{"id":31883,"date":"2025-08-07T17:31:19","date_gmt":"2025-08-07T09:31:19","guid":{"rendered":"http:\/\/192.168.10.115\/?p=31883"},"modified":"2025-08-07T17:31:19","modified_gmt":"2025-08-07T09:31:19","slug":"2025-08-07-%e8%87%aa%e5%8a%a8%e9%a9%be%e9%a9%b6%e5%a4%a7%e6%a8%a1%e5%9e%8b%e6%96%b9%e6%a1%88%ef%bc%9a%e8%a7%86%e8%a7%89%e8%af%ad%e8%a8%80%e6%a8%a1%e5%9e%8bvlm%e5%b7%a5%e4%bd%9c%e4%b8%80%e8%a7%88","status":"publish","type":"post","link":"http:\/\/222.128.65.89:18086\/index.php\/2025\/08\/07\/31883\/","title":{"rendered":"2025-08-07 \u81ea\u52a8\u9a7e\u9a76\u5927\u6a21\u578b\u65b9\u6848\uff1a\u89c6\u89c9\u8bed\u8a00\u6a21\u578bVLM\u5de5\u4f5c\u4e00\u89c8\uff0c\u9762\u5411\u91cf\u4ea7\u548c\u7814\u7a76~"},"content":{"rendered":"\n<p>\u539f\u6587\u94fe\u63a5\uff1a<a href=\"https:\/\/zhuanlan.zhihu.com\/p\/1936719591561724387\">https:\/\/zhuanlan.zhihu.com\/p\/1936719591561724387<\/a><\/p>\n\n\n\n<p>\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08Vision-Language Model, VLM\uff09 \u6b63\u4ee5\u5176\u72ec\u7279\u7684\u8de8\u6a21\u6001\u7406\u89e3\u4e0e\u63a8\u7406\u80fd\u529b\uff0c\u6210\u4e3a\u8d4b\u80fd\u4e0b\u4e00\u4ee3\u81ea\u52a8\u9a7e\u9a76\u7cfb\u7edf\u7684\u5173\u952e\u5f15\u64ce\u3002VLM\u7684\u6838\u5fc3\u5728\u4e8e\u6253\u901a\u89c6\u89c9\u4e0e\u8bed\u8a00\u4e4b\u95f4\u7684\u58c1\u5792\uff0c\u8ba9\u81ea\u52a8\u9a7e\u9a76\u4e0d\u4ec5\u80fd\u201c\u770b\u89c1\u201d\u9053\u8def\uff0c\u66f4\u80fd\u50cf\u4eba\u7c7b\u4e00\u6837\u201c\u7406\u89e3\u201d\u573a\u666f\u3001\u610f\u56fe\u5e76\u8fdb\u884c\u6df1\u5c42\u6b21\u7684\u63a8\u7406\u3002<\/p>\n\n\n\n<p>\u5728\u81ea\u52a8\u9a7e\u9a76\u7684\u590d\u6742\u73af\u5883\u4e2d\uff0cVLM\u5c55\u73b0\u51fa\u5f3a\u5927\u7684\u5e94\u7528\u6f5c\u529b\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u73af\u5883\u611f\u77e5\u4e0e\u6df1\u5ea6\u7406\u89e3<\/strong>\uff1a VLM\u80fd\u591f\u8d85\u8d8a\u4f20\u7edf\u89c6\u89c9\u6a21\u578b\uff0c\u7ed3\u5408\u76f8\u673a\u56fe\u50cf\u6216\u89c6\u9891\u6d41\uff0c\u7406\u89e3\u4ea4\u901a\u573a\u666f\u4e2d\u7684\u8bed\u4e49\u4fe1\u606f\u3002\u4f8b\u5982\uff0c\u5b83\u4e0d\u4ec5\u80fd\u8bc6\u522b\u7269\u4f53\u662f\u201c\u8f66\u201d\u6216\u201c\u4eba\u201d\uff0c\u66f4\u80fd\u7406\u89e3\u201c\u884c\u4eba\u6b63\u5728\u6325\u624b\u793a\u610f\u8fc7\u9a6c\u8def\u201d\u3001\u201c\u524d\u65b9\u8f66\u8f86\u6b63\u5728\u6253\u5f00\u53cc\u95ea\u53ef\u80fd\u629b\u951a\u201d\u3001\u201c\u8def\u724c\u4e0a\u7684\u6587\u5b57\u662f\u2018\u5b66\u6821\u533a\u57df\u51cf\u901f\u6162\u884c\u2019\u201d\u7b49\u590d\u6742\u8bed\u4e49\uff0c\u4e3a\u7cfb\u7edf\u63d0\u4f9b\u66f4\u4e30\u5bcc\u3001\u66f4\u8d34\u8fd1\u4eba\u7c7b\u8ba4\u77e5\u7684\u73af\u5883\u6a21\u578b\u3002<\/li>\n\n\n\n<li><strong>\u573a\u666f\u63cf\u8ff0\u4e0e\u51b3\u7b56\u89e3\u91ca<\/strong>\uff1a VLM\u53ef\u4ee5\u5c06\u590d\u6742\u7684\u89c6\u89c9\u573a\u666f\u8f6c\u5316\u4e3a\u6e05\u6670\u7684\u81ea\u7136\u8bed\u8a00\u63cf\u8ff0\uff0c\u4e3a\u81ea\u52a8\u9a7e\u9a76\u7cfb\u7edf\u7684\u51b3\u7b56\u63d0\u4f9b\u53ef\u89e3\u91ca\u6027\u3002\u8fd9\u4e0d\u4ec5\u6709\u52a9\u4e8e\u5f00\u53d1\u8c03\u8bd5\uff08\u7406\u89e3\u7cfb\u7edf\u201c\u4e3a\u4f55\u201d\u505a\u51fa\u67d0\u4e2a\u51b3\u7b56\uff09\uff0c\u4e5f\u4e3a\u4e58\u5ba2\u6216\u76d1\u7ba1\u8005\u63d0\u4f9b\u900f\u660e\u7684\u9a7e\u9a76\u610f\u56fe\u8bf4\u660e\uff08\u5982\u201c\u6211\u6b63\u5728\u53d8\u9053\u4ee5\u907f\u5f00\u524d\u65b9\u65bd\u5de5\u533a\u57df\u201d\uff09\uff0c\u589e\u5f3a\u4fe1\u4efb\u611f\u3002<\/li>\n\n\n\n<li>\u590d\u6742\u6307\u4ee4\u7406\u89e3\u4e0e\u4ea4\u4e92\uff1a \u9762\u5411\u672a\u6765\u7684\u667a\u80fd\u5ea7\u8231\u548c\u4eba\u8f66\u4ea4\u4e92\uff0cVLM\u662f\u5b9e\u73b0\u81ea\u7136\u8bed\u8a00\u4ea4\u4e92\u7684\u6838\u5fc3\u3002\u4e58\u5ba2\u6216\u8fdc\u7a0b\u76d1\u63a7\u5458\u53ef\u4ee5\u901a\u8fc7\u53e3\u8bed\u5316\u6307\u4ee4\uff08\u5982\u201c\u5728\u4e0b\u4e00\u4e2a\u4fbf\u5229\u5e97\u95e8\u53e3\u9760\u8fb9\u505c\u4e00\u4e0b\u201d\u3001\u201c\u5c0f\u5fc3\u53f3\u8fb9\u7a81\u7136\u51b2\u51fa\u6765\u7684\u81ea\u884c\u8f66\u201d\uff09\u4e0e\u8f66\u8f86\u6c9f\u901a\uff0cVLM\u80fd\u51c6\u786e\u7406\u89e3\u610f\u56fe\u5e76\u6307\u5bfc\u7cfb\u7edf\u6267\u884c\u6216\u9884\u8b66\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u53ef\u4ee5\u8bf4\uff0cVLM\u6b63\u5c06\u81ea\u52a8\u9a7e\u9a76\u7cfb\u7edf\u7684\u611f\u77e5\u4e0e\u8ba4\u77e5\u80fd\u529b\u63a8\u5411\u65b0\u9ad8\u5ea6\uff0c\u4f7f\u5176\u4ece\u201c\u770b\u5f97\u6e05\u201d\u8d70\u5411\u201c\u61c2\u5f97\u6df1\u201d\u3002\u5b83\u4e0d\u4ec5\u662f\u7406\u89e3\u590d\u6742\u5f00\u653e\u4e16\u754c\u7684\u5173\u952e\uff0c\u4e5f\u662f\u5b9e\u73b0\u4eba\u8f66\u81ea\u7136\u534f\u540c\u3001\u6784\u5efa\u53ef\u4fe1\u8d56\u81ea\u52a8\u9a7e\u9a76\u4f53\u9a8c\u7684\u91cd\u8981\u6865\u6881\u3002\u63a5\u4e0b\u6765\uff0c\u81ea\u52a8\u9a7e\u9a76\u4e4b\u5fc3\u5c06\u4e3a\u5927\u5bb6\u7cfb\u7edf\u68b3\u7406VLM\u5728\u57fa\u7840\u6a21\u578b\u4e2d\u7684\u6700\u65b0\u7814\u7a76\u4e0e\u5e94\u7528\u8fdb\u5c55\uff0c\u63a2\u7d22\u5176\u5982\u4f55\u91cd\u5851\u81ea\u52a8\u9a7e\u9a76\u7684\u672a\u6765\u56fe\u666f\u3002\u672c\u6587\u5168\u90e8\u6587\u7ae0\u5df2\u6c47\u603b\u81f3\u300e\u81ea\u52a8\u9a7e\u9a76\u4e4b\u5fc3\u77e5\u8bc6\u661f\u7403\u300f\uff0c\u6b22\u8fce\u52a0\u5165\u6211\u4eec\uff0c\u5171\u540c\u63a2\u7d22AI\u9a71\u52a8\u7684\u81ea\u52a8\u9a7e\u9a76\u524d\u6cbf\uff01<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_0\">CrashAgent<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aCrashAgent: Crash Scenario Generation via Multi-modal Reasoning<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/www.arxiv.org\/abs\/2505.18341\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/www.arxiv.org\/abs\/2505.18341<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u5361\u5185\u57fa\u6885\u9686\u5927\u5b66\u3001NVIDIA\u3001\u897f\u5317\u5927\u5b66<\/li>\n<\/ul>\n\n\n\n<p>\u81ea\u52a8\u9a7e\u9a76\u7b97\u6cd5\u7684\u8bad\u7ec3\u4e0e\u8bc4\u4f30\u9700\u8981\u591a\u6837\u5316\u7684\u573a\u666f\u3002\u7136\u800c\uff0c\u73b0\u6709\u6570\u636e\u96c6\u4e3b\u8981\u5305\u542b\u4eba\u7c7b\u9a7e\u9a76\u5458\u5c55\u793a\u7684\u6b63\u5e38\u9a7e\u9a76\u884c\u4e3a\uff0c\u5bfc\u81f4\u5b89\u5168\u5173\u952e\u6848\u4f8b\u6570\u91cf\u6709\u9650\u3002\u8fd9\u79cd\u901a\u5e38\u88ab\u79f0\u4e3a\u957f\u5c3e\u5206\u5e03\u7684\u4e0d\u5e73\u8861\u6027\uff0c\u9650\u5236\u4e86\u9a7e\u9a76\u7b97\u6cd5\u4ece\u6d89\u53ca\u98ce\u9669\u6216\u5931\u6548\u7684\u5173\u952e\u573a\u666f\u4e2d\u5b66\u4e60\u7684\u80fd\u529b\u2014\u2014\u800c\u8fd9\u4e9b\u573a\u666f\u5bf9\u4e8e\u4eba\u7c7b\u9ad8\u6548\u63d0\u5347\u9a7e\u9a76\u6280\u80fd\u81f3\u5173\u91cd\u8981\u3002\u4e3a\u751f\u6210\u6b64\u7c7b\u573a\u666f\uff0c\u672c\u6587\u5229\u7528\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b (Multi-modal Large Language Models) \u5c06\u4e8b\u6545\u62a5\u544a\u8f6c\u6362\u4e3a\u7ed3\u6784\u5316\u573a\u666f\u683c\u5f0f\uff0c\u8be5\u683c\u5f0f\u53ef\u76f4\u63a5\u5728\u4eff\u771f\u73af\u5883\u4e2d\u6267\u884c\u3002\u5177\u4f53\u800c\u8a00\uff0c\u672c\u6587\u63d0\u51fa\u4e86 CrashAgent\uff0c\u4e00\u4e2a\u591a\u667a\u80fd\u4f53\u6846\u67b6 (multi-agent framework)\uff0c\u65e8\u5728\u89e3\u6790\u591a\u6a21\u6001\u7684\u771f\u5b9e\u4e16\u754c\u4ea4\u901a\u4e8b\u6545\u62a5\u544a\uff0c\u4ee5\u751f\u6210\u9053\u8def\u5e03\u5c40\u4ee5\u53ca\u81ea\u8f66\uff08ego vehicle\uff09\u548c\u5468\u56f4\u4ea4\u901a\u53c2\u4e0e\u8005\u7684\u884c\u4e3a\u3002\u672c\u6587\u4ece\u591a\u4e2a\u89d2\u5ea6\u5168\u9762\u8bc4\u4f30\u751f\u6210\u7684\u78b0\u649e\u573a\u666f\uff0c\u5305\u62ec\u5e03\u5c40\u91cd\u5efa\u51c6\u786e\u6027\u3001\u78b0\u649e\u53d1\u751f\u7387\u548c\u591a\u6837\u6027\u3002\u6700\u7ec8\u4ea7\u751f\u7684\u9ad8\u8d28\u91cf\u3001\u5927\u89c4\u6a21\u78b0\u649e\u6570\u636e\u96c6\u5c06\u516c\u5f00\u63d0\u4f9b (publicly available)\uff0c\u4ee5\u652f\u6301\u5f00\u53d1\u80fd\u591f\u5904\u7406\u5b89\u5168\u5173\u952e\u60c5\u51b5\u7684\u81ea\u52a8\u9a7e\u9a76\u7b97\u6cd5\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pica.zhimg.com\/v2-30a275c4b4ccd4e9b1d06c783c2ec0b0_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/picx.zhimg.com\/v2-a29479e1913ff7fe14bf100ae698556b_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_1\">CurricuVLM<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aCurricuVLM: Towards Safe Autonomous Driving via Personalized Safety-Critical Curriculum Learning with Vision-Language Models<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2502.15119\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2502.15119<\/a><\/li>\n\n\n\n<li>\u9879\u76ee\u4e3b\u9875\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/zihaosheng.github.io\/CurricuVLM\/\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/zihaosheng.github.io\/CurricuVLM\/<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u7f8e\u56fd\u5a01\u65af\u5eb7\u661f\u5927\u5b66\u9ea6\u8fea\u900a\u5206\u6821\uff0c\u666e\u6e21\u5927\u5b66\uff0c\u8c37\u6b4c<\/li>\n<\/ul>\n\n\n\n<p>\u786e\u4fdd\u81ea\u52a8\u9a7e\u9a76\u7cfb\u7edf\u7684\u5b89\u5168\u4ecd\u7136\u662f\u4e00\u4e2a\u5173\u952e\u6311\u6218\uff0c\u5c24\u5176\u662f\u5728\u5904\u7406\u7f55\u89c1\u4f46\u53ef\u80fd\u707e\u96be\u6027\u7684\u5b89\u5168\u5173\u952e\u573a\u666f\u65f6\u3002\u867d\u7136\u73b0\u6709\u7814\u7a76\u63a2\u7d22\u4e86\u4e3a\u81ea\u52a8\u9a7e\u9a76\u8f66\u8f86\uff08AV\uff09\u6d4b\u8bd5\u751f\u6210\u5b89\u5168\u5173\u952e\u573a\u666f\uff0c\u4f46\u5982\u4f55\u6709\u6548\u5730\u5c06\u8fd9\u4e9b\u573a\u666f\u7eb3\u5165\u7b56\u7565\u5b66\u4e60\u4ee5\u589e\u5f3a\u5b89\u5168\u6027\u7684\u5de5\u4f5c\u4ecd\u7136\u6709\u9650\u3002\u6b64\u5916\uff0c\u5f00\u53d1\u9002\u5e94\u81ea\u52a8\u9a7e\u9a76\u8f66\u8f86\u4e0d\u65ad\u6f14\u5316\u7684\u884c\u4e3a\u6a21\u5f0f\u4e0e\u6027\u80fd\u74f6\u9888\u7684\u8bad\u7ec3\u8bfe\u7a0b\u5728\u5f88\u5927\u7a0b\u5ea6\u4e0a\u5c1a\u672a\u88ab\u63a2\u7d22\u3002\u4e3a\u5e94\u5bf9\u8fd9\u4e9b\u6311\u6218\uff0c\u672c\u6587\u63d0\u51fa\u4e86 