{"id":3887,"date":"2025-09-15T10:39:11","date_gmt":"2025-09-15T02:39:11","guid":{"rendered":"https:\/\/www.15zhi.net\/blog\/?p=3887"},"modified":"2025-09-15T10:39:11","modified_gmt":"2025-09-15T02:39:11","slug":"202509-%e8%ae%ba%e6%96%87%e7%a0%94%e8%af%bb-endowing-embodied-agents-with-spatial-reasoning-capabilities-for-vision-and-language-navigation","status":"publish","type":"post","link":"https:\/\/www.15zhi.net\/blog\/202509-%e8%ae%ba%e6%96%87%e7%a0%94%e8%af%bb-endowing-embodied-agents-with-spatial-reasoning-capabilities-for-vision-and-language-navigation\/","title":{"rendered":"202509 \u8bba\u6587\u7814\u8bfb-Endowing Embodied Agents with Spatial Reasoning Capabilities for Vision-and-language Navigation"},"content":{"rendered":"\n

\u4f5c\u8005\uff1a<\/a>Luo Ling<\/a>,\u00a0Bai Qianqian<\/a>
\u5355\u4f4d\uff1a\u897f\u5357\u8d22\u7ecf\u5927\u5b66
\u6765\u6e90\uff1aarxiv Artificial Intelligence
\u65f6\u95f4\uff1a2025.04.09<\/p>\n\n\n\n

\u4e00\u3001\u7814\u7a76\u80cc\u666f<\/strong><\/h2>\n\n\n\n

\u7814\u7a76\u65b9\u5411\u4e3a\u89c6\u89c9-\u8bed\u8a00\u5bfc\u822a\uff08Vision-and-Language Navigation, VLN\uff09\uff0c\u5373\u667a\u80fd\u4f53\u901a\u8fc7\u89c6\u89c9\u8f93\u5165\u548c\u81ea\u7136\u8bed\u8a00\u6307\u4ee4\u5728\u590d\u6742\u73af\u5883\u4e2d\u5b8c\u6210\u5bfc\u822a\u3002\u867d\u7136\u5728\u6a21\u62df\u73af\u5883\u4e2d\u5df2\u6709\u663e\u8457\u8fdb\u5c55\uff0c\u4f46\u8fc1\u79fb\u5230\u73b0\u5b9e\u73af\u5883\u65f6\uff0c\u5e38\u51fa\u73b0\u7a7a\u95f4\u5e7b\u89c9\uff08spatial hallucination\uff09<\/strong>\u2014\u2014\u667a\u80fd\u4f53\u5bf9\u73af\u5883\u7684\u611f\u77e5\u4e0e\u771f\u5b9e\u4e0d\u7b26\uff0c\u5bfc\u81f4\u8def\u5f84\u9009\u62e9\u9519\u8bef\u3001\u78b0\u649e\u6216\u8ff7\u5931\u65b9\u5411\u3002\u73b0\u6709\u65b9\u6cd5\u4f9d\u8d56\u5355\u4e00\u63d0\u793a\u7cfb\u7edf\u6216\u5355\u4e00\u5730\u56fe\u5efa\u6a21\uff0c\u7f3a\u4e4f\u591a\u5c42\u6b21\u73af\u5883\u5efa\u6a21\u548c\u52a8\u6001\u66f4\u65b0\u80fd\u529b\uff0c\u96be\u4ee5\u5e94\u5bf9\u8fde\u7eed\u7a7a\u95f4\u548c\u52a8\u6001\u969c\u788d\u3002<\/p>\n\n\n\n

\u4e8c\u3001\u6838\u5fc3\u5185\u5bb9<\/strong><\/h2>\n\n\n\n

\u8bba\u6587\u63d0\u51fa BrainNav \u6846\u67b6<\/strong>\uff0c\u4e00\u79cd\u57fa\u4e8e\u751f\u7269\u8ba4\u77e5\u7406\u8bba\u7684\u7a7a\u95f4\u5bfc\u822a\u65b9\u6cd5\uff1a<\/p>\n\n\n\n

