{"id":13633,"date":"2026-10-20T21:30:41","date_gmt":"2026-10-20T13:30:41","guid":{"rendered":"https:\/\/infernews.com\/blog\/cope\/"},"modified":"2026-10-20T21:34:09","modified_gmt":"2026-10-20T13:34:09","slug":"cope","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/cope\/","title":{"rendered":"CoPE \u8b93 Agent \u9010\u6b65\u9810\u6e2c\u884c\u52d5\uff0c\u7bc0\u7701\u4e00\u534a\u8a13\u7df4\u6a23\u672c"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/10\/CoPE.jpg\" alt=\"CoPE\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u9577\u8ecc\u8de1\u53ea\u9760\u5b8c\u7d50\u6642\u4e00\u6b21\u734e\u52f5\u5b78\u7fd2\uff0c\u5f80\u5f80\u6703\u4ee4\u591a\u6578\u6b65\u9a5f\u6b20\u7f3a\u76f4\u63a5\u56de\u994b\u3002CoPE \u662f\u5efa\u57fa\u65bc AgentGym-RL \u53ca verl \u5206\u652f\u7684\u5f37\u5316\u5b78\u7fd2\u6846\u67b6\uff0c\u900f\u904e\u9010\u6b65\u9810\u6e2c\u667a\u80fd\u9ad4\u63a5\u4e0b\u4f86\u6703\u63a1\u53d6\u7684 K \u500b\u884c\u52d5\uff0c\u70ba\u6bcf\u500b\u6c7a\u7b56\u4f4d\u7f6e\u52a0\u5165\u66f4\u7d30\u7dfb\u7684\u8a13\u7df4\u8a0a\u865f\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u6a21\u578b\u9aa8\u5e79\u540c\u6642\u8ca0\u8cac acting \u8207 forecasting\uff1a\u667a\u80fd\u9ad4\u6309\u65e2\u6709 policy \u884c\u52d5\uff0c\u4e26\u5728\u6bcf\u4e00\u6b65\u9810\u6e2c\u672a\u4f86\u884c\u52d5\uff1b\u5b8c\u6210 rollout \u5f8c\uff0c\u771f\u5be6\u884c\u52d5\u5e8f\u5217\u6703\u76e3\u7763\u9810\u6e2c\u7d50\u679c\uff0c\u518d\u628a\u5b78\u7fd2\u6210\u679c\u84b8\u993e\u56de\u540c\u4e00\u500b policy\u3002\u95dc\u9589\u76f8\u95dc\u529f\u80fd\u5f8c\uff0c\u8a13\u7df4\u8def\u5f91\u4fbf\u7dad\u6301\u6a19\u6e96 Group Relative Policy Optimization\uff08GRPO\uff09\uff0c\u65b9\u4fbf\u6309\u7814\u7a76\u9700\u8981\u6bd4\u8f03\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6bcf\u4e00\u6b65\u9810\u6e2c K \u500b\u5f8c\u7e8c\u884c\u52d5\uff0c\u88dc\u5145\u7d42\u7d50\u734e\u52f5\u6b20\u7f3a\u7684\u5206\u6b65\u76e3\u7763\u3002<\/li>\n<li>acting \u8207 forecasting \u5171\u7528\u4e00\u500b\u6a21\u578b\uff0c\u4e0d\u9700\u984d\u5916\u6a21\u578b\u6216\u9810\u8a13\u7df4\u968e\u6bb5\u3002<\/li>\n<li>\u63a8\u7406\u6642\u4ecd\u9010\u6b65\u884c\u52d5\uff0c\u4e0d\u52a0\u5165\u984d\u5916\u904b\u7b97\u958b\u652f\u3002<\/li>\n<li>\u652f\u63f4 SciWorld\u3001ALFWorld\u3001WebShop\u3001AppWorld \u53ca \u03c4\u00b2-bench \u4e94\u500b\u4e92\u52d5\u74b0\u5883\u3002<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u7d50\u679c\u986f\u793a\uff0cCoPE \u5728\u4e0a\u8ff0\u74b0\u5883\u64ca\u6557 GRPO \u8207 predictive-auxiliary baselines\uff1b\u6a23\u672c\u6548\u7387\u6700\u9ad8\u63d0\u9ad8 2 \u500d\uff0c\u5176\u4e2d ScienceWorld \u70ba 2.0 \u500d\u3001AppWorld \u70ba 1.6 \u500d\u3002\u4efb\u52d9\u6210\u529f\u7387\u76f8\u5c0d GRPO \u5728 ALFWorld\u3001WebShop \u8207 ScienceWorld \u5206\u5225\u589e\u52a0 5.0\u300112.5 \u53ca 8.0 \u500b\u767e\u5206\u9ede\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u914d\u7f6e\u9700\u8981 Python 3.10 \u8a13\u7df4\u74b0\u5883\u53ca\u7368\u7acb\u74b0\u5883\u670d\u52d9\uff1bWebShop \u4f7f\u7528 Python 3.8\uff0c\u03c4\u00b2-bench \u9700\u8981 Python 3.12 \u6216\u4ee5\u4e0a\uff0c\u4e26\u9808\u53d6\u5f97 AgentGym submodule\u3002\u9019\u5957\u67b6\u69cb\u9069\u5408\u7814\u7a76\u4e92\u52d5\u5f0f\u8a9e\u8a00\u667a\u80fd\u9ad4\u3001\u8a08\u7b97\u6210\u672c\u53d7\u63a7\u7684\u5718\u968a\uff0c\u4ee5\u53ca\u6bd4\u8f03 action forecasting \u8207\u7a00\u758f\u734e\u52f5\u8a13\u7df4\u65b9\u6cd5\u7684\u958b\u767c\u8005\uff0c\u4f46\u591a\u74b0\u5883\u914d\u7f6e\u8207 GPU \u9700\u6c42\u589e\u52a0\u4e86\u5b89\u88dd\u9580\u6abb\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/ulab-uiuc.github.io\/CoPE\/\" rel=\"noopener noreferrer\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/github.com\/ulab-uiuc\/CoPE\" rel=\"noopener noreferrer\"><strong>GitHub<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>CoPE \u4ee5\u591a\u6b65\u884c\u52d5\u9810\u6e2c\u88dc\u5f37 Agentic RL\uff0c\u5728\u4e0d\u589e\u52a0\u63a8\u8ad6\u6210\u672c\u4e0b\uff0c\u63d0\u5347\u8a9e\u8a00\u667a\u80fd\u9ad4\u7684\u5b78\u7fd2\u6548\u7387\u3002<\/p>\n","protected":false},"author":8,"featured_media":13632,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"wpai_generated_summary":"","wpai_meta_description":"","footnotes":""},"categories":[133,116,76,127,197,152,188,214],"tags":[],"class_list":["post-13633","post","type-post","status-publish","format-standard","hentry","category-133","category-agentic","category-76","category-127","category-framework","category-python","category-meta","category-reinforcement-learning"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/13633","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=13633"}],"version-history":[{"count":1,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/13633\/revisions"}],"predecessor-version":[{"id":13636,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/13633\/revisions\/13636"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/13632"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=13633"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=13633"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=13633"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}