CurricuVLM\uff0c\u4e00\u4e2a\u5229\u7528\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08VLMs\uff09 \u4e3a\u81ea\u52a8\u9a7e\u9a76\u667a\u80fd\u4f53\u5b9e\u73b0\u4e2a\u6027\u5316\u8bfe\u7a0b\u5b66\u4e60\u7684\u65b0\u9896\u6846\u67b6\u3002\u672c\u6587\u7684\u65b9\u6cd5\u72ec\u7279\u5730\u5229\u7528\u4e86VLMs\u7684\u591a\u6a21\u6001\u7406\u89e3\u80fd\u529b\u6765\u5206\u6790\u667a\u80fd\u4f53\u884c\u4e3a\u3001\u8bc6\u522b\u6027\u80fd\u5f31\u70b9\uff0c\u5e76\u52a8\u6001\u751f\u6210\u91cf\u8eab\u5b9a\u5236\u7684\u8bad\u7ec3\u573a\u666f\u4ee5\u8fdb\u884c\u8bfe\u7a0b\u9002\u5e94\u3002\u901a\u8fc7\u5bf9\u4e0d\u5b89\u5168\u9a7e\u9a76\u60c5\u5883\u7684\u53d9\u4e8b\u63cf\u8ff0\u8fdb\u884c\u7efc\u5408\u5206\u6790\uff0cCurricuVLM\u6267\u884c\u6df1\u5165\u63a8\u7406\u4ee5\u8bc4\u4f30\u81ea\u52a8\u9a7e\u9a76\u8f66\u8f86\u7684\u80fd\u529b\u5e76\u8bc6\u522b\u5173\u952e\u884c\u4e3a\u6a21\u5f0f\u3002\u8be5\u6846\u67b6\u968f\u540e\u5408\u6210\u9488\u5bf9\u8fd9\u4e9b\u5df2\u8bc6\u522b\u5c40\u9650\u6027\u7684\u5b9a\u5236\u5316\u8bad\u7ec3\u573a\u666f\uff0c\u4ece\u800c\u5b9e\u73b0\u6709\u6548\u4e14\u4e2a\u6027\u5316\u7684\u8bfe\u7a0b\u5b66\u4e60\u3002\u5728Waymo Open Motion\u6570\u636e\u96c6\u4e0a\u8fdb\u884c\u7684\u5927\u91cf\u5b9e\u9a8c\u8868\u660e\uff0cCurricuVLM\u5728\u5e38\u89c4\u548c\u5b89\u5168\u5173\u952e\u573a\u666f\u4e0b\u5747\u4f18\u4e8e\u6700\u5148\u8fdb\u7684\u57fa\u7ebf\u65b9\u6cd5\uff0c\u5728\u5bfc\u822a\u6210\u529f\u7387\u3001\u9a7e\u9a76\u6548\u7387\u548c\u5b89\u5168\u6027\u6307\u6807\u65b9\u9762\u5747\u53d6\u5f97\u4e86\u5353\u8d8a\u6027\u80fd\u3002\u8fdb\u4e00\u6b65\u5206\u6790\u63ed\u793a\uff0cCurricuVLM\u53ef\u4f5c\u4e3a\u901a\u7528\u65b9\u6cd5\uff0c\u4e0e\u5404\u79cd\u5f3a\u5316\u5b66\u4e60\uff08RL\uff09\u7b97\u6cd5\u96c6\u6210\u4ee5\u589e\u5f3a\u81ea\u52a8\u9a7e\u9a76\u7cfb\u7edf\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic1.zhimg.com\/v2-b3bd83563d2be96c1ad3d554dd6afd3a_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic3.zhimg.com\/v2-ea0ea4d450a9a404b7a921f7b0580690_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_2\">From Accidents to Insights<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aFrom Accidents to Insights: Leveraging Multimodal Data for Scenario-Driven ADS Testing<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2502.02025\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2502.02025<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u9ea6\u8003\u745e\u5927\u5b66\uff0c\u5317\u5fb7\u514b\u8428\u65af\u5927\u5b66<\/li>\n<\/ul>\n\n\n\n<p>\u81ea\u52a8\u9a7e\u9a76\u7cfb\u7edf\uff08ADS\uff09\u7684\u5feb\u901f\u53d1\u5c55\u4e9f\u9700\u9c81\u68d2\u7684\u8f6f\u4ef6\u6d4b\u8bd5\u6765\u786e\u4fdd\u5b89\u5168\u6027\u548c\u53ef\u9760\u6027\u3002\u7136\u800c\uff0c\u81ea\u52a8\u5316\u751f\u6210\u53ef\u6269\u5c55\u4e14\u5177\u4f53\u7684\u6d4b\u8bd5\u573a\u666f\u4ecd\u662f\u91cd\u5927\u6311\u6218\u3002\u5f53\u524d\u57fa\u4e8e\u573a\u666f\u7684\u6d4b\u8bd5\u7528\u4f8b\u751f\u6210\u65b9\u6cd5\u5e38\u9762\u4e34\u573a\u666f\u4e0d\u771f\u5b9e\u3001\u8f66\u8f86\u8f68\u8ff9\u4e0d\u51c6\u786e\u7b49\u5c40\u9650\uff0c\u4e3b\u8981\u6e90\u4e8e\u6570\u636e\u63d0\u53d6\u8fc7\u7a0b\u4e2d\u7684\u5730\u56fe\u4fe1\u606f\u4e22\u5931\uff0c\u4ee5\u53ca\u7f3a\u4e4f\u6709\u6548\u673a\u5236\u7f13\u89e3\u5927\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u7684\u5e7b\u89c9\u95ee\u9898\u3002\u672c\u6587\u63d0\u51fa <a href=\"https:\/\/zhida.zhihu.com\/search?content_id=261363339&amp;content_type=Article&amp;match_order=1&amp;q=TRACE&amp;zhida_source=entity\" target=\"_blank\" rel=\"noreferrer noopener\">TRACE<\/a><\/p>\n\n\n\n<p>\uff08\u57fa\u4e8e\u5173\u952e\u573a\u666f\u7684ADS\u6d4b\u8bd5\u7528\u4f8b\u751f\u6210\u6846\u67b6\uff09\uff0c\u901a\u8fc7\u5229\u7528\u591a\u6a21\u6001\u6570\u636e\u4ece\u771f\u5b9e\u8f66\u7978\u62a5\u544a\u4e2d\u63d0\u53d6\u9ad8\u6311\u6218\u6027\u573a\u666f\uff0c\u4ee5\u8f83\u5c11\u6570\u636e\u6784\u5efa\u5927\u91cf\u5173\u952e\u6d4b\u8bd5\u7528\u4f8b\uff0c\u663e\u8457\u63d0\u5347ADS\u7f3a\u9677\u68c0\u6d4b\u6548\u7387\u3002\u7ed3\u5408\u4e0a\u4e0b\u6587\u5b66\u4e60\u3001\u601d\u7ef4\u94fe\u63d0\u793a\u548c\u81ea\u9a8c\u8bc1\u6280\u672f\uff0c\u672c\u6587\u5229\u7528LLM\u4ece\u4e8b\u6545\u62a5\u544a\u4e2d\u63d0\u53d6\u73af\u5883\u4e0e\u8def\u7f51\u4fe1\u606f\uff1b\u5728\u8f66\u8f86\u8f68\u8ff9\u89c4\u5212\u4e2d\uff0c\u96c6\u6210\u542b\u5730\u56fe\u4fe1\u606f\u548c\u8f66\u8f86\u5750\u6807\u7684\u6570\u636e\u4f5c\u4e3a\u77e5\u8bc6\u5e93\uff0c\u6784\u5efa\u5177\u5907\u8def\u5f84\u89c4\u5212\u80fd\u529b\u7684LLM\u6a21\u578b TraceMate\u3002\u57fa\u4e8e50\u4efd\u771f\u5b9e\u4e8b\u6545\u62a5\u544a\uff0c\u672c\u65b9\u6cd5\u5728MetaDrive\u548cBeamNG\u4eff\u771f\u5e73\u53f0\u4e0a\u6210\u529f\u6d4b\u8bd5\u4e09\u79cdADS\u6a21\u578b\u3002\u5728\u751f\u6210\u7684290\u4e2a\u6d4b\u8bd5\u573a\u666f\u4e2d\uff0c127\u4e2a\u88ab\u8bc6\u522b\u4e3a\u5173\u952e\u573a\u666f\uff08\u5f15\u53d1\u8f66\u8f86\u78b0\u649e\uff09\u3002\u7528\u6237\u53cd\u9988\u8868\u660e\uff0cTRACE\u7684\u573a\u666f\u91cd\u5efa\u51c6\u786e\u7387\u663e\u8457\u4f18\u4e8e\u73b0\u6709\u6280\u672f\u2014\u201477.5%\u7684\u573a\u666f\u88ab\u8bc4\u4e3a&#8221;\u57fa\u672c&#8221;\u6216&#8221;\u5b8c\u5168&#8221;\u4e00\u81f4\uff0c\u800c\u6700\u4f73\u57fa\u7ebf\u65b9\u6cd5SOTA-LCTGen\u4ec5\u8fbe27%\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic1.zhimg.com\/v2-30248e5880a8693971d9584e41f9ddc4_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic4.zhimg.com\/v2-c890bb508644089409dd21630894b58b_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_3\">Generating OOD<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aGenerating Out-Of-Distribution Scenarios Using Language Models<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2411.16554\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2411.16554<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u9ebb\u7701\u7406\u5de5\u5b66\u9662\uff0c\u9a6c\u8428\u8bf8\u585e\u5927\u5b66\u963f\u9ed8\u65af\u7279\u5206\u6821 \uff0c\u4e30\u7530\u7814\u7a76\u9662<\/li>\n<\/ul>\n\n\n\n<p>\u7531\u673a\u5668\u5b66\u4e60\u6280\u672f\u9a71\u52a8\u7684\u81ea\u52a8\u9a7e\u9a76\u6c7d\u8f66\u7684\u90e8\u7f72\uff0c\u9700\u8981\u5728\u591a\u6837\u5316\u7684\u73b0\u5b9e\u4e16\u754c\u73af\u5883\u4e2d\u8fdb\u884c\u5e7f\u6cdb\u6d4b\u8bd5\uff0c\u7a33\u5065\u5904\u7406\u8fb9\u7f18\u6848\u4f8b\u548c\u5206\u5e03\u5916\uff08Out-Of-Distribution, OOD\uff09\u573a\u666f\uff0c\u5e76\u8fdb\u884c\u5168\u9762\u7684\u5b89\u5168\u9a8c\u8bc1\uff0c\u4ee5\u786e\u4fdd\u8fd9\u4e9b\u7cfb\u7edf\u5728\u4e0d\u53ef\u9884\u6d4b\u6761\u4ef6\u4e0b\u80fd\u591f\u5b89\u5168\u6709\u6548\u5730\u884c\u9a76\u3002\u89e3\u51b3\u5206\u5e03\u5916\uff08OOD\uff09\u9a7e\u9a76\u573a\u666f\u5bf9\u4e8e\u63d0\u5347\u5b89\u5168\u6027\u81f3\u5173\u91cd\u8981\uff0c\u56e0\u4e3aOOD\u573a\u666f\u6709\u52a9\u4e8e\u9a8c\u8bc1\u8f66\u8f86\u81ea\u4e3b\u7cfb\u7edf\u6280\u672f\u6808\u4e2d\u6a21\u578b\u7684\u53ef\u9760\u6027\u3002\u7136\u800c\uff0c\u7531\u4e8eOOD\u573a\u666f\u5728\u957f\u5c3e\u5206\u5e03\u4e2d\u7684\u7a00\u758f\u6027\u53ca\u5176\u5728\u57ce\u5e02\u9a7e\u9a76\u6570\u636e\u96c6\u4e2d\u7684\u7f55\u89c1\u6027\uff0c\u751f\u6210\u6b64\u7c7b\u573a\u666f\u6781\u5177\u6311\u6218\u6027\u3002\u8fd1\u671f\uff0c\u5927\u8bed\u8a00\u6a21\u578b\uff08Large Language Models, LLMs\uff09\u5728\u81ea\u52a8\u9a7e\u9a76\u9886\u57df\u5c55\u73b0\u51fa\u6f5c\u529b\uff0c\u7279\u522b\u662f\u5728\u96f6\u6837\u672c\u6cdb\u5316\uff08zero-shot generalization\uff09\u548c\u5e38\u8bc6\u63a8\u7406\uff08common-sense reasoning\uff09\u80fd\u529b\u65b9\u9762\u3002\u672c\u6587\u5229\u7528LLMs\u7684\u8fd9\u4e9b\u4f18\u52bf\uff0c\u63d0\u51fa\u4e00\u4e2a\u7528\u4e8e\u751f\u6210\u591a\u6837\u5316OOD\u9a7e\u9a76\u573a\u666f\u7684\u6846\u67b6\u3002\u672c\u6587\u7684\u65b9\u6cd5\u5229\u7528LLMs\u6784\u5efa\u4e00\u4e2a\u5206\u652f\u6811\u7ed3\u6784\uff08branching tree\uff09\uff0c\u5176\u4e2d\u6bcf\u6761\u5206\u652f\u4ee3\u8868\u4e00\u4e2a\u72ec\u7279\u7684OOD\u573a\u666f\u3002\u8fd9\u4e9b\u573a\u666f\u968f\u540e\u901a\u8fc7\u4e00\u4e2a\u81ea\u52a8\u5316\u6846\u67b6\u5728CARLA\u4eff\u771f\u5668\u4e2d\u5b9e\u73b0\uff0c\u8be5\u6846\u67b6\u786e\u4fdd\u573a\u666f\u589e\u5f3a\uff08scene augmentation\uff09\u4e0e\u76f8\u5e94\u7684\u6587\u672c\u63cf\u8ff0\u4fdd\u6301\u4e00\u81f4\u3002\u672c\u6587\u901a\u8fc7\u5e7f\u6cdb\u7684\u4eff\u771f\u5b9e\u9a8c\u8bc4\u4f30\u4e86\u8be5\u6846\u67b6\uff0c\u5e76\u91c7\u7528\u4e00\u4e2a\u8861\u91cf\u573a\u666f\u4e30\u5bcc\u5ea6\u7684\u591a\u6837\u6027\u6307\u6807\uff08diversity metric\uff09\u6765\u8bc4\u4f30\u5176\u6027\u80fd\u3002\u6b64\u5916\uff0c\u672c\u6587\u5f15\u5165\u4e86\u4e00\u79cd\u65b0\u7684\u201cOOD\u504f\u79bb\u5ea6\u201d\uff08OOD-ness\uff09\u6307\u6807\uff0c\u7528\u4e8e\u91cf\u5316\u6240\u751f\u6210\u573a\u666f\u504f\u79bb\u5178\u578b\u57ce\u5e02\u9a7e\u9a76\u6761\u4ef6\u7684\u7a0b\u5ea6\u3002\u8fdb\u4e00\u6b65\u5730\uff0c\u672c\u6587\u63a2\u7d22\u4e86\u73b0\u4ee3\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08Vision-Language Models, VLMs\uff09\u5728\u89e3\u8bfb\u6240\u4eff\u771f\u7684OOD\u573a\u666f\u5e76\u5b9e\u73b0\u5b89\u5168\u5bfc\u822a\u65b9\u9762\u7684\u80fd\u529b\u3002\u672c\u6587\u7684\u7814\u7a76\u7ed3\u679c\u4e3a\u8bed\u8a00\u6a21\u578b\u5728\u89e3\u51b3\u57ce\u5e02\u9a7e\u9a76\u80cc\u666f\u4e0bOOD\u573a\u666f\u95ee\u9898\u4e2d\u7684\u53ef\u9760\u6027\u63d0\u4f9b\u4e86\u6709\u4ef7\u503c\u7684\u89c1\u89e3\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic2.zhimg.com\/v2-0420bde01a1ad036d62c0642878696e5_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic2.zhimg.com\/v2-a01e811fc361b25bd5912b6fc3ab27bf_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_4\">From Dashcam Videos to Driving Simulations<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aFrom Dashcam Videos to Driving Simulations: Stress Testing Automated Vehicles against Rare