    \n
  • \u501f\u9274\u751f\u7269\u7684 \u8ba4\u77e5\u5730\u56fe\u7406\u8bba<\/strong> \u4e0e\u5927\u8111\u591a\u533a\u57df\u534f\u4f5c\u673a\u5236\u3002<\/li>\n\n\n\n
  • \u8bbe\u8ba1\u4e86 \u53cc\u5730\u56fe\uff08\u5750\u6807\u56fe + \u62d3\u6251\u56fe\uff09<\/strong> \u4e0e \u53cc\u65b9\u5411\uff08\u7edd\u5bf9\u65b9\u5411 + \u76f8\u5bf9\u65b9\u5411\uff09<\/strong> \u7b56\u7565\u3002<\/li>\n\n\n\n
  • \u5f15\u5165\u4e94\u5927\u7c7b\u201c\u4eff\u751f\u6a21\u5757\u201d\uff1a\u6d77\u9a6c\u4f53\u8bb0\u5fc6\u3001\u89c6\u89c9\u76ae\u5c42\u611f\u77e5\u3001\u9876\u53f6\u7a7a\u95f4\u5efa\u6784\u3001\u524d\u989d\u53f6\u51b3\u7b56\u3001\u4e2d\u5c0f\u8111\u8fd0\u52a8\u6267\u884c\u3002<\/li>\n\n\n\n
  • \u6846\u67b6\u4e0e\u5927\u8bed\u8a00\u6a21\u578b\uff08\u5982 GPT-4\uff09\u517c\u5bb9\uff0c\u652f\u6301\u96f6\u6837\u672c\uff08zero-shot\uff09\u5728\u73b0\u5b9e\u73af\u5883\u4e2d\u5bfc\u822a\u3002<\/li>\n<\/ul>\n\n\n\n

    \u4e09\u3001\u6838\u5fc3\u6846\u67b6\u8d21\u732e<\/strong><\/h2>\n\n\n\n
    \"\"<\/figure>\n\n\n\n

    BrainNav \u7684\u4e94\u5927\u6a21\u5757\u5bf9\u5e94\u5927\u8111\u529f\u80fd\uff1a<\/p>\n\n\n\n

      \n
    1. Hippocampal Memory Center\uff08\u6d77\u9a6c\u4f53\u8bb0\u5fc6\u4e2d\u5fc3\uff09<\/strong>\uff1a\u5386\u53f2\u8f68\u8ff9\u5b58\u50a8\u4e0e\u56de\u6eaf\uff0c\u652f\u6301\u8bb0\u5fc6\u53ec\u56de\u3002<\/li>\n\n\n\n
    2. Visual Cortex Perception Engine\uff08\u89c6\u89c9\u76ae\u5c42\u611f\u77e5\u5f15\u64ce\uff09<\/strong>\uff1a\u57fa\u4e8e LLM \u5206\u6790\u573a\u666f\u56fe\u50cf\uff0c\u8bc6\u522b\u969c\u788d\u7269\u3001\u53ef\u884c\u8d70\u8def\u5f84\u3001\u5730\u6807\u3002<\/li>\n\n\n\n
    3. Parietal Spatial Builder\uff08\u9876\u53f6\u7a7a\u95f4\u5efa\u6784\u5668\uff09<\/strong>\uff1a\u6784\u5efa\u53cc\u5730\u56fe\uff08\u7cbe\u786e\u4f4d\u7f6e\u7684\u5750\u6807\u56fe + \u7a7a\u95f4\u8fde\u901a\u6027\u7684\u62d3\u6251\u56fe\uff09\u3002<\/li>\n\n\n\n
    4. Prefrontal Decision Center\uff08\u524d\u989d\u53f6\u51b3\u7b56\u4e2d\u5fc3\uff09<\/strong>\uff1a\u9ad8\u5c42\u89c4\u5212\u4e0e\u51b3\u7b56\uff0c\u6574\u5408\u611f\u77e5\u3001\u5730\u56fe\u548c\u8bb0\u5fc6\u4fe1\u606f\uff0c\u8f93\u51fa\u4e0b\u4e00\u6b65\u52a8\u4f5c\u3002<\/li>\n\n\n\n
    5. Cerebellum Motor Execution Unit\uff08\u5c0f\u8111\u8fd0\u52a8\u6267\u884c\u5355\u5143\uff09<\/strong>\uff1a\u5c06\u9ad8\u5c42\u6307\u4ee4\u8f6c\u5316\u4e3a\u5177\u4f53\u673a\u5668\u4eba\u52a8\u4f5c\uff0c\u7ed3\u5408\u76f8\u5bf9\u4e0e\u7edd\u5bf9\u65b9\u5411\u51cf\u5c11\u5e7b\u89c9\u3002<\/li>\n<\/ol>\n\n\n\n

