
{"id":8970,"date":"2026-06-10T00:42:03","date_gmt":"2026-06-09T16:42:03","guid":{"rendered":"https:\/\/infernews.com\/blog\/code-and-data-for-paper-quot-spatialworld-benchmarking-interactive-spatial-reaso\/"},"modified":"2026-06-10T00:42:03","modified_gmt":"2026-06-09T16:42:03","slug":"code-and-data-for-paper-quot-spatialworld-benchmarking-interactive-spatial-reaso","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/code-and-data-for-paper-quot-spatialworld-benchmarking-interactive-spatial-reaso\/","title":{"rendered":"SpatialWorld\uff1a\u6e2c\u8a66\u591a\u6a21\u614b\u4ee3\u7406\u7a7a\u9593\u7406\u89e3\u7684\u7d71\u4e00\u57fa\u6e96"},"content":{"rendered":"<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/06\/pasted-2809be6d5598.jpg\" alt=\"SpatialWorld framework overview\"><\/figure>\n<p>SpatialWorld \u662f\u4e00\u500b\u7528\u4f86\u6e2c\u8a66 Multimodal Large Language Models\uff08MLLMs\uff09\u8207\u4ee3\u7406\u80fd\u529b\u7684 benchmark\u3002\u5b83\u628a 8 \u500b\u4e0d\u540c\u7684 3D \u6a21\u64ec\u5f8c\u7aef\u6574\u5408\u6210\u540c\u4e00\u5957 observation\u2013action \u4ecb\u9762\uff0c\u8b93\u6a21\u578b\u53ea\u9760\u81ea\u7136\u8a9e\u8a00\u6307\u4ee4\u3001\u7b2c\u4e00\u8eab egocentric RGB \u756b\u9762\uff0c\u4ee5\u53ca\u7d71\u4e00\u7684\u6587\u5b57\u52d5\u4f5c\u6307\u4ee4\u5b8c\u6210\u4efb\u52d9\u3002<\/p>\n<p>\u5982\u679c\u4f60\u60f3\u77e5\u9053\u4e00\u500b\u6a21\u578b\u662f\u5426\u771f\u7684\u61c2\u5f97\u5728\u74b0\u5883\u4e2d\u63a2\u7d22\u3001\u8f49\u5411\u3001\u79fb\u52d5\u3001\u5224\u65b7\u4f4d\u7f6e\u8207\u5b8c\u6210\u591a\u6b65\u9a5f\u4efb\u52d9\uff0cSpatialWorld \u63d0\u4f9b\u4e86\u8f03\u4e00\u81f4\u7684\u6e2c\u8a66\u65b9\u6cd5\u3002\u5b83\u5305\u542b 760 \u500b\u4eba\u5de5\u6a19\u8a3b\u4efb\u52d9\uff0c\u8986\u84cb\u5bb6\u5c45\u3001\u51fa\u884c\u3001\u5354\u4f5c\u8207\u6578\u78bc 3D \u904a\u6232\u7b49\u5834\u666f\uff0c\u4e26\u4ee5 human-validated terminal-state verifiers \u5224\u5b9a\u7d50\u679c\u3002<\/p>\n<p>\u8b93\u4ee3\u7406\u8f38\u5165\u6587\u5b57\u52d5\u4f5c\uff0c\u4f8b\u5982 Move\u3001Rotate\uff0c\u518d\u7531 action parser \u8f49\u6210\u5404\u500b\u6a21\u64ec\u5668\u539f\u751f\u6307\u4ee4\u3002\u9019\u7a2e\u505a\u6cd5\u7684\u91cd\u9ede\uff0c\u662f\u907f\u514d\u6bcf\u500b simulator \u5404\u6709\u4e00\u5957\u6d41\u7a0b\uff0c\u4ee4\u4e0d\u540c\u6a21\u578b\u4e4b\u9593\u8f03\u5bb9\u6613\u4f5c\u6a6b\u5411\u6bd4\u8f03\u3002<\/p>\n<ul>\n<li>\u7d71\u4e00 8 \u500b 3D backends\uff0c\u6e1b\u5c11 simulator-specific pipelines \u5e36\u4f86\u7684\u6bd4\u8f03\u56f0\u96e3<\/li>\n<li>\u53ea\u63d0\u4f9b vision-only partial observability\uff0c\u66f4\u63a5\u8fd1\u4ee3\u7406\u9010\u6b65\u63a2\u7d22\u7684\u60c5\u6cc1<\/li>\n<li>\u9664\u4e86 task success rate\uff08TSR\uff09\uff0c\u4ea6\u6703\u770b step