Events<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2411.16027\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2411.16027<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u4e30\u7530\u5317\u7f8e\u7814\u7a76\u9662\uff0c\u4f0a\u5229\u8bfa\u4f0a\u5927\u5b66\u5384\u5df4\u7eb3 &#8211; \u9999\u69df\u5206\u6821<\/li>\n<\/ul>\n\n\n\n<p>\u5728\u4eff\u771f\u73af\u5883\u4e2d\u4f7f\u7528\u771f\u5b9e\u7684\u9a7e\u9a76\u573a\u666f\u6d4b\u8bd5\u81ea\u52a8\u9a7e\u9a76\u7cfb\u7edf\uff08ADS\uff09\u5bf9\u4e8e\u9a8c\u8bc1\u5176\u6027\u80fd\u81f3\u5173\u91cd\u8981\u3002\u7136\u800c\uff0c\u5c06\u771f\u5b9e\u4e16\u754c\u7684\u9a7e\u9a76\u89c6\u9891\u8f6c\u5316\u4e3a\u4eff\u771f\u573a\u666f\u9762\u4e34\u91cd\u5927\u6311\u6218\uff0c\u8fd9\u6e90\u4e8e\u9ad8\u7ef4\u89c6\u9891\u6570\u636e\u89e3\u91ca\u7684\u590d\u6742\u6027\u4ee5\u53ca\u7cbe\u786e\u624b\u52a8\u573a\u666f\u91cd\u5efa\u7684\u8017\u65f6\u6027\u3002\u5728\u672c\u7814\u7a76\u4e2d\uff0c\u672c\u6587\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u9896\u7684\u6846\u67b6\uff0c\u53ef\u81ea\u52a8\u5c06\u771f\u5b9e\u4e16\u754c\u7684\u8f66\u8f86\u78b0\u649e\u89c6\u9891\u8f6c\u6362\u4e3a\u7528\u4e8eADS\u6d4b\u8bd5\u7684\u8be6\u7ec6\u4eff\u771f\u573a\u666f\u3002\u672c\u6587\u7684\u65b9\u6cd5\u5229\u7528\u63d0\u793a\u5de5\u7a0b\u4f18\u5316\u7684\u89c6\u9891\u8bed\u8a00\u6a21\u578b\uff08VLM\uff09\uff0c\u5c06\u884c\u8f66\u8bb0\u5f55\u4eea\uff08dashcam\uff09\u89c6\u9891\u7247\u6bb5\u8f6c\u6362\u4e3aSCENIC\u811a\u672c\u3002\u8fd9\u4e9b\u811a\u672c\u5b9a\u4e49\u4e86CARLA\u4eff\u771f\u5668\u4e2d\u7684\u73af\u5883\u548c\u9a7e\u9a76\u884c\u4e3a\uff0c\u4ece\u800c\u80fd\u591f\u751f\u6210\u903c\u771f\u7684\u4eff\u771f\u573a\u666f\u3002\u91cd\u8981\u7684\u662f\uff0c\u672c\u6587\u7684\u6846\u67b6\u5e76\u975e\u5355\u7eaf\u8ffd\u6c42\u4e00\u5bf9\u4e00\u7684\u573a\u666f\u91cd\u5efa\uff0c\u800c\u662f\u4fa7\u91cd\u4e8e\u4ece\u539f\u59cb\u89c6\u9891\u4e2d\u6355\u6349\u6838\u5fc3\u9a7e\u9a76\u884c\u4e3a\uff0c\u540c\u65f6\u63d0\u4f9b\u5929\u6c14\u6216\u9053\u8def\u6761\u4ef6\u7b49\u53c2\u6570\u7684\u7075\u6d3b\u6027\uff0c\u4ee5\u652f\u6301\u57fa\u4e8e\u641c\u7d22\u7684\u6d4b\u8bd5\u3002\u6b64\u5916\uff0c\u672c\u6587\u5f15\u5165\u4e86\u4e00\u79cd\u76f8\u4f3c\u6027\u5ea6\u91cf\u6307\u6807\uff0c\u901a\u8fc7\u6bd4\u8f83\u771f\u5b9e\u89c6\u9891\u4e0e\u4eff\u771f\u89c6\u9891\u4e4b\u95f4\u9a7e\u9a76\u884c\u4e3a\u7684\u5173\u952e\u7279\u5f81\u6765\u63d0\u4f9b\u53cd\u9988\uff0c\u4ece\u800c\u8fed\u4ee3\u5730\u4f18\u5316\u751f\u6210\u7684\u573a\u666f\u3002\u672c\u6587\u7684\u521d\u6b65\u7ed3\u679c\u8868\u660e\uff0c\u8be5\u65b9\u6cd5\u5177\u6709\u663e\u8457\u7684\u65f6\u95f4\u6548\u7387\uff0c\u53ef\u5728\u6570\u5206\u949f\u5185\u5b8c\u6210\u5168\u81ea\u52a8\u3001\u65e0\u9700\u4eba\u5de5\u5e72\u9884\u7684\u771f\u5b9e\u5230\u4eff\u771f\uff08real-to-sim\uff09\u8f6c\u6362\uff0c\u540c\u65f6\u4fdd\u6301\u5bf9\u539f\u59cb\u9a7e\u9a76\u4e8b\u4ef6\u7684\u9ad8\u4fdd\u771f\u5ea6\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic4.zhimg.com\/v2-905220786c58b48b78768f02a88c829b_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic4.zhimg.com\/v2-26ff4f5fd088e574bc7965a24a269f8d_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_5\"><a href=\"https:\/\/zhida.zhihu.com\/search?content_id=261363339&amp;content_type=Article&amp;match_order=1&amp;q=OmniTester&amp;zhida_source=entity\" target=\"_blank\" rel=\"noreferrer noopener\">OmniTester<\/a><\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aMultimodal Large Language Model Driven Scenario Testing for Autonomous Vehicles<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2409.06450\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2409.06450<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u6e05\u534e\u5927\u5b66\uff0c\u516c\u5b89\u90e8\u9053\u8def\u4ea4\u901a\u5b89\u5168\u7814\u7a76\u9662<\/li>\n<\/ul>\n\n\n\n<p>\u6781\u7aef\u6848\u4f8b\uff08corner cases\uff09\u7684\u751f\u6210\u5bf9\u4e8e\u81ea\u52a8\u9a7e\u9a76\u8f66\u8f86\uff08AV\uff09\u9053\u8def\u90e8\u7f72\u524d\u7684\u9ad8\u6548\u6d4b\u8bd5\u81f3\u5173\u91cd\u8981\u3002\u7136\u800c\uff0c\u73b0\u6709\u65b9\u6cd5\u96be\u4ee5\u6ee1\u8db3\u591a\u6837\u5316\u7684\u6d4b\u8bd5\u9700\u6c42\uff0c\u4e14\u666e\u904d\u7f3a\u4e4f\u5bf9\u672a\u77e5\u573a\u666f\u7684\u6cdb\u5316\u80fd\u529b\uff0c\u4ece\u800c\u964d\u4f4e\u4e86\u751f\u6210\u573a\u666f\u7684\u5b9e\u7528\u6027\u4e0e\u53ef\u7528\u6027\u3002\u672c\u7814\u7a76\u63d0\u51fa OmniTester\uff1a\u4e00\u79cd\u57fa\u4e8e\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u7684\u6846\u67b6\uff0c\u5145\u5206\u5229\u7528LLM\u7684\u4e16\u754c\u77e5\u8bc6\u4e0e\u63a8\u7406\u80fd\u529b\uff0c\u5728\u4eff\u771f\u73af\u5883\u4e2d\u751f\u6210\u9ad8\u771f\u5b9e\u6027\u4e0e\u591a\u6837\u5316\u7684\u6d4b\u8bd5\u573a\u666f\u3002\u9664\u63d0\u793a\u5de5\u7a0b\u5916\uff0c\u672c\u6587\u96c6\u6210\u4ea4\u901a\u4eff\u771f\u5de5\u5177SUMO\u4ee5\u7b80\u5316LLM\u751f\u6210\u7684\u4ee3\u7801\u590d\u6742\u5ea6\uff0c\u5e76\u5f15\u5165\u68c0\u7d22\u589e\u5f3a\u751f\u6210\uff08RAG\uff09\u4e0e\u81ea\u6539\u8fdb\u673a\u5236\uff0c\u589e\u5f3aLLM\u5bf9\u573a\u666f\u7684\u7406\u89e3\u80fd\u529b\uff0c\u63d0\u5347\u751f\u6210\u573a\u666f\u7684\u771f\u5b9e\u6027\u3002\u5b9e\u9a8c\u8bc1\u660e\uff0c\u8be5\u65b9\u6cd5\u5728\u751f\u6210\u4e09\u7c7b\u590d\u6742\u6311\u6218\u6027\u573a\u666f\u65f6\u5177\u5907\u4f18\u5f02\u7684\u53ef\u63a7\u6027\u4e0e\u771f\u5b9e\u6027\uff0c\u540c\u65f6\u4f9d\u6258LLM\u7684\u6cdb\u5316\u80fd\u529b\uff0c\u6210\u529f\u5b9e\u73b0\u4e86\u57fa\u4e8e\u4e8b\u6545\u62a5\u544a\u7684\u65b0\u573a\u666f\u91cd\u5efa\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/picx.zhimg.com\/v2-4e132a78f2bc5f305315ece153f5161d_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic4.zhimg.com\/v2-0b141048da1ee99ed1a7b2af1471a8a7_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_6\">DriveGenVLM<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aDriveGenVLM: Real-world Video Generation for Vision Language Model based Autonomous Driving<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/ieeexplore.ieee.org\/abstract\/document\/10786438\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/ieeexplore.ieee.org\/abstract\/document\/10786438<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u54e5\u4f26\u6bd4\u4e9a\u5927\u5b66<\/li>\n<\/ul>\n\n\n\n<p>\u81ea\u52a8\u9a7e\u9a76\u6280\u672f\u7684\u8fdb\u6b65\u4e9f\u9700\u66f4\u5148\u8fdb\u7684\u65b9\u6cd5\u6765\u7406\u89e3\u548c\u9884\u6d4b\u73b0\u5b9e\u573a\u666f\u3002\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08VLMs\uff09\u4f5c\u4e3a\u9769\u547d\u6027\u5de5\u5177\u5d2d\u9732\u5934\u89d2\uff0c\u5728\u81ea\u52a8\u9a7e\u9a76\u9886\u57df\u5c55\u73b0\u51fa\u5de8\u5927\u6f5c\u529b\u3002\u672c\u6587\u63d0\u51faDriveGenVLM\u6846\u67b6\uff0c\u901a\u8fc7\u751f\u6210\u9a7e\u9a76\u89c6\u9891\u5e76\u5229\u7528VLMs\u8fdb\u884c\u573a\u666f\u7406\u89e3\u3002\u4e3a\u5b9e\u73b0\u8fd9\u4e00\u76ee\u6807\uff0c\u672c\u6587\u91c7\u7528\u57fa\u4e8e\u53bb\u566a\u6269\u6563\u6982\u7387\u6a21\u578b\uff08DDPM\uff09\u7684\u89c6\u9891\u751f\u6210\u6846\u67b6\uff0c\u65e8\u5728\u9884\u6d4b\u771f\u5b9e\u4e16\u754c\u89c6\u9891\u5e8f\u5217\u3002\u968f\u540e\u901a\u8fc7\u9884\u8bad\u7ec3\u6a21\u578b&#8221;\u7b2c\u4e00\u4eba\u79f0\u89c6\u89d2\u89c6\u9891\u9ad8\u6548\u4e0a\u4e0b\u6587\u5b66\u4e60&#8221;\uff08EILEV\uff09\u8bc4\u4f30\u751f\u6210\u89c6\u9891\u5728VLMs\u4e2d\u7684\u9002\u7528\u6027\u3002\u8be5\u6269\u6563\u6a21\u578b\u57fa\u4e8eWaymo\u5f00\u653e\u6570\u636e\u96c6\u8bad\u7ec3\uff0c\u5e76\u91c7\u7528\u5f17\u96f7\u6b47\u89c6\u9891\u8ddd\u79bb\uff08FVD\uff09\u6307\u6807\u8bc4\u4f30\u751f\u6210\u89c6\u9891\u7684\u8d28\u91cf\u4e0e\u771f\u5b9e\u6027\u3002EILEV\u4e3a\u751f\u6210\u89c6\u9891\u63d0\u4f9b\u573a\u666f\u63cf\u8ff0\uff0c\u8fd9\u4e9b\u63cf\u8ff0\u5728\u81ea\u52a8\u9a7e\u9a76\u9886\u57df\u5177\u6709\u91cd\u8981\u4ef7\u503c\uff1a\u53ef\u589e\u5f3a\u4ea4\u901a\u573a\u666f\u7406\u89e3\u80fd\u529b\u3001\u8f85\u52a9\u5bfc\u822a\u51b3\u7b56\u5e76\u63d0\u5347\u8def\u5f84\u89c4\u5212\u80fd\u529b\u3002DriveGenVLM\u6846\u67b6\u5c06\u89c6\u9891\u751f\u6210\u4e0eVLMs\u6df1\u5ea6\u878d\u5408\uff0c\u6807\u5fd7\u7740\u5229\u7528\u5148\u8fdb\u4eba\u5de5\u667a\u80fd\u6a21\u578b\u89e3\u51b3\u81ea\u52a8\u9a7e\u9a76\u590d\u6742\u6311\u6218\u7684\u91cd\u5927\u7a81\u7834\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_7\">WEDGE<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aWEDGE: A multi-weather autonomous driving dataset built from generative vision-language models<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/pdf\/2305.07528\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/pdf\/2305.07528<\/a><\/li>\n\n\n\n<li>\u9879\u76ee\u4e3b\u9875\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/infernolia.github.io\/WEDGE\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/infernolia.github.io\/WEDGE<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u5361\u5185\u57fa\u6885\u9686\u5927\u5b66\uff0c\u5370\u5ea6\u5171\u751f\u56fd\u9645\u5927\u5b66<\/li>\n<\/ul>\n\n\n\n<p>\u5f00\u653e\u9053\u8def\u73af\u5883\u5bf9\u81ea\u52a8\u9a7e\u9a76\u611f\u77e5\u63d0\u51fa\u4e86\u8bf8\u591a\u6311\u6218\uff0c\u5176\u4e2d\u6781\u7aef\u5929\u6c14\u6761\u4ef6\u5bfc\u81f4\u7684\u4f4e\u80fd\u89c1\u5ea6\u5c24\u4e3a\u7a81\u51fa\u3002\u5728\u826f\u597d\u5929\u6c14\u6570\u636e\u96c6\u4e0a\u8bad\u7ec3\u7684\u6a21\u578b\uff0c\u5728\u9762\u5bf9\u8fd9\u4e9b\u5206\u5e03\u5916\uff08out-of-distribution\uff09\u573a\u666f\u65f6\uff0c\u5176\u68c0\u6d4b\u6027\u80fd\u5f80\u5f80\u663e\u8457\u4e0b\u964d\u3002\u4e3a\u589e\u5f3a\u611f\u77e5\u7cfb\u7edf\u7684\u5bf9\u6297\u9c81\u68d2\u6027\uff08adversarial robustness\uff09\uff0c\u672c\u6587\u63d0\u51fa\u4e86 WEDGE\uff08WEather images by DALL-E GEneration\uff09\uff1a\u4e00\u4e2a\u901a\u8fc7\u63d0\u793a\uff08prompting\uff09\u5229\u7528\u89c6\u89c9\u8bed\u8a00\u751f\u6210\u6a21\u578b\u6784\u5efa\u7684\u5408\u6210\u6570\u636e\u96c6\u3002WEDGE \u5305\u542b 3360 \u5f20\u56fe\u50cf\uff0c\u6db5\u76d6 16 \u79cd\u6781\u7aef\u5929\u6c14\u6761\u4ef6\uff0c\u5e76\u624b\u5de5\u6807\u6ce8\u4e86 