      \u56db\u3001\u5b9e\u9a8c\u90e8\u5206<\/strong><\/h2>\n\n\n\n

      \u5b9e\u9a8c\u8bbe\u7f6e<\/strong>\uff1a<\/p>\n\n\n\n

        \n
      • \u4f7f\u7528 AgileX Limo PRO \u79fb\u52a8\u673a\u5668\u4eba\uff0c\u4ec5\u4f9d\u8d56 RGB \u6444\u50cf\u5934\uff08\u4e0d\u4f7f\u7528\u6df1\u5ea6\u76f8\u673a\u3001LiDAR \u7b49\uff09\u3002<\/li>\n\n\n\n
      • \u5728\u73b0\u5b9e\u5ba4\u5185\u573a\u666f\uff08\u8d70\u5eca\u3001\u4f1a\u8bae\u5ba4\u3001\u5b9e\u9a8c\u5ba4\u3001\u5f00\u653e\u7a7a\u95f4\uff09\u8fdb\u884c\u6d4b\u8bd5\u3002<\/li>\n\n\n\n
      • \u5bf9\u6bd4\u57fa\u7ebf\uff1aMapGPT\uff08\u57fa\u4e8e\u6a21\u62df\u73af\u5883\u7684\u65b9\u6cd5\uff09\u3002<\/li>\n<\/ul>\n\n\n\n

        \u8bc4\u4f30\u6307\u6807<\/strong>\uff1a\u8f68\u8ff9\u957f\u5ea6 (TL)\u3001\u5bfc\u822a\u8bef\u5dee (NE)\u3001\u6210\u529f\u7387 (SR)\u3001\u8def\u5f84\u6548\u7387 (SPL)\u3002<\/p>\n\n\n\n

        \"\"<\/figure>\n\n\n\n

        \u7ed3\u679c<\/strong>\uff1a<\/p>\n\n\n\n

          \n
        • \u7b80\u5355\u6307\u4ee4\u4efb\u52a1<\/strong>\uff08\u76ee\u6807\u641c\u7d22\u3001\u8def\u5f84\u5bfc\u822a\uff09\uff1aBrainNav \u6210\u529f\u7387\u8fdc\u9ad8\u4e8e MapGPT\uff08\u5982\u76ee\u6807\u641c\u7d22 SR = 88% vs. 0%\uff09\u3002<\/li>\n\n\n\n
        • \u590d\u6742\u4efb\u52a1<\/strong>\uff08\u591a\u6b65\u5bfc\u822a\u3001\u591a\u76ee\u6807\u641c\u7d22\u3001\u52a8\u6001\u969c\u788d\u89c4\u907f\uff09\uff1aBrainNav \u5728\u52a8\u6001\u969c\u788d\u89c4\u907f\u4e2d\u6210\u529f\u7387\u8fbe\u5230 87.5%\uff0c\u6574\u4f53\u663e\u8457\u4f18\u4e8e\u57fa\u7ebf\u3002<\/li>\n\n\n\n
        • \u56de\u6eaf\u7387<\/strong>\uff1aBrainNav \u7684\u56de\u6eaf\u7387\u4ec5 10%\uff08MapGPT \u4e3a 75%\uff09\uff0c\u4e14\u6210\u529f\u4fee\u6b63\u7387\u9ad8\u8fbe 73.3%\u3002<\/li>\n\n\n\n
        • \u6848\u4f8b\u7814\u7a76\u663e\u793a\uff0cBrainNav \u80fd\u5728\u771f\u5b9e\u573a\u666f\u4e2d\u987a\u5229\u5b8c\u6210\u591a\u9636\u6bb5\u5bfc\u822a\u3002<\/li>\n<\/ul>\n\n\n\n

