628a\u81ea\u56de\u6b78\u5f71\u7247\u751f\u6210\u8b8a\u6210\u8996\u89ba\u63a8\u7406\u6d41\u7a0b\uff0c\u517c\u9867\u6700\u7d42\u76ee\u6a19\u8207\u4e2d\u9593\u6b65\u9a5f\u7684\u5408\u7406\u6027\u3002<\/p>\n","protected":false},"author":8,"featured_media":12966,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"wpai_generated_summary":"","footnotes":""},"categories":[133,183,162,76,127,173,31,163,38,152,150,199,213],"tags":[],"class_list":["post-12967","post","type-post","status-publish","format-standard","hentry","category-133","category-183","category-ai-productions","category-76","category-127","category-173","category-video","category-163","category-38","category-python","category-150","category-dataset-","category-tokenize"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/12967","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=12967"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/12967\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/12966"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=12967"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=12967"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=12967"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}