16513 \u4e2a\u8fb9\u754c\u6846\uff0c\u652f\u6301\u5929\u6c14\u5206\u7c7b\u548c 2D \u76ee\u6807\u68c0\u6d4b\u4efb\u52a1\u7684\u7814\u7a76\u3002\u672c\u6587\u4ece\u7814\u7a76\u89d2\u5ea6\u5206\u6790\u4e86 WEDGE\uff0c\u9a8c\u8bc1\u4e86\u5176\u5bf9\u4e8e\u6781\u7aef\u5929\u6c14\u81ea\u52a8\u9a7e\u9a76\u611f\u77e5\u7684\u6709\u6548\u6027\u3002\u672c\u6587\u4e3a\u5206\u7c7b\u548c\u68c0\u6d4b\u4efb\u52a1\u5efa\u7acb\u4e86\u57fa\u7ebf\u6027\u80fd\uff0c\u6d4b\u8bd5\u51c6\u786e\u7387\u8fbe\u5230 53.87%\uff0c\u5e73\u5747\u7cbe\u5ea6\uff08mAP\uff09\u8fbe\u5230 45.41\u3002\u6700\u91cd\u8981\u7684\u662f\uff0cWEDGE \u53ef\u7528\u4e8e\u5fae\u8c03\uff08fine-tune\uff09 \u6700\u5148\u8fdb\u7684\u68c0\u6d4b\u5668\uff0c\u5728\u771f\u5b9e\u4e16\u754c\u5929\u6c14\u57fa\u51c6\u6d4b\u8bd5\uff08\u5982 DAWN\uff09\u4e0a\uff0c\u5c06 SOTA \u6027\u80fd\u63d0\u5347\u4e86 4.48 AP\uff08\u5bf9\u4e8e\u5361\u8f66\u7b49\u751f\u6210\u6548\u679c\u826f\u597d\u7684\u7c7b\u522b\uff09\u3002WEDGE \u7684\u6536\u96c6\u9075\u5faa OpenAI \u7684\u4f7f\u7528\u6761\u6b3e\uff0c\u5e76\u4ee5 CC BY-NC-SA 4.0 \u8bb8\u53ef\u534f\u8bae\u516c\u5f00\u53d1\u5e03\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/picx.zhimg.com\/v2-67f0431c0ddaf427777cb07cb2fd9e05_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic1.zhimg.com\/v2-0d42c5272832ef4631e1bdbd2d19b5bc_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_8\">CBR-LLM<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aCase-based Reasoning Augmented Large Language Model Framework for Decision Making in Realistic Safety-Critical Driving Scenarios<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2506.20531\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2506.20531<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1aNICT<\/li>\n<\/ul>\n\n\n\n<p>\u5728\u5b89\u5168\u5173\u952e\u573a\u666f\u4e2d\u9a7e\u9a76\u9700\u8981\u5feb\u901f\u3001\u5177\u6709\u60c5\u5883\u611f\u77e5\u7684\u51b3\u7b56\uff0c\u8fd9\u79cd\u51b3\u7b56\u65e2\u8981\u57fa\u4e8e\u60c5\u5883\u7406\u89e3\uff0c\u53c8\u8981\u4f9d\u6258\u7ecf\u9a8c\u63a8\u7406\u3002\u5927\u8bed\u8a00\u6a21\u578b\uff08LLMs\uff09\u51ed\u501f\u5176\u5f3a\u5927\u7684\u901a\u7528\u63a8\u7406\u80fd\u529b\uff0c\u4e3a\u8fd9\u7c7b\u51b3\u7b56\u63d0\u4f9b\u4e86\u9887\u5177\u524d\u666f\u7684\u57fa\u7840\u3002\u7136\u800c\uff0c\u7531\u4e8e\u5728\u9886\u57df\u9002\u914d\u3001\u60c5\u5883\u951a\u5b9a\u4ee5\u53ca\u7f3a\u4e4f\u5728\u52a8\u6001\u3001\u9ad8\u98ce\u9669\u73af\u5883\u4e2d\u505a\u51fa\u53ef\u9760\u4e14\u53ef\u89e3\u91ca\u51b3\u7b56\u6240\u9700\u7684\u7ecf\u9a8c\u77e5\u8bc6\u7b49\u65b9\u9762\u5b58\u5728\u6311\u6218\uff0c\u5b83\u4eec\u5728\u81ea\u52a8\u9a7e\u9a76\u4e2d\u7684\u76f4\u63a5\u5e94\u7528\u4ecd\u53d7\u5230\u9650\u5236\u3002\u4e3a\u586b\u8865\u8fd9\u4e00\u7a7a\u767d\uff0c\u672c\u6587\u63d0\u51fa\u4e86\u4e00\u79cd\u57fa\u4e8e\u6848\u4f8b\u63a8\u7406\u589e\u5f3a\u7684\u5927\u8bed\u8a00\u6a21\u578b\uff08CBR-LLM\uff09\u6846\u67b6\uff0c\u7528\u4e8e\u590d\u6742\u98ce\u9669\u573a\u666f\u4e2d\u7684\u89c4\u907f\u673a\u52a8\u51b3\u7b56\u3002\u6211\u4eec\u7684\u65b9\u6cd5\u5c06\u884c\u8f66\u8bb0\u5f55\u4eea\u89c6\u9891\u8f93\u5165\u4e2d\u7684\u8bed\u4e49\u573a\u666f\u7406\u89e3\u4e0e\u76f8\u5173\u8fc7\u5f80\u9a7e\u9a76\u6848\u4f8b\u7684\u68c0\u7d22\u76f8\u7ed3\u5408\uff0c\u4f7f LLMs \u80fd\u591f\u751f\u6210\u65e2\u7b26\u5408\u60c5\u5883\u53c8\u4e0e\u4eba\u7c7b\u884c\u4e3a\u4e00\u81f4\u7684\u673a\u52a8\u5efa\u8bae\u3002\u5728\u591a\u4e2a\u5f00\u6e90 LLMs \u4e0a\u8fdb\u884c\u7684\u5b9e\u9a8c\u8868\u660e\uff0c\u6211\u4eec\u7684\u6846\u67b6\u63d0\u9ad8\u4e86\u51b3\u7b56\u51c6\u786e\u6027\u3001\u63a8\u7406\u5408\u7406\u6027\u4ee5\u53ca\u4e0e\u4eba\u7c7b\u4e13\u5bb6\u884c\u4e3a\u7684\u4e00\u81f4\u6027\u3002\u98ce\u9669\u611f\u77e5\u63d0\u793a\u7b56\u7565\u8fdb\u4e00\u6b65\u63d0\u5347\u4e86\u5728\u4e0d\u540c\u98ce\u9669\u7c7b\u578b\u4e0b\u7684\u6027\u80fd\uff0c\u800c\u57fa\u4e8e\u76f8\u4f3c\u5ea6\u7684\u6848\u4f8b\u68c0\u7d22\u5728\u5f15\u5bfc\u4e0a\u4e0b\u6587\u5b66\u4e60\u65b9\u9762\u59cb\u7ec8\u4f18\u4e8e\u968f\u673a\u62bd\u6837\u3002\u6848\u4f8b\u7814\u7a76\u8fdb\u4e00\u6b65\u8bc1\u660e\u4e86\u8be5\u6846\u67b6\u5728\u5177\u6709\u6311\u6218\u6027\u7684\u73b0\u5b9e\u4e16\u754c\u6761\u4ef6\u4e0b\u7684\u7a33\u5065\u6027\uff0c\u7a81\u663e\u4e86\u5176\u4f5c\u4e3a\u667a\u80fd\u9a7e\u9a76\u7cfb\u7edf\u4e2d\u4e00\u79cd\u81ea\u9002\u5e94\u4e14\u53ef\u4fe1\u7684\u51b3\u7b56\u652f\u6301\u5de5\u5177\u7684\u6f5c\u529b\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/picx.zhimg.com\/v2-9b78f0f8472b2e6108c10a7a18b32c09_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/picx.zhimg.com\/v2-33968e1c6dd516c695d7f9fd46ec996b_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_9\">FutureSightDrive<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aFutureSightDrive: Thinking Visually with Spatio-Temporal CoT for Autonomous Driving<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2505.17685\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2505.17685<\/a><\/li>\n\n\n\n<li>\u9879\u76ee\u4e3b\u9875\uff1a<\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u963f\u91cc\u5df4\u5df4\u96c6\u56e2\u9ad8\u5fb7\uff0c\u897f\u5b89\u4ea4\u901a\u5927\u5b66\uff0c\u963f\u91cc\u5df4\u5df4\u96c6\u56e2\u8fbe\u6469\u9662<\/li>\n<\/ul>\n\n\n\n<p>\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08VLMs\uff09\u51ed\u501f\u5176\u5f3a\u5927\u7684\u63a8\u7406\u80fd\u529b\u5728\u81ea\u52a8\u9a7e\u9a76\u9886\u57df\u5f15\u8d77\u4e86\u8d8a\u6765\u8d8a\u591a\u7684\u5173\u6ce8\u3002\u7136\u800c\uff0c\u73b0\u6709\u7684\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u901a\u5e38\u91c7\u7528\u9488\u5bf9\u5f53\u524d\u573a\u666f\u7684\u79bb\u6563\u6587\u672c\u601d\u7ef4\u94fe\uff08CoT\uff09\uff0c\u8fd9\u672c\u8d28\u4e0a\u662f\u5bf9\u89c6\u89c9\u4fe1\u606f\u7684\u9ad8\u5ea6\u62bd\u8c61\u548c\u7b26\u53f7\u5316\u538b\u7f29\uff0c\u53ef\u80fd\u5bfc\u81f4\u65f6\u7a7a\u5173\u7cfb\u6a21\u7cca\u548c\u7ec6\u7c92\u5ea6\u4fe1\u606f\u4e22\u5931\u3002\u81ea\u52a8\u9a7e\u9a76\u7684\u5efa\u6a21\u662f\u5426\u66f4\u9002\u5408\u57fa\u4e8e\u771f\u5b9e\u4e16\u754c\u7684\u6a21\u62df\u4e0e\u60f3\u8c61\uff0c\u800c\u975e\u7eaf\u7cb9\u7684\u7b26\u53f7\u903b\u8f91\u3002\u672c\u6587\u63d0\u51fa\u4e00\u79cd\u65f6\u7a7a\u601d\u7ef4\u94fe\u63a8\u7406\u65b9\u6cd5\uff0c\u4f7f\u6a21\u578b\u80fd\u591f\u8fdb\u884c\u89c6\u89c9\u5316\u601d\u8003\u3002\u9996\u5148\uff0c\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u4f5c\u4e3a\u4e16\u754c\u6a21\u578b\u751f\u6210\u7edf\u4e00\u56fe\u50cf\u5e27\u4ee5\u9884\u6d4b\u672a\u6765\u4e16\u754c\u72b6\u6001\uff1a\u5176\u4e2d\u611f\u77e5\u7ed3\u679c\uff08\u5982\u8f66\u9053\u5206\u9694\u7ebf\u548c 3D \u68c0\u6d4b\u7ed3\u679c\uff09\u8868\u5f81\u672a\u6765\u7a7a\u95f4\u5173\u7cfb\uff0c\u666e\u901a\u672a\u6765\u5e27\u8868\u5f81\u65f6\u95f4\u6f14\u5316\u5173\u7cfb\u3002\u8fd9\u79cd\u65f6\u7a7a\u601d\u7ef4\u94fe\u968f\u540e\u4f5c\u4e3a\u4e2d\u95f4\u63a8\u7406\u6b65\u9aa4\uff0c\u4f7f\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u80fd\u591f\u4f5c\u4e3a\u9006\u52a8\u529b\u5b66\u6a21\u578b\uff0c\u57fa\u4e8e\u5f53\u524d\u89c2\u6d4b\u548c\u672a\u6765\u9884\u6d4b\u8fdb\u884c\u8f68\u8ff9\u89c4\u5212\u3002\u4e3a\u4e86\u5728\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u4e2d\u5b9e\u73b0\u89c6\u89c9\u751f\u6210\uff0c\u6211\u4eec\u63d0\u51fa\u4e00\u79cd\u878d\u5408\u89c6\u89c9\u751f\u6210\u4e0e\u7406\u89e3\u7684\u7edf\u4e00\u9884\u8bad\u7ec3\u8303\u5f0f\uff0c\u4ee5\u53ca\u4e00\u79cd\u589e\u5f3a\u81ea\u56de\u5f52\u56fe\u50cf\u751f\u6210\u7684\u6e10\u8fdb\u5f0f\u89c6\u89c9\u601d\u7ef4\u94fe\u3002\u5927\u91cf\u5b9e\u9a8c\u7ed3\u679c\u8bc1\u660e\u4e86\u6240\u63d0\u65b9\u6cd5\u7684\u6709\u6548\u6027\uff0c\u63a8\u52a8\u81ea\u52a8\u9a7e\u9a76\u5411\u89c6\u89c9\u63a8\u7406\u65b9\u5411\u53d1\u5c55\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic3.zhimg.com\/v2-15cd1c3a0fe383e536f8e17ed50f6b96_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic1.zhimg.com\/v2-403cde2cba5aa4921305d014d87f4b78_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_10\"><a href=\"https:\/\/zhida.zhihu.com\/search?content_id=261363339&amp;content_type=Article&amp;match_order=1&amp;q=OpenLKA-Alert&amp;zhida_source=entity\" target=\"_blank\" rel=\"noreferrer noopener\">OpenLKA-Alert<\/a><\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aBridging Human Oversight and Black-box Driver Assistance: Vision-Language Models for Predictive Alerting in Lane Keeping Assist systems<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2505.11535\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2505.11535<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u7f8e\u56fd\u5357\u4f5b\u7f57\u91cc\u8fbe\u5927\u5b66<\/li>\n<\/ul>\n\n\n\n<p>\u8f66\u9053\u4fdd\u6301\u8f85\u52a9\uff08LKA\uff09\u7cfb\u7edf\u867d\u7136\u65e5\u76ca\u666e\u53ca\uff0c\u4f46\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\u5e38\u51fa\u73b0\u4e0d\u53ef\u9884\u6d4b\u7684\u5931\u6548\u60c5\u51b5\uff0c\u8fd9\u5728\u5f88\u5927\u7a0b\u5ea6\u4e0a\u5f52\u56e0\u4e8e\u5176\u4e0d\u900f\u660e\u7684\u9ed1\u7bb1\u7279\u6027\uff0c\u9650\u5236\u4e86\u9a7e\u9a76\u5458\u7684\u9884\u5224\u80fd\u529b\u548c\u4fe1\u4efb\u5ea6\u3002\u4e3a\u5f25\u5408\u81ea\u52a8\u5316\u8f85\u52a9\u4e0e\u6709\u6548\u4eba\u7c7b\u76d1\u7763\u4e4b\u95f4\u7684\u5dee\u8ddd\uff0c\u6211\u4eec\u63d0\u51fa\u4e86 LKAlert\u2014\u2014 \u4e00\u79cd\u65b0\u578b\u76d1\u7763\u8b66\u62a5\u7cfb\u7edf\uff0c\u8be5\u7cfb\u7edf\u5229\u7528\u89c6\u89c9 &#8211; \u8bed\u8a00\u6a21\u578b\uff08VLM\uff09\u63d0\u524d 1-3 \u79d2\u9884\u6d4b\u6f5c\u5728\u7684 LKA \u98ce\u9669\u3002LKAlert \u5904\u7406\u884c\u8f66\u8bb0\u5f55\u4eea\u89c6\u9891\u548c CAN \u6570\u636e\uff0c\u6574\u5408\u6765\u81ea\u5e76\u884c\u53ef\u89e3\u91ca\u6a21\u578b\u7684\u66ff\u4ee3\u8f66\u9053\u5206\u5272\u7279\u5f81\u4f5c\u4e3a\u81ea\u52a8\u5316\u5f15\u5bfc\u6ce8\u610f\u529b\u3002\u4e0e\u4f20\u7edf\u7684\u4e8c\u5143\u5206\u7c7b\u5668\u4e0d\u540c\uff0cLKAlert \u4e0d\u4ec5\u53d1\u51fa\u9884\u6d4b\u6027\u8b66\u62a5\uff0c\u8fd8\u63d0\u4f9b\u7b80\u6d01\u7684\u81ea\u7136\u8bed\u8a00\u89e3\u91ca\uff0c\u4ece\u800c\u589e\u5f3a\u9a7e\u9a76\u5458\u7684\u60c5\u5883\u611f\u77e5\u80fd\u529b\u548c\u4fe1\u4efb\u5ea6\u3002\u4e3a\u652f\u6301\u6b64\u7c7b\u7cfb\u7edf\u7684\u5f00\u53d1\u548c\u8bc4\u4f30\uff0c\u6211\u4eec\u5f15\u5165\u4e86 OpenLKA-Alert\uff0c\u8fd9\u662f\u9996\u4e2a\u4e13\u4e3a\u9884\u6d4b\u6027\u548c\u53ef\u89e3\u91ca\u6027 LKA \u5931\u6548\u9884\u8b66\u8bbe\u8ba1\u7684\u57fa\u51c6\u6570\u636e\u96c6\u3002\u8be5\u6570\u636e\u96c6\u5305\u542b\u540c\u6b65\u7684\u591a\u6a21\u6001\u8f93\u5165\u4ee5\u53ca\u5728\u5e26\u6ce8\u91ca\u7684\u65f6\u95f4\u7a97\u53e3\u5185\u7531\u4eba\u5de5\u7f16\u5199\u7684\u89e3\u91ca\u3002\u6211\u4eec\u8fd8\u8d21\u732e\u4e86\u4e00\u4e2a\u53ef\u63a8\u5e7f\u7684\u57fa\u4e8e VLM \u7684\u9ed1\u7bb1\u884c\u4e3a\u9884\u6d4b\u65b9\u6cd5\u6846\u67b6\uff0c\u8be5\u6846\u67b6\u5c06\u66ff\u4ee3\u7279\u5f81\u5f15\u5bfc\u4e0e LoRA \u76f8\u7ed3\u5408\u3002\u6b64\u6846\u67b6\u4f7f VLM \u80fd\u591f\u5728\u4e0d\u6539\u53d8\u5176\u89c6\u89c9\u4e3b\u5e72\u7684\u60c5\u51b5\u4e0b\u5bf9\u7ed3\u6784\u5316\u89c6\u89c9\u4e0a\u4e0b\u6587\u8fdb\u884c\u63a8\u7406\uff0c\u4f7f\u5176\u5e7f\u6cdb\u9002\u7528\u4e8e\u5176\u4ed6\u9700\u8981\u53ef\u89e3\u91ca\u76d1\u7763\u7684\u590d\u6742\u3001\u4e0d\u900f\u660e\u7cfb\u7edf\u3002\u5b9e\u8bc1\u7ed3\u679c\u663e\u793a\uff0c\u8be5\u7cfb\u7edf\u5bf9\u5373\u5c06\u53d1\u751f\u7684 LKA \u5931\u6548\u7684\u9884\u6d4b\u51c6\u786e\u7387\u4e3a 69.8%\uff0cF1 \u5206\u6570\u4e3a 58.6%\u3002\u8be5\u7cfb\u7edf\u8fd8\u80fd\u4e3a\u9a7e\u9a76\u5458\u751f\u6210\u9ad8\u8d28\u91cf\u7684\u6587\u672c\u89e3\u91ca\uff08ROUGE-L \u5f97\u5206\u4e3a 71.7\uff09\uff0c\u4e14\u8fd0\u884c\u6548\u7387\u7ea6\u4e3a 2 Hz\uff0c\u8bc1\u5b9e\u5176\u9002\u7528\u4e8e\u5b9e\u65f6\u8f66\u8f7d\u573a\u666f\u3002\u6211\u4eec\u7684\u7814\u7a76\u7ed3\u679c\u8868\u660e\uff0cLKAlert \u662f\u63d0\u5347\u73b0\u6709\u5148\u8fdb\u9a7e\u9a76\u8f85\u52a9\u7cfb\u7edf\uff08ADAS\uff09\u5b89\u5168\u6027\u548c\u53ef\u7528\u6027\u7684\u5b9e\u7528\u89e3\u51b3\u65b9\u6848\uff0c\u5e76\u4e3a\u5c06 VLM \u5e94\u7528\u4e8e\u4ee5\u4eba\u7c7b\u4e3a\u4e2d\u5fc3\u7684\u9ed1\u7bb1\u81ea\u52a8\u5316\u76d1\u7763\u63d0\u4f9b\u4e86\u53ef\u6269\u5c55\u7684\u8303\u5f0f\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/picx.zhimg.com\/v2-3d4968db5318af456b7a6ca2e0c13547_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic1.zhimg.com\/v2-0e97605b8aa0a8f1ea33ca1540f6f318_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_11\">OpenLKA<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aOpenLKA: An Open Dataset of Lane Keeping Assist from Recent Car Models under Real-world Driving Conditions<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2505.09092\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2505.09092<\/a><\/li>\n\n\n\n<li>\u9879\u76ee\u4e3b\u9875\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/github.com\/OpenLKA\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/github.com\/OpenLKA<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u7f8e\u56fd\u5357\u4f5b\u7f57\u91cc\u8fbe\u5927\u5b66<\/li>\n<\/ul>\n\n\n\n<p>\u8f66\u9053\u4fdd\u6301\u8f85\u52a9\uff08LKA\uff09\u5728\u73b0\u4ee3\u91cf\u4ea7\u8f66\u8f86\u4e2d\u5f97\u5230\u5e7f\u6cdb\u5e94\u7528\uff0c\u4f46\u5176\u5b9e\u9645\u6027\u80fd\u4ecd\u4e0d\u900f\u660e\uff0c\u8fd9\u662f\u7531\u4e8e\u8be5\u7cfb\u7edf\u7684\u4e13\u6709\u63a7\u5236\u6808\u5bfc\u81f4\u7814\u7a76\u4eba\u5458\u65e0\u6cd5\u5bf9\u8fd9\u9879\u6280\u672f\u8fdb\u884c\u8bca\u65ad\u6216\u6539\u8fdb\u3002\u4e3a\u586b\u8865\u8fd9\u4e00\u7a7a\u767d\uff0c\u672c\u6587\u63d0\u51fa\u4e86 OpenLKA\uff0c\u8fd9\u662f\u9996\u4e2a\u7528\u4e8e LKA \u8bc4\u4f30\u548c\u6539\u8fdb\u7684\u5f00\u653e\u5f0f\u5927\u89c4\u6a21\u6570\u636e\u96c6\u3002\u8be5\u6570\u636e\u96c6\u5305\u542b\u6765\u81ea 62 \u6b3e\u91cf\u4ea7\u8f66\u578b\u7684 389.1 \u5c0f\u65f6 LKA \u63a7\u5236\u6570\u636e\uff0c\u8fd9\u4e9b\u6570\u636e\u901a\u8fc7\u5728\u4f5b\u7f57\u91cc\u8fbe\u5dde\u5766\u5e15\u5e02\u8fdb\u884c\u7684\u5e7f\u6cdb\u9053\u8def\u6d4b\u8bd5\u4ee5\u53ca\u5f00\u6e90\u793e\u533a\u7684\u5168\u7403\u8d21\u732e\u8005\u6536\u96c6\u800c\u6765\u3002\u6570\u636e\u96c6\u6db5\u76d6\u4e86\u591a\u79cd\u5177\u6709\u6311\u6218\u6027\u7684\u573a\u666f\uff0c\u5305\u62ec\u8f66\u9053\u6807\u8bb0\u9000\u5316\u3001\u590d\u6742\u9053\u8def\u51e0\u4f55\u5f62\u72b6\u3001\u6076\u52a3\u5929\u6c14\u3001\u5149\u7167\u6761\u4ef6\u4ee5\u53ca\u5404\u79cd\u5468\u8fb9\u4ea4\u901a\u72b6\u51b5\u3002\u8be5\u6570\u636e\u96c6\u662f\u591a\u6a21\u6001\u7684\uff0c\u5305\u62ec\uff1ai\uff09\u89e3\u7801\u7684\u8f66\u8f86\u5185\u90e8\u6d88\u606f\uff0c\u5176\u4e2d\u5305\u542b\u5173\u952e\u7684 LKA \u4fe1\u53f7\uff08\u5982\u7cfb\u7edf\u9000\u51fa\u3001\u8f66\u9053\u68c0\u6d4b\u5931\u8d25\u7b49\uff09\uff1bii\uff09\u6765\u81ea\u8f66\u8f7d dash \u6444\u50cf\u5934\u7684\u540c\u6b65\u9ad8\u5206\u8fa8\u7387\u89c6\u9891\uff1biii\uff09\u7531\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u751f\u6210\u7684\u4e30\u5bcc\u573a\u666f\u6807\u6ce8\uff0c\u6db5\u76d6\u8f66\u9053\u53ef\u89c1\u6027\u3001\u8def\u9762\u8d28\u91cf\u3001\u5929\u6c14\u3001\u5149\u7167\u548c\u4ea4\u901a\u72b6\u51b5\u7b49\u3002\u603b\u4f53\u800c\u8a00\uff0cOpenLKA \u63d0\u4f9b\u4e86\u4e00\u4e2a\u5168\u9762\u7684\u5e73\u53f0\uff0c\u7528\u4e8e\u57fa\u51c6\u6d4b\u8bd5\u91cf\u4ea7 LKA \u7cfb\u7edf\u7684\u5b9e\u9645\u6027\u80fd\u3001\u8bc6\u522b\u5b89\u5168\u5173\u952e\u64cd\u4f5c\u573a\u666f\u4ee5\u53ca\u8bc4\u4f30\u5f53\u524d\u9053\u8def\u57fa\u7840\u8bbe\u65bd\u5bf9\u81ea\u52a8\u9a7e\u9a76\u7684\u5c31\u7eea\u7a0b\u5ea6\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic1.zhimg.com\/v2-495232179e4d4777391dfb440aed4d3c_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic2.zhimg.com\/v2-6a7e2ff73212de16ff352debfeb192d9_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_12\"><a href=\"https:\/\/zhida.zhihu.com\/search?content_id=261363339&amp;content_type=Article&amp;match_order=1&amp;q=Vision+Foundation+Model&amp;zhida_source=entity\" target=\"_blank\" rel=\"noreferrer noopener\">Vision Foundation Model<\/a><\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aVision Foundation Model Embedding-Based Semantic Anomaly Detection<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2505.07998\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2505.07998<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u65af\u5766\u798f\u5927\u5b66\uff0cNVIDIA\u7b49<\/li>\n<\/ul>\n\n\n\n<p>ICRA 2025\u4e2d\u7a3f\u7684\u5de5\u4f5c\uff0c\u8bed\u4e49\u5f02\u5e38\u662f\u719f\u6089\u89c6\u89c9\u5143\u7d20\u5728\u8bed\u5883\u4e2d\u65e0\u6548\u6216\u4e0d\u5bfb\u5e38\u7684\u7ec4\u5408\uff0c\u53ef\u80fd\u5bfc\u81f4\u81ea\u4e3b\u7cfb\u7edf\u5728\u7cfb\u7edf\u7ea7\u63a8\u7406\u4e2d\u51fa\u73b0\u672a\u5b9a\u4e49\u884c\u4e3a\u548c\u6545\u969c\u3002\u672c\u7814\u7a76\u901a\u8fc7\u5229\u7528\u6700\u5148\u8fdb\u7684\u89c6\u89c9\u57fa\u7840\u6a21\u578b\u7684\u8bed\u4e49\u5148\u9a8c\u77e5\u8bc6\uff0c\u76f4\u63a5\u5bf9\u56fe\u50cf\u8fdb\u884c\u5904\u7406\uff0c\u63a2\u7d22\u8bed\u4e49\u5f02\u5e38\u68c0\u6d4b\u3002\u6211\u4eec\u63d0\u51fa\u4e86\u4e00\u4e2a\u6846\u67b6\uff0c\u5c06\u8fd0\u884c\u65f6\u56fe\u50cf\u7684\u5c40\u90e8\u89c6\u89c9\u5d4c\u5165\u4e0e\u81ea\u4e3b\u7cfb\u7edf\u88ab\u8ba4\u4e3a\u5b89\u5168\u4e14\u6027\u80fd\u826f\u597d\u7684\u6b63\u5e38\u573a\u666f\u6570\u636e\u5e93\u8fdb\u884c\u6bd4\u8f83\u3002\u5728\u672c\u7814\u7a76\u4e2d\uff0c\u6211\u4eec\u8003\u8651\u4e86\u6240\u63d0\u6846\u67b6\u7684\u4e24\u79cd\u53d8\u4f53\uff1a\u4e00\u79cd\u4f7f\u7528\u539f\u59cb\u7684\u57fa\u4e8e\u7f51\u683c\u7684\u5d4c\u5165\uff0c\u53e6\u4e00\u79cd\u5229\u7528\u5b9e\u4f8b\u5206\u5272\u83b7\u53d6\u4ee5\u5bf9\u8c61\u4e3a\u4e2d\u5fc3\u7684\u8868\u793a\u3002\u4e3a\u8fdb\u4e00\u6b65\u63d0\u9ad8\u9c81\u68d2\u6027\uff0c\u6211\u4eec\u5f15\u5165\u4e86\u4e00\u79cd\u7b80\u5355\u7684\u8fc7\u6ee4\u673a\u5236\u6765\u6291\u5236\u8bef\u62a5\u3002\u6211\u4eec\u5728 CARLA \u6a21\u62df\u5f02\u5e38\u4e0a\u7684\u8bc4\u4f30\u8868\u660e\uff0c\u7ed3\u5408\u8fc7\u6ee4\u7684\u57fa\u4e8e\u5b9e\u4f8b\u7684\u65b9\u6cd5\u6027\u80fd\u53ef\u4e0e GPT-4o \u5ab2\u7f8e\uff0c\u540c\u65f6\u80fd\u63d0\u4f9b\u7cbe\u786e\u7684\u5f02\u5e38\u5b9a\u4f4d\u3002\u8fd9\u4e9b\u7ed3\u679c\u51f8\u663e\u4e86\u57fa\u7840\u6a21\u578b\u7684\u89c6\u89c9\u5d4c\u5165\u5728\u81ea\u4e3b\u7cfb\u7edf\u5b9e\u65f6\u5f02\u5e38\u68c0\u6d4b\u4e2d\u7684\u6f5c\u5728\u6548\u7528\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pica.zhimg.com\/v2-796af9e97a894e542bd106d5e77366dc_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic2.zhimg.com\/v2-e1c81d5f60cc8874e2a53b0cc3096865_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_13\">ORION<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aORION: A Holistic End-to-End Autonomous Driving Framework by Vision-Language Instructed Action Generation<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/www.arxiv.org\/abs\/2503.19755\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/www.arxiv.org\/abs\/2503.19755<\/a><\/li>\n\n\n\n<li>\u9879\u76ee\u4e3b\u9875\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/xiaomi-mlab.github.io\/Orion\/\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/xiaomi-mlab.github.io\/Orion\/<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u534e\u4e2d\u79d1\u6280\u5927\u5b66\uff0c\u5c0f\u7c73\u6c7d\u8f66<\/li>\n<\/ul>\n\n\n\n<p>\u7aef\u5230\u7aef\uff08E2E\uff09\u81ea\u52a8\u9a7e\u9a76\u65b9\u6cd5\u7531\u4e8e\u56e0\u679c\u63a8\u7406\u80fd\u529b\u6709\u9650\uff0c\u5728\u4ea4\u4e92\u5f0f\u95ed\u73af\u8bc4\u4f30\u4e2d\u4ecd\u96be\u4ee5\u505a\u51fa\u6b63\u786e\u51b3\u7b56\u3002\u73b0\u6709\u65b9\u6cd5\u5c1d\u8bd5\u5229\u7528\u89c6\u89c9 &#8211; \u8bed\u8a00\u6a21\u578b\uff08VLMs\uff09\u5f3a\u5927\u7684\u7406\u89e3\u4e0e\u63a8\u7406\u80fd\u529b\u6765\u89e3\u51b3\u8fd9\u4e00\u56f0\u5883\u3002\u7136\u800c\uff0c\u7531\u4e8e\u8bed\u4e49\u63a8\u7406\u7a7a\u95f4\u4e0e\u52a8\u4f5c\u7a7a\u95f4\u4e2d\u7eaf\u6570\u503c\u8f68\u8ff9\u8f93\u51fa\u4e4b\u95f4\u5b58\u5728\u5dee\u8ddd\uff0c\u5f88\u5c11\u6709\u9002\u7528\u4e8e\u7aef\u5230\u7aef\u65b9\u6cd5\u7684 VLMs \u80fd\u5728\u95ed\u73af\u8bc4\u4f30\u4e2d\u8868\u73b0\u826f\u597d\uff0c\u8fd9\u4e00\u95ee\u9898\u4ecd\u672a\u5f97\u5230\u89e3\u51b3\u3002\u4e3a\u89e3\u51b3\u8be5\u95ee\u9898\uff0c\u6211\u4eec\u63d0\u51fa ORION\u2014\u2014 \u4e00\u79cd\u57fa\u4e8e\u89c6\u89c9 &#8211; \u8bed\u8a00\u6307\u4ee4\u52a8\u4f5c\u751f\u6210\u7684\u6574\u4f53\u7aef\u5230\u7aef\u81ea\u52a8\u9a7e\u9a76\u6846\u67b6\u3002ORION \u72ec\u7279\u5730\u878d\u5408\u4e86\u7528\u4e8e\u805a\u5408\u957f\u671f\u5386\u53f2\u4e0a\u4e0b\u6587\u7684 QT-Former\u3001\u7528\u4e8e\u9a7e\u9a76\u573a\u666f\u63a8\u7406\u7684\u5927\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u4ee5\u53ca\u7528\u4e8e\u7cbe\u786e\u8f68\u8ff9\u9884\u6d4b\u7684\u751f\u6210\u5f0f\u89c4\u5212\u5668\u3002\u6b64\u5916\uff0cORION \u901a\u8fc7\u5bf9\u9f50\u63a8\u7406\u7a7a\u95f4\u4e0e\u52a8\u4f5c\u7a7a\u95f4\uff0c\u5b9e\u73b0\u4e86\u89c6\u89c9\u95ee\u7b54\uff08VQA\uff09\u548c\u89c4\u5212\u4efb\u52a1\u7684\u7edf\u4e00\u7aef\u5230\u7aef\u4f18\u5316\u3002\u6211\u4eec\u7684\u65b9\u6cd5\u5728 Bench2Drive \u6311\u6218\u6570\u636e\u96c6\u4e0a\u53d6\u5f97\u4e86\u4ee4\u4eba\u77a9\u76ee\u7684\u95ed\u73af\u6027\u80fd\uff0c\u9a7e\u9a76\u5f97\u5206\uff08DS\uff09\u8fbe 77.74\uff0c\u6210\u529f\u7387\uff08SR\uff09\u8fbe 54.62%\uff0c\u5927\u5e45\u8d85\u8d8a\u4e86\u6700\u5148\u8fdb\uff08SOTA\uff09\u65b9\u6cd5\uff0c\u9886\u5148 14.28 \u4e2a\u9a7e\u9a76\u5f97\u5206\u548c 19.61% \u7684\u6210\u529f\u7387\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic3.zhimg.com\/v2-71b259b067c1f3861a6e4a1943fee124_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pica.zhimg.com\/v2-c38c675b1e2171de0ee7d52f22eb9524_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_14\">DriveLMM-o1<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aDriveLMM-o1: A Step-by-Step Reasoning Dataset and Large Multimodal Model for Driving Scenario Understanding<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2503.10621\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2503.10621<\/a><\/li>\n\n\n\n<li>\u9879\u76ee\u4e3b\u9875\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/github.com\/ayesha-ishaq\/DriveLMM-o1\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/github.com\/ayesha-ishaq\/DriveLMM-o1<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u7a46\u7f55\u9ed8\u5fb7\u30fb\u672c\u30fb\u624e\u8036\u5fb7\u4eba\u5de5\u667a\u80fd\u5927\u5b66\uff0c\u6fb3\u5927\u5229\u4e9a\u56fd\u7acb\u5927\u5b66<\/li>\n<\/ul>\n\n\n\n<p>\u5c3d\u7ba1\u5927\u578b\u591a\u6a21\u6001\u6a21\u578b\uff08LMMs\uff09\u5728\u5404\u79cd\u89c6\u89c9\u95ee\u7b54\uff08VQA\uff09\u4efb\u52a1\u4e2d\u8868\u73b0\u51fa\u4e86\u8f83\u5f3a\u7684\u6027\u80fd\uff0c\u4f46\u67d0\u4e9b\u6311\u6218\u9700\u8981\u590d\u6742\u7684\u591a\u6b65\u9aa4\u63a8\u7406\u624d\u80fd\u5f97\u51fa\u51c6\u786e\u7b54\u6848\u3002\u81ea\u52a8\u9a7e\u9a76\u5c31\u662f\u4e00\u9879\u6781\u5177\u6311\u6218\u6027\u7684\u4efb\u52a1\uff0c\u5728\u505a\u51fa\u51b3\u7b56\u4e4b\u524d\u9700\u8981\u8fdb\u884c\u5168\u9762\u7684\u8ba4\u77e5\u5904\u7406\u3002\u5728\u8fd9\u4e00\u9886\u57df\uff0c\u5bf9\u89c6\u89c9\u7ebf\u7d22\u7684\u987a\u5e8f\u6027\u548c\u89e3\u91ca\u6027\u7406\u89e3\u5bf9\u4e8e\u6709\u6548\u7684\u611f\u77e5\u3001\u9884\u6d4b\u548c\u89c4\u5212\u81f3\u5173\u91cd\u8981\u3002\u7136\u800c\uff0c\u5e38\u89c1\u7684 VQA \u57fa\u51c6\u5f80\u5f80\u4fa7\u91cd\u4e8e\u6700\u7ec8\u7b54\u6848\u7684\u51c6\u786e\u6027\uff0c\u5374\u5ffd\u89c6\u4e86\u751f\u6210\u51c6\u786e\u54cd\u5e94\u6240\u4f9d\u8d56\u7684\u63a8\u7406\u8fc7\u7a0b\u3002\u6b64\u5916\uff0c\u73b0\u6709\u65b9\u6cd5\u7f3a\u4e4f\u4e00\u4e2a\u5168\u9762\u7684\u6846\u67b6\u6765\u8bc4\u4f30\u771f\u5b9e\u9a7e\u9a76\u573a\u666f\u4e2d\u7684\u9010\u6b65\u63a8\u7406\u80fd\u529b\u3002\u4e3a\u4e86\u586b\u8865\u8fd9\u4e00\u7a7a\u767d\uff0c\u6211\u4eec\u63d0\u51fa\u4e86 DriveLMM-o1\uff0c\u8fd9\u662f\u4e00\u4e2a\u4e13\u95e8\u4e3a\u63a8\u8fdb\u81ea\u52a8\u9a7e\u9a76\u9886\u57df\u7684\u9010\u6b65\u89c6\u89c9\u63a8\u7406\u800c\u8bbe\u8ba1\u7684\u65b0\u6570\u636e\u96c6\u548c\u57fa\u51c6\u3002\u6211\u4eec\u7684\u57fa\u51c6\u5728\u8bad\u7ec3\u96c6\u4e2d\u5305\u542b\u8d85\u8fc7 18k \u4e2a VQA \u793a\u4f8b\uff0c\u5728\u6d4b\u8bd5\u96c6\u4e2d\u5305\u542b\u8d85\u8fc7 4k \u4e2a\uff0c\u6db5\u76d6\u4e86\u5173\u4e8e\u611f\u77e5\u3001\u9884\u6d4b\u548c\u89c4\u5212\u7684\u5404\u79cd\u95ee\u9898\uff0c\u6bcf\u4e2a\u95ee\u9898\u90fd\u914d\u6709\u9010\u6b65\u63a8\u7406\u8fc7\u7a0b\uff0c\u4ee5\u786e\u4fdd\u5728\u81ea\u52a8\u9a7e\u9a76\u573a\u666f\u4e2d\u7684\u903b\u8f91\u63a8\u7406\u3002\u6211\u4eec\u8fdb\u4e00\u6b65\u4ecb\u7ecd\u4e86\u4e00\u4e2a\u5728\u6211\u4eec\u7684\u63a8\u7406\u6570\u636e\u96c6\u4e0a\u8fdb\u884c\u5fae\u8c03\u7684\u5927\u578b\u591a\u6a21\u6001\u6a21\u578b\uff0c\u8be5\u6a21\u578b\u5728\u590d\u6742\u9a7e\u9a76\u573a\u666f\u4e2d\u8868\u73b0\u51fa\u7a33\u5065\u7684\u6027\u80fd\u3002\u6b64\u5916\uff0c\u6211\u4eec\u5728\u63d0\u51fa\u7684\u6570\u636e\u96c6\u4e0a\u5bf9\u5404\u79cd\u5f00\u6e90\u548c\u95ed\u6e90\u65b9\u6cd5\u8fdb\u884c\u4e86\u57fa\u51c6\u6d4b\u8bd5\uff0c\u7cfb\u7edf\u5730\u6bd4\u8f83\u4e86\u5b83\u4eec\u5728\u81ea\u52a8\u9a7e\u9a76\u4efb\u52a1\u4e2d\u7684\u63a8\u7406\u80fd\u529b\u3002\u6211\u4eec\u7684\u6a21\u578b\u76f8\u8f83\u4e8e\u4e4b\u524d\u6700\u4f73\u7684\u5f00\u6e90\u6a21\u578b\uff0c\u5728\u6700\u7ec8\u7b54\u6848\u51c6\u786e\u6027\u4e0a\u63d0\u5347\u4e86 7.49%\uff0c\u5728\u63a8\u7406\u5f97\u5206\u4e0a\u63d0\u5347\u4e86 3.62%\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic1.zhimg.com\/v2-3d7794c9912fffd158c5cd0f886d931a_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic2.zhimg.com\/v2-0ce3eb28bd15addca1b2676f680e9053_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_15\">AutoDrive-QA<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aAutoDrive-QA &#8211; Automated Generation of Multiple-Choice Questions for Autonomous Driving Datasets Using Large Vision-Language Models<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2503.15778\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2503.15778<\/a><\/li>\n\n\n\n<li>\u9879\u76ee\u4e3b\u9875\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/github.com\/Boshrakh\/AutoDrive-QA\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/github.com\/Boshrakh\/AutoDrive-QA<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u54e5\u4f26\u6bd4\u4e9a\u5927\u5b66<\/li>\n<\/ul>\n\n\n\n<p>\u5728\u81ea\u52a8\u9a7e\u9a76\u9886\u57df\uff0c\u5f00\u653e\u5f0f\u95ee\u7b54\u5f80\u5f80\u56e0\u81ea\u7531\u5f62\u5f0f\u7684\u56de\u7b54\u9700\u8981\u590d\u6742\u7684\u8bc4\u4f30\u6307\u6807\u6216\u4e3b\u89c2\u7684\u4eba\u5de5\u5224\u65ad\u800c\u9762\u4e34\u8bc4\u4f30\u4e0d\u53ef\u9760\u7684\u95ee\u9898\u3002\u4e3a\u5e94\u5bf9\u8fd9\u4e00\u6311\u6218\uff0c\u6211\u4eec\u63d0\u51fa\u4e86 AutoDrive-QA \u2014\u2014 \u4e00\u4e2a\u81ea\u52a8\u6d41\u6c34\u7ebf\uff0c\u53ef\u5c06\u73b0\u6709\u7684\u9a7e\u9a76\u95ee\u7b54\u6570\u636e\u96c6\uff08\u5305\u62ec DriveLM\u3001NuScenes-QA \u548c LingoQA\uff09\u8f6c\u6362\u4e3a\u7ed3\u6784\u5316\u7684\u591a\u9879\u9009\u62e9\u9898\uff08MCQ\uff09\u683c\u5f0f\u3002\u8be5\u57fa\u51c6\u7cfb\u7edf\u5730\u8bc4\u4f30\u611f\u77e5\u3001\u9884\u6d4b\u548c\u89c4\u5212\u4efb\u52a1\uff0c\u63d0\u4f9b\u4e86\u4e00\u4e2a\u6807\u51c6\u5316\u4e14\u5ba2\u89c2\u7684\u8bc4\u4f30\u6846\u67b6\u3002AutoDrive-QA \u91c7\u7528\u81ea\u52a8\u6d41\u6c34\u7ebf\uff0c\u5229\u7528\u5927\u8bed\u8a00\u6a21\u578b\uff08LLMs\uff09\uff0c\u57fa\u4e8e\u81ea\u52a8\u9a7e\u9a76\u573a\u666f\u4e2d\u5e38\u89c1\u7684\u7279\u5b9a\u9886\u57df\u9519\u8bef\u6a21\u5f0f\u751f\u6210\u9ad8\u8d28\u91cf\u3001\u4e0e\u4e0a\u4e0b\u6587\u76f8\u5173\u7684\u5e72\u6270\u9879\u3002\u4e3a\u8bc4\u4f30\u6a21\u578b\u7684\u901a\u7528\u80fd\u529b\u548c\u6cdb\u5316\u6027\u80fd\uff0c\u6211\u4eec\u5728\u4e09\u4e2a\u516c\u5f00\u6570\u636e\u96c6\u4e0a\u6d4b\u8bd5\u4e86\u8be5\u57fa\u51c6\uff0c\u5e76\u5728\u4e00\u4e2a\u672a\u89c1\u8fc7\u7684\u6570\u636e\u96c6\u4e0a\u8fdb\u884c\u4e86\u96f6\u6837\u672c\u5b9e\u9a8c\u3002\u96f6\u6837\u672c\u8bc4\u4f30\u663e\u793a\uff0cGPT-4V \u4ee5 69.57% \u7684\u51c6\u786e\u7387\u9886\u5148 \u2014\u2014 \u5728\u611f\u77e5\u4efb\u52a1\u4e2d\u8fbe\u5230 74.94%\uff0c\u9884\u6d4b\u4efb\u52a1\u4e2d\u8fbe\u5230 65.33%\uff0c\u89c4\u5212\u4efb\u52a1\u4e2d\u8fbe\u5230 68.45%\uff0c\u8fd9\u8868\u660e\u6240\u6709\u6a21\u578b\u5728\u611f\u77e5\u4efb\u52a1\u4e2d\u8868\u73b0\u51fa\u8272\uff0c\u4f46\u5728\u9884\u6d4b\u4efb\u52a1\u4e2d\u5b58\u5728\u56f0\u96be\u3002\u56e0\u6b64\uff0cAutoDrive-QA \u5efa\u7acb\u4e86\u4e00\u4e2a\u4e25\u683c\u3001\u65e0\u504f\u7684\u6807\u51c6\uff0c\u7528\u4e8e\u6574\u5408\u548c\u8bc4\u4f30\u4e0d\u540c\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u5728\u5404\u79cd\u81ea\u52a8\u9a7e\u9a76\u6570\u636e\u96c6\u4e0a\u7684\u8868\u73b0\uff0c\u4ece\u800c\u63d0\u9ad8\u8be5\u9886\u57df\u7684\u6cdb\u5316\u80fd\u529b\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic4.zhimg.com\/v2-6635ee604acc41174122b55160d0a61d_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic4.zhimg.com\/v2-13f25d7b21065fe48d29d3742d930639_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_16\">NuGrounding<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aNuGrounding: A Multi-View 3D Visual Grounding Framework in Autonomous Driving<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2503.22436\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2503.22436<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u6e05\u534e\u5927\u5b66\uff0c\u534e\u4e3a\u8bfa\u4e9a\u65b9\u821f\u5b9e\u9a8c\u5ba4\uff0c\u9999\u6e2f\u5927\u5b66<\/li>\n<\/ul>\n\n\n\n<p>\u591a\u89c6\u56fe 3D \u89c6\u89c9\u5b9a\u4f4d\u5bf9\u4e8e\u81ea\u52a8\u9a7e\u9a76\u8f66\u8f86\u5728\u590d\u6742\u73af\u5883\u4e2d\u7406\u89e3\u81ea\u7136\u8bed\u8a00\u5e76\u5b9a\u4f4d\u76ee\u6807\u7269\u4f53\u81f3\u5173\u91cd\u8981\u3002\u7136\u800c\uff0c\u73b0\u6709\u6570\u636e\u96c6\u548c\u65b9\u6cd5\u5b58\u5728\u8bed\u8a00\u6307\u4ee4\u7c92\u5ea6\u7c97\u7cd9\u30013D \u51e0\u4f55\u63a8\u7406\u4e0e\u8bed\u8a00\u7406\u89e3\u878d\u5408\u4e0d\u8db3\u7b49\u95ee\u9898\u3002\u4e3a\u6b64\uff0c\u6211\u4eec\u63d0\u51fa\u4e86 NuGrounding\uff0c\u8fd9\u662f\u9996\u4e2a\u9762\u5411\u81ea\u52a8\u9a7e\u9a76\u591a\u89c6\u56fe 3D \u89c6\u89c9\u5b9a\u4f4d\u7684\u5927\u89c4\u6a21\u57fa\u51c6\u6570\u636e\u96c6\u3002\u6211\u4eec\u8bbe\u8ba1\u4e86\u5b9a\u4f4d\u5c42\u6b21\u7ed3\u6784\uff08HoG\uff09\u65b9\u6cd5\u6765\u6784\u5efa\u8be5\u6570\u636e\u96c6\uff0c\u751f\u6210\u5177\u6709\u5c42\u6b21\u7ed3\u6784\u7684\u591a\u7ea7\u6307\u4ee4\uff0c\u786e\u4fdd\u5168\u9762\u8986\u76d6\u4eba\u7c7b\u6307\u4ee4\u6a21\u5f0f\u3002\u4e3a\u5e94\u5bf9\u8fd9\u4e00\u5177\u6709\u6311\u6218\u6027\u7684\u6570\u636e\u96c6\uff0c\u6211\u4eec\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u8303\u5f0f\uff0c\u5c06\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b\uff08MLLMs\uff09\u7684\u6307\u4ee4\u7406\u89e3\u80fd\u529b\u4e0e\u4e13\u4e1a\u68c0\u6d4b\u6a21\u578b\u7684\u7cbe\u786e\u5b9a\u4f4d\u80fd\u529b\u65e0\u7f1d\u7ed3\u5408\u3002\u6211\u4eec\u7684\u65b9\u6cd5\u5f15\u5165\u4e86\u4e24\u4e2a\u89e3\u8026\u7684\u4efb\u52a1\u4ee4\u724c\u548c\u4e00\u4e2a\u4e0a\u4e0b\u6587\u67e5\u8be2\uff0c\u7528\u4e8e\u805a\u5408 3D \u51e0\u4f55\u4fe1\u606f\u548c\u8bed\u4e49\u6307\u4ee4\uff0c\u968f\u540e\u901a\u8fc7\u878d\u5408\u89e3\u7801\u5668\u4f18\u5316\u7a7a\u95f4 &#8211; \u8bed\u4e49\u7279\u5f81\u878d\u5408\u4ee5\u5b9e\u73b0\u7cbe\u786e\u5b9a\u4f4d\u3002\u5927\u91cf\u5b9e\u9a8c\u8868\u660e\uff0c\u6211\u4eec\u7684\u65b9\u6cd5\u663e\u8457\u4f18\u4e8e\u4ece\u4ee3\u8868\u6027 3D \u573a\u666f\u7406\u89e3\u65b9\u6cd5\u6539\u7f16\u800c\u6765\u7684\u57fa\u7ebf\u65b9\u6cd5\uff0c\u7cbe\u786e\u7387\u8fbe\u5230 0.59\uff0c\u53ec\u56de\u7387\u8fbe\u5230 0.64\uff0c\u5206\u522b\u63d0\u5347\u4e86 50.8% \u548c 54.7%\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/picx.zhimg.com\/v2-2d06b3a64f54b673118091e472ce3d1d_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic2.zhimg.com\/v2-3c346608477d7292802d63ed02ea32a9_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_17\">ViLA<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aEvaluating Multimodal Vision-Language Model Prompting Strategies for Visual Question Answering in Road Scene Understanding<\/li>\n\n\n\n<li>HTML\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/openaccess.thecvf.com\/content\/WACV2025W\/LLVMAD\/html\/Keskar_Evaluating_Multimodal_Vision-Language_Model_Prompting_Strategies_for_Visual_Question_Answering_WACVW_2025_paper.html\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/openaccess.thecvf.com\/content\/WACV2025W\/LLVMAD\/html\/Keskar_Evaluating_Multimodal_Vision-Language_Model_Prompting_Strategies_for_Visual_Question_Answering_WACVW_2025_paper.html<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u52a0\u5dde\u5927\u5b66\u9ed8\u585e\u5fb7\u5206\u6821<\/li>\n<\/ul>\n\n\n\n<p>\u7406\u89e3\u590d\u6742\u4ea4\u901a\u573a\u666f\u662f\u63a8\u52a8\u81ea\u52a8\u9a7e\u9a76\u7cfb\u7edf\u53d1\u5c55\u7684\u5173\u952e\u6311\u6218\u3002\u89c6\u89c9\u95ee\u7b54\uff08VQA\uff09\u4efb\u52a1\u4f5c\u4e3a\u4e00\u79cd\u65b0\u5174\u65b9\u6cd5\uff0c\u80fd\u591f\u4ece\u591a\u6a21\u6001\u4ea4\u901a\u6570\u636e\u4e2d\u63d0\u53d6\u53ef\u64cd\u4f5c\u4fe1\u606f\uff0c\u4f7f\u8f66\u8f86\u505a\u51fa\u7cbe\u51c6\u7684\u5b9e\u65f6\u51b3\u7b56\u3002\u4f5c\u4e3a2025\u5e74WACV\u81ea\u52a8\u9a7e\u9a76\u5927\u8bed\u8a00\u89c6\u89c9\u6a21\u578b\u6311\u6218\u8d5b\uff08LLVM-AD\uff09\u7684\u91cd\u8981\u7ec4\u6210\u90e8\u5206\uff0cMAPLM-QA\u6570\u636e\u96c6\u4e3a\u6b64\u4efb\u52a1\u63d0\u4f9b\u4e86\u5f3a\u6709\u529b\u7684\u57fa\u51c6\u3002\u8be5\u6570\u636e\u96c6\u5305\u542b14,000\u4e2a\u591a\u6a21\u6001\u5e27\uff0c\u878d\u5408\u4e86\u9ad8\u5206\u8fa8\u7387\u5168\u666f\u56fe\u50cf\u4e0eLiDAR\u4e09\u7ef4\u70b9\u4e91\u6e32\u67d3\u751f\u6210\u7684\u9e1f\u77b0\u56fe\uff08BEV\uff09\u3002\u672c\u7814\u7a76\u63a2\u7d22\u4e86\u5e94\u7528\u82f1\u4f1f\u8fbe\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08ViLA\uff09\u89e3\u51b3MAPLM-QA\u4e2dVQA\u4efb\u52a1\u7684\u65b9\u6cd5\u3002\u901a\u8fc7\u9488\u5bf9\u6570\u636e\u96c6\u8bbe\u8ba1\u7cbe\u7ec6\u63d0\u793a\u5de5\u7a0b\uff0c\u6211\u4eec\u7cfb\u7edf\u8bc4\u4f30\u4e86ViLA\u7684\u6027\u80fd\uff1a\u5728\u8d28\u91cf\u8bc4\u4f30\u7b49\u6307\u6807\u4e0a\u5c55\u73b0\u51fa\u4f18\u52bf\uff0c\u540c\u65f6\u63ed\u793a\u4e86\u5176\u5728\u8f66\u9053\u8ba1\u6570\u3001\u4ea4\u53c9\u8def\u53e3\u8bc6\u522b\u53ca\u573a\u666f\u7ec6\u5fae\u5dee\u5f02\u7406\u89e3\u65b9\u9762\u7684\u6311\u6218\u3002\u7814\u7a76\u7ed3\u679c\u8bc1\u5b9e\u4e86\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08VLM\uff09\u5728\u589e\u5f3a\u4ea4\u901a\u573a\u666f\u5206\u6790\u4e0e\u81ea\u52a8\u9a7e\u9a76\u80fd\u529b\u65b9\u9762\u7684\u6f5c\u529b\uff0c\u4e3a\u672a\u6765\u5229\u7528VLM\u548c\u591a\u6a21\u6001\u6570\u636e\u96c6\u5b9e\u73b0\u53ef\u6269\u5c55\u3001\u9c81\u68d2\u7684\u4ea4\u901a\u573a\u666f\u7406\u89e3\u5960\u5b9a\u4e86\u575a\u5b9e\u57fa\u7840\uff0c\u5e76\u660e\u786e\u4e86\u6280\u672f\u5c40\u9650\u6027\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic2.zhimg.com\/v2-29b026c650f81e97657c2a7ee55aa8db_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pica.zhimg.com\/v2-2e7a29248b6794daf66df34d4c927bda_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_18\">INSIGHT<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aINSIGHT: Enhancing Autonomous Driving Safety through Vision-Language Models on Context-Aware Hazard Detection and Edge Case Evaluation<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/www.arxiv.org\/abs\/2502.00262\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/www.arxiv.org\/abs\/2502.00262<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u7f8e\u56fd\u9a6c\u91cc\u5170\u5927\u5b66\u5e15\u514b\u5206\u6821\uff0c\u7f8e\u56fd\u5317\u5361\u7f57\u6765\u7eb3\u5dde\u7acb\u5927\u5b66<\/li>\n<\/ul>\n\n\n\n<p>\u81ea\u52a8\u9a7e\u9a76\u7cfb\u7edf\u5728\u5904\u7406\u4e0d\u53ef\u9884\u6d4b\u7684\u8fb9\u7f18\u6848\u4f8b\u573a\u666f\u65f6\u9762\u4e34\u91cd\u5927\u6311\u6218\uff0c\u4f8b\u5982\u5177\u6709\u5bf9\u6297\u6027\u7684\u884c\u4eba\u8fd0\u52a8\u3001\u5371\u9669\u7684\u8f66\u8f86\u64cd\u4f5c\u4ee5\u53ca\u7a81\u53d1\u7684\u73af\u5883\u53d8\u5316\u3002\u5f53\u524d\u7684\u7aef\u5230\u7aef\u9a7e\u9a76\u6a21\u578b\u7531\u4e8e\u4f20\u7edf\u68c0\u6d4b\u548c\u9884\u6d4b\u65b9\u6cd5\u7684\u5c40\u9650\u6027\uff0c\u5728\u6cdb\u5316\u5230\u8fd9\u4e9b\u7f55\u89c1\u4e8b\u4ef6\u65f6\u8868\u73b0\u4e0d\u4f73\u3002\u4e3a\u89e3\u51b3\u8fd9\u4e00\u95ee\u9898\uff0c\u6211\u4eec\u63d0\u51fa\u4e86 INSIGHT\uff08\u7528\u4e8e\u5e7f\u4e49\u5371\u9669\u8ddf\u8e2a\u7684\u8bed\u4e49\u4e0e\u89c6\u89c9\u8f93\u5165\u96c6\u6210\uff09\uff0c\u8fd9\u662f\u4e00\u79cd\u5206\u5c42\u89c6\u89c9 &#8211; \u8bed\u8a00\u6a21\u578b\uff08VLM\uff09\u6846\u67b6\uff0c\u65e8\u5728\u589e\u5f3a\u5371\u9669\u68c0\u6d4b\u548c\u8fb9\u7f18\u6848\u4f8b\u8bc4\u4f30\u80fd\u529b\u3002\u901a\u8fc7\u591a\u6a21\u6001\u6570\u636e\u878d\u5408\uff0c\u6211\u4eec\u7684\u65b9\u6cd5\u6574\u5408\u4e86\u8bed\u4e49\u548c\u89c6\u89c9\u8868\u5f81\uff0c\u80fd\u591f\u7cbe\u786e\u89e3\u8bfb\u9a7e\u9a76\u573a\u666f\u5e76\u51c6\u786e\u9884\u6d4b\u6f5c\u5728\u5371\u9669\u3002\u901a\u8fc7\u5bf9\u89c6\u89c9 &#8211; \u8bed\u8a00\u6a21\u578b\u8fdb\u884c\u76d1\u7763\u5fae\u8c03\uff0c\u6211\u4eec\u5229\u7528\u57fa\u4e8e\u6ce8\u610f\u529b\u7684\u673a\u5236\u548c\u5750\u6807\u56de\u5f52\u6280\u672f\u4f18\u5316\u4e86\u7a7a\u95f4\u5371\u9669\u5b9a\u4f4d\u3002\u5728 BDD100K \u6570\u636e\u96c6\u4e0a\u7684\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0c\u4e0e\u73b0\u6709\u6a21\u578b\u76f8\u6bd4\uff0c\u8be5\u6a21\u578b\u5728\u5371\u9669\u9884\u6d4b\u7684\u76f4\u89c2\u6027\u548c\u51c6\u786e\u6027\u65b9\u9762\u6709\u663e\u8457\u63d0\u5347\uff0c\u6cdb\u5316\u6027\u80fd\u4e5f\u5f97\u5230\u4e86\u660e\u663e\u6539\u5584\u3002\u8fd9\u4e00\u8fdb\u5c55\u589e\u5f3a\u4e86\u81ea\u52a8\u9a7e\u9a76\u7cfb\u7edf\u7684\u9c81\u68d2\u6027\u548c\u5b89\u5168\u6027\uff0c\u786e\u4fdd\u5728\u590d\u6742\u7684\u73b0\u5b9e\u4e16\u754c\u573a\u666f\u4e2d\u63d0\u5347\u6001\u52bf\u611f\u77e5\u80fd\u529b\u548c\u6f5c\u5728\u51b3\u7b56\u80fd\u529b\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic4.zhimg.com\/v2-957a0f99fad18c3b4e1ae651963712e9_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic4.zhimg.com\/v2-abdf642dea3366cbe21adfc7d67dfa49_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_19\">Scenario Understanding<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aScenario Understanding of Traffic Scenes Through Large Visual Language Models<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/pdf\/2501.17131\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/pdf\/2501.17131<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u6155\u5c3c\u9ed1\u5de5\u4e1a\u5927\u5b66<\/li>\n<\/ul>\n\n\n\n<p>WACV2025\u4e2d\u7a3f\u7684\u5de5\u4f5c\uff0c\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u5728\u81ea\u52a8\u9a7e\u9a76\u9886\u57df\uff08\u5305\u62ec\u611f\u77e5\u3001\u89c4\u5212\u548c\u63a7\u5236\uff09\u7684\u9ad8\u6027\u80fd\u4f9d\u8d56\u4e8e\u6d77\u91cf\u6570\u636e\u96c6\u3002\u7136\u800c\uff0c\u7531\u4e8e\u7279\u5b9a\u9886\u57df\u7684\u6570\u636e\u5206\u5e03\uff0c\u8fd9\u4e9b\u6a21\u578b\u7684\u6cdb\u5316\u80fd\u529b\u5f80\u5f80\u53d7\u9650\uff0c\u56e0\u6b64\u9700\u8981\u5bf9\u6837\u672c\u8fdb\u884c\u6709\u6548\u7684\u57fa\u4e8e\u573a\u666f\u7684\u5206\u7c7b\uff0c\u4ee5\u63d0\u9ad8\u5176\u5728\u4e0d\u540c\u9886\u57df\u7684\u53ef\u9760\u6027\u3002\u4eba\u5de5\u6807\u6ce8\u867d\u6709\u4ef7\u503c\uff0c\u4f46\u65e2\u8017\u8d39\u4eba\u529b\u53c8\u8017\u65f6\uff0c\u6210\u4e3a\u6570\u636e\u6807\u6ce8\u8fc7\u7a0b\u4e2d\u7684\u74f6\u9888\u3002\u5927\u578b\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08LVLMs\uff09\u901a\u8fc7\u4e0a\u4e0b\u6587\u67e5\u8be2\u5b9e\u73b0\u56fe\u50cf\u5206\u6790\u548c\u5206\u7c7b\u7684\u81ea\u52a8\u5316\uff0c\u4e14\u901a\u5e38\u65e0\u9700\u4e3a\u65b0\u7c7b\u522b\u91cd\u65b0\u8bad\u7ec3\uff0c\u4e3a\u89e3\u51b3\u8fd9\u4e00\u95ee\u9898\u63d0\u4f9b\u4e86\u6781\u5177\u5438\u5f15\u529b\u7684\u65b9\u6848\u3002\u5728\u672c\u7814\u7a76\u4e2d\uff0c\u6211\u4eec\u8bc4\u4f30\u4e86\u5305\u62ec