          \u4e94\u3001\u603b\u7ed3<\/strong><\/h2>\n\n\n\n
            \n
          • \u8d21\u732e<\/strong>\uff1a\n
              \n
            1. \u63d0\u51fa\u4e86\u9996\u4e2a\u57fa\u4e8e\u751f\u7269\u8ba4\u77e5\u7406\u8bba\u7684\u5206\u5c42\u51b3\u7b56\u5bfc\u822a\u6846\u67b6 BrainNav\u3002<\/li>\n\n\n\n
            2. \u8bbe\u8ba1\u4e86\u53cc\u5730\u56fe + \u53cc\u65b9\u5411\u673a\u5236\uff0c\u6709\u6548\u7f13\u89e3\u7a7a\u95f4\u5e7b\u89c9\u95ee\u9898\u3002<\/li>\n\n\n\n
            3. \u5728\u73b0\u5b9e\u96f6\u6837\u672c\u73af\u5883\u4e0b\u5b9e\u73b0\u4e86\u8d85\u8d8a\u73b0\u6709 SOTA \u7684\u6027\u80fd\u3002<\/li>\n<\/ol>\n<\/li>\n\n\n\n
            4. \u5c40\u9650\u6027<\/strong>\uff1a\n
                \n
              • \u591a\u76ee\u6807 (>3) \u7684\u590d\u6742\u6307\u4ee4\u6210\u529f\u7387\u4e0b\u964d\u660e\u663e\u3002<\/li>\n\n\n\n
              • \u5728\u52a8\u6001\u969c\u788d\u573a\u666f\u4e2d\u7684\u957f\u671f\u7a33\u5b9a\u6027\u4ecd\u9700\u6539\u8fdb\u3002<\/li>\n<\/ul>\n<\/li>\n\n\n\n
              • \u672a\u6765\u65b9\u5411<\/strong>\uff1a\u589e\u5f3a\u591a\u6a21\u6001\u8bed\u4e49\u7406\u89e3\uff0c\u4f18\u5316\u52a8\u6001\u73af\u5883\u4e0b\u7684\u8def\u5f84\u89c4\u5212\u4e0e\u76ee\u6807\u8ddf\u8e2a\u3002<\/li>\n<\/ul>\n\n\n\n

                <\/p>\n","protected":false},"excerpt":{"rendered":"

                \u4f5c\u8005\uff1aLuo Ling,\u00a0Bai Qianqian\u5355\u4f4d\uff1a\u897f\u5357\u8d22\u7ecf\u5927\u5b66\u6765\u6e90\uff1aarxiv Artificial I […]<\/p>\n","protected":false},"author":63,"featured_media":0,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-3887","post","type-post","status-publish","format-standard","hentry","category-events"],"_links":{"self":[{"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts\/3887","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/users\/63"}],"replies":[{"embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/comments?post=3887"}],"version-history":[{"count":1,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts\/3887\/revisions"}],"predecessor-version":[{"id":3890,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts\/3887\/revisions\/3890"}],"wp:attachment":[{"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/media?parent=3887"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/categories?post=3887"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/tags?post=3887"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}