efficiency\uff08SE\uff09\uff0c\u4e0d\u53ea\u6bd4\u8f03\u6709\u6c92\u6709\u5b8c\u6210<\/li>\n<li>\u5df2\u8a55\u4f30 15 \u500b\u4ee3\u7406\uff0c\u65b9\u4fbf\u5c0d\u7167\u73fe\u6709\u6a21\u578b\u8868\u73fe<\/li>\n<\/ul>\n<p>\u5f9e\u516c\u958b\u7d50\u679c\u770b\uff0c\u9019\u500b\u9805\u76ee\u63ed\u793a\u4e86\u76ee\u524d\u6a21\u578b\u7684\u9650\u5236\u3002GPT-5 \u7684\u5e73\u5747 TSR \u70ba 17.4%\uff0c\u9818\u5148\u7684 open-source \u6a21\u578b Qwen-3.5 \u70ba 14.1%\uff1b\u82e5\u770b Physical Overall TSR\uff0cGPT-5 \u53ea\u6709 14.4%\uff0cQwen-3.5-397B-A17B \u70ba 12.2%\u3002\u9019\u8868\u793a\u6a21\u578b\u5373\u4f7f\u80fd\u7406\u89e3\u5716\u7247\u8207\u6587\u5b57\uff0c\u4e5f\u672a\u5fc5\u80fd\u7a69\u5b9a\u5b8c\u6210\u9700\u8981\u7a7a\u9593\u63a8\u7406\u8207\u9577\u6b65\u9a5f\u898f\u5283\u7684\u4efb\u52d9\u3002<\/p>\n<p>\u76f8\u95dc\u6a21\u578b\u6709 GPT-5\u3001Qwen-3.5\u3001Qwen-3.5-397B-A17B \u8207 Gemini-3.1-Pro\uff0c\u5176\u4e2d Gemini-3.1-Pro \u5728 digital 3D games \u9054\u5230 39.0% TSR\u3002\u82e5\u4f60\u662f\u505a agent\u3001embodied AI\u3001MLLM \u8a55\u6e2c\uff0c\u6216\u8005\u60f3\u6bd4\u8f03\u4e0d\u540c\u6a21\u578b\u5728\u4e92\u52d5\u5f0f\u7a7a\u9593\u4efb\u52d9\u7684\u5dee\u7570\uff0cSpatialWorld \u6703\u662f\u4e00\u500b\u5f88\u6709\u53c3\u8003\u50f9\u503c\u7684\u9805\u76ee\u3002<\/p>\n<p><strong>GitHub\uff1a<\/strong> <a href=\"https:\/\/github.com\/Hongcheng-Gao\/SpatialWorld\" rel=\"noopener noreferrer\">https:\/\/github.com\/Hongcheng-Gao\/SpatialWorld<\/a><\/p>\n<p><strong>\u9805\u76ee\uff1a<\/strong> <a href=\"https:\/\/spatial-world.github.io\/\" rel=\"noopener noreferrer\">https:\/\/spatial-world.github.io\/<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>SpatialWorld \u7528\u7d71\u4e00\u4ecb\u9762\u8a55\u4f30\u591a\u6a21\u614b\u4ee3\u7406\u5728 3D \u4efb\u52d9\u4e2d\u7684\u7a7a\u9593\u7406\u89e3\u3002\u6578\u64da\u4ea6\u986f\u793a\uff0c\u73fe\u6709\u6a21\u578b\u8ddd\u96e2\u53ef\u9760\u5b8c\u6210\u4efb\u52d9\u4ecd\u6709\u660e\u986f\u5dee\u8ddd\u3002<\/p>\n","protected":false},"author":8,"featured_media":8969,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","footnotes":""},"categories":[169,119,195,196,197],"tags":[],"class_list":["post-8970","post","type-post","status-publish","format-standard","hentry","category-169","category-119","category-195","category-196","category-framework"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/8970","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=8970"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/8970\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/8969"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=8970"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=8970"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=8970"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}