GPT-4 \u548c LLaVA \u5728\u5185\u7684\u5927\u578b\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u5728\u5185\u90e8\u6570\u636e\u96c6\u548c BDD100K \u6570\u636e\u96c6\u4e0a\u5bf9\u57ce\u5e02\u4ea4\u901a\u573a\u666f\u7684\u7406\u89e3\u4e0e\u5206\u7c7b\u80fd\u529b\u3002\u6211\u4eec\u63d0\u51fa\u4e86\u4e00\u4e2a\u53ef\u6269\u5c55\u7684\u6807\u6ce8\u7ba1\u9053\uff0c\u6574\u5408\u4e86\u6700\u5148\u8fdb\u7684\u6a21\u578b\uff0c\u80fd\u591f\u7075\u6d3b\u90e8\u7f72\u4e8e\u65b0\u6570\u636e\u96c6\u3002\u901a\u8fc7\u5b9a\u91cf\u6307\u6807\u4e0e\u5b9a\u6027\u5206\u6790\u76f8\u7ed3\u5408\u7684\u65b9\u5f0f\uff0c\u6211\u4eec\u7684\u7814\u7a76\u8bc1\u5b9e\u4e86\u5927\u578b\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u5728\u7406\u89e3\u57ce\u5e02\u4ea4\u901a\u573a\u666f\u65b9\u9762\u7684\u6709\u6548\u6027\uff0c\u5e76\u51f8\u663e\u4e86\u5176\u4f5c\u4e3a\u81ea\u52a8\u9a7e\u9a76\u9886\u57df\u6570\u636e\u9a71\u52a8\u8fdb\u5c55\u7684\u9ad8\u6548\u5de5\u5177\u7684\u6f5c\u529b\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic1.zhimg.com\/v2-b96d05af8ddaf6a5a60f3628089bb854_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic3.zhimg.com\/v2-e268ae37c819383c69a9c5cc93e3ed8c_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_20\">DriveLM<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aDriveLM: Driving with Graph Visual Question Answering<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2312.14150\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2312.14150<\/a><\/li>\n\n\n\n<li>\u9879\u76ee\u4e3b\u9875\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/github.com\/OpenDriveLab\/DriveLM\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/github.com\/OpenDriveLab\/DriveLM<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u4e0a\u6d77\u4eba\u5de5\u667a\u80fd\u5b9e\u9a8c\u5ba4\uff0c\u9999\u6e2f\u5927\u5b66\uff0c\u56fe\u5bbe\u6839\u5927\u5b66<\/li>\n<\/ul>\n\n\n\n<p>ECCV 2024\u4e2d\u7a3f\u7684\u5de5\u4f5c\uff0c\u6211\u4eec\u7814\u7a76\u4e86\u5982\u4f55\u5c06\u57fa\u4e8e\u7f51\u7edc\u89c4\u6a21\u6570\u636e\u8bad\u7ec3\u7684\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08VLMs\uff09\u96c6\u6210\u5230\u7aef\u5230\u7aef\u9a7e\u9a76\u7cfb\u7edf\u4e2d\uff0c\u4ee5\u589e\u5f3a\u6cdb\u5316\u80fd\u529b\u5e76\u5b9e\u73b0\u4e0e\u4eba\u7c7b\u7528\u6237\u7684\u4ea4\u4e92\u3002\u5c3d\u7ba1\u8fd1\u5e74\u6765\u7684\u65b9\u6cd5\u901a\u8fc7\u5355\u8f6e\u89c6\u89c9\u95ee\u7b54\uff08VQA\uff09\u4f7f VLMs \u9002\u5e94\u9a7e\u9a76\u573a\u666f\uff0c\u4f46\u4eba\u7c7b\u9a7e\u9a76\u5458\u5728\u505a\u51b3\u7b56\u65f6\u4f1a\u8fdb\u884c\u591a\u6b65\u9aa4\u63a8\u7406\uff1a\u4ece\u5173\u952e\u76ee\u6807\u7684\u5b9a\u4f4d\u5f00\u59cb\uff0c\u5728\u91c7\u53d6\u884c\u52a8\u524d\u5148\u4f30\u8ba1\u76ee\u6807\u95f4\u7684\u4ea4\u4e92\u3002\u6838\u5fc3\u89c1\u89e3\u662f\uff0c\u901a\u8fc7\u6211\u4eec\u63d0\u51fa\u7684\u56fe\u89c6\u89c9\u95ee\u7b54\uff08Graph VQA\uff09\u4efb\u52a1 \u2014\u2014 \u8be5\u4efb\u52a1\u901a\u8fc7\u611f\u77e5\u3001\u9884\u6d4b\u548c\u89c4\u5212\u95ee\u7b54\u5bf9\u6765\u5efa\u6a21\u56fe\u7ed3\u6784\u63a8\u7406 \u2014\u2014 \u6211\u4eec\u83b7\u5f97\u4e86\u4e00\u4e2a\u5408\u9002\u7684\u4ee3\u7406\u4efb\u52a1\uff0c\u4ee5\u6a21\u62df\u4eba\u7c7b\u7684\u63a8\u7406\u8fc7\u7a0b\u3002\u6211\u4eec\u57fa\u4e8e nuScenes \u548c CARLA \u6784\u5efa\u4e86\u6570\u636e\u96c6\uff08DriveLM-Data\uff09\uff0c\u5e76\u63d0\u51fa\u4e86\u4e00\u79cd\u57fa\u4e8e VLM \u7684\u57fa\u7ebf\u65b9\u6cd5\uff08DriveLM-Agent\uff09\uff0c\u7528\u4e8e\u8054\u5408\u6267\u884c\u56fe\u89c6\u89c9\u95ee\u7b54\u548c\u7aef\u5230\u7aef\u9a7e\u9a76\u3002\u5b9e\u9a8c\u8868\u660e\uff0c\u56fe\u89c6\u89c9\u95ee\u7b54\u4e3a\u9a7e\u9a76\u573a\u666f\u63a8\u7406\u63d0\u4f9b\u4e86\u4e00\u4e2a\u7b80\u5355\u3001\u6709\u539f\u5219\u7684\u6846\u67b6\uff0c\u800c DriveLM-Data \u4e3a\u8be5\u4efb\u52a1\u63d0\u4f9b\u4e86\u4e00\u4e2a\u5177\u6709\u6311\u6218\u6027\u7684\u57fa\u51c6\u3002\u6211\u4eec\u7684 DriveLM-Agent \u57fa\u7ebf\u5728\u7aef\u5230\u7aef\u81ea\u52a8\u9a7e\u9a76\u65b9\u9762\u7684\u8868\u73b0\u4e0e\u6700\u5148\u8fdb\u7684\u7279\u5b9a\u9a7e\u9a76\u67b6\u6784\u76f8\u5f53\u3002\u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c\u5728\u5bf9\u672a\u89c1\u8fc7\u7684\u4f20\u611f\u5668\u914d\u7f6e\u8fdb\u884c\u96f6\u6837\u672c\u8bc4\u4f30\u65f6\uff0c\u5176\u4f18\u52bf\u5c24\u4e3a\u660e\u663e\u3002\u6211\u4eec\u7684\u9010\u95ee\u9898\u6d88\u878d\u7814\u7a76\u8868\u660e\uff0c\u6027\u80fd\u63d0\u5347\u6765\u81ea\u4e8e\u56fe\u7ed3\u6784\u4e2d\u9884\u6d4b\u548c\u89c4\u5212\u95ee\u7b54\u5bf9\u7684\u4e30\u5bcc\u6807\u6ce8\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic2.zhimg.com\/v2-a5dbe2c261cfbc0d47aa910b640c779b_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic4.zhimg.com\/v2-565c1930fe56fd667777d3aa177208c5_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_21\">POTGui<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aEnhancing Large Vision Model in Street Scene Semantic Understanding through Leveraging Posterior Optimization Trajectory<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2501.01710\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2501.01710<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u9999\u6e2f\u5927\u5b66\uff0c\u5357\u65b9\u79d1\u6280\u5927\u5b66\uff0c\u4e2d\u56fd\u79d1\u5b66\u9662\u6df1\u5733\u5148\u8fdb\u6280\u672f\u7814\u7a76\u9662<\/li>\n<\/ul>\n\n\n\n<p>\u4e3a\u63d0\u9ad8\u81ea\u52a8\u9a7e\u9a76\uff08AD\uff09\u611f\u77e5\u6a21\u578b\u7684\u6cdb\u5316\u80fd\u529b\uff0c\u8f66\u8f86\u9700\u8981\u57fa\u4e8e\u6301\u7eed\u6536\u96c6\u7684\u6570\u636e\u968f\u65f6\u95f4\u66f4\u65b0\u6a21\u578b\u3002\u968f\u7740\u65f6\u95f4\u63a8\u79fb\uff0cAD \u6a21\u578b\u62df\u5408\u7684\u6570\u636e\u91cf\u4e0d\u65ad\u6269\u5927\uff0c\u8fd9\u663e\u8457\u6709\u52a9\u4e8e\u63d0\u5347\u6a21\u578b\u7684\u6cdb\u5316\u80fd\u529b\u3002\u7136\u800c\uff0c\u8fd9\u79cd\u4e0d\u65ad\u81a8\u80c0\u7684\u6570\u636e\u5bf9 AD \u6a21\u578b\u800c\u8a00\u662f\u628a\u53cc\u5203\u5251\u3002\u5177\u4f53\u6765\u8bf4\uff0c\u5f53\u62df\u5408\u7684\u6570\u636e\u91cf\u8d85\u8fc7 AD \u6a21\u578b\u7684\u62df\u5408\u80fd\u529b\u65f6\uff0c\u6a21\u578b\u5bb9\u6613\u51fa\u73b0\u6b20\u62df\u5408\u95ee\u9898\u3002\u4e3a\u89e3\u51b3\u8fd9\u4e00\u95ee\u9898\uff0c\u6211\u4eec\u63d0\u51fa\u91c7\u7528\u9884\u8bad\u7ec3\u7684\u5927\u578b\u89c6\u89c9\u6a21\u578b\uff08LVMs\uff09\u4f5c\u4e3a\u4e3b\u5e72\u7f51\u7edc\uff0c\u7ed3\u5408\u4e0b\u6e38\u611f\u77e5\u5934\u6765\u7406\u89e3 AD \u8bed\u4e49\u4fe1\u606f\u3002\u8fd9\u79cd\u8bbe\u8ba1\u4e0d\u4ec5\u80fd\u51ed\u501f LVMs \u5f3a\u5927\u7684\u62df\u5408\u80fd\u529b\u514b\u670d\u4e0a\u8ff0\u6b20\u62df\u5408\u95ee\u9898\uff0c\u8fd8\u80fd\u501f\u52a9 LVMs \u6d77\u91cf\u4e14\u591a\u6837\u7684\u8bad\u7ec3\u6570\u636e\u589e\u5f3a\u611f\u77e5\u6cdb\u5316\u80fd\u529b\u3002\u53e6\u4e00\u65b9\u9762\uff0c\u4e3a\u51cf\u8f7b\u8f66\u8f86\u5728\u8fd0\u884c LVM \u4e3b\u5e72\u7f51\u7edc\u65f6\u8bad\u7ec3\u611f\u77e5\u5934\u7684\u8ba1\u7b97\u8d1f\u62c5\uff0c\u6211\u4eec\u5f15\u5165\u540e\u9a8c\u4f18\u5316\u8f68\u8ff9\uff08POT\uff09\u5f15\u5bfc\u7684\u4f18\u5316\u65b9\u6848\uff08POTGui\uff09\u4ee5\u52a0\u901f\u6536\u655b\u3002\u5177\u4f53\u800c\u8a00\uff0c\u6211\u4eec\u8bbe\u8ba1\u4e86 POT \u751f\u6210\u5668\uff08POTGen\uff09\uff0c\u63d0\u524d\u751f\u6210\u540e\u9a8c\uff08\u672a\u6765\uff09\u4f18\u5316\u65b9\u5411\uff0c\u7528\u4e8e\u6307\u5bfc\u5f53\u524d\u7684\u4f18\u5316\u8fed\u4ee3\uff0c\u4f7f\u6a21\u578b\u901a\u5e38\u80fd\u5728 10 \u4e2a epoch \u5185\u6536\u655b\u3002\u5927\u91cf\u5b9e\u9a8c\u8868\u660e\uff0c\u4e0e\u73b0\u6709\u7684\u6700\u5148\u8fdb\u65b9\u6cd5\u76f8\u6bd4\uff0c\u6240\u63d0\u65b9\u6cd5\u6027\u80fd\u63d0\u5347\u8d85\u8fc7 66.48%\uff0c\u6536\u655b\u901f\u5ea6\u52a0\u5feb 6 \u500d\u4ee5\u4e0a\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic3.zhimg.com\/v2-87e54f93807b978cb2117ac70f0e3204_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic3.zhimg.com\/v2-8c43f74858533f4076dfe2c4ffc46a66_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_22\">VLM\u7efc\u8ff0<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aAre VLMs Ready for Autonomous Driving? An Empirical Study from the Reliability, Data, and Metric Perspectives<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2501.04003\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2501.04003<\/a><\/li>\n\n\n\n<li>\u9879\u76ee\u4e3b\u9875\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/drive-bench.github.io\/\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/drive-bench.github.io\/<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u52a0\u5dde\u5927\u5b66\u6b27\u6587\u5206\u6821\uff0c\u4e0a\u6d77\u4eba\u5de5\u667a\u80fd\u5b9e\u9a8c\u5ba4\uff0c\u5357\u6d0b\u7406\u5de5\u5927\u5b66S-Lab\u7b49<\/li>\n<\/ul>\n\n\n\n<p>\u8fd1\u5e74\u6765\uff0c\u89c6\u89c9 &#8211; \u8bed\u8a00\u6a21\u578b\uff08VLMs\uff09\u7684\u8fdb\u6b65\u5f15\u53d1\u4e86\u5c06\u5176\u5e94\u7528\u4e8e\u81ea\u52a8\u9a7e\u9a76\u7684\u5174\u8da3\uff0c\u5c24\u5176\u662f\u5728\u901a\u8fc7\u81ea\u7136\u8bed\u8a00\u751f\u6210\u53ef\u89e3\u91ca\u7684\u9a7e\u9a76\u51b3\u7b56\u65b9\u9762\u3002\u7136\u800c\uff0cVLMs \u662f\u5426\u5929\u751f\u80fd\u4e3a\u9a7e\u9a76\u63d0\u4f9b\u57fa\u4e8e\u89c6\u89c9\u7684\u3001\u53ef\u9760\u4e14\u53ef\u89e3\u91ca\u7684\u89e3\u91ca\u8fd9\u4e00\u5047\u8bbe\u5728\u5f88\u5927\u7a0b\u5ea6\u4e0a\u5c1a\u672a\u5f97\u5230\u68c0\u9a8c\u3002\u4e3a\u586b\u8865\u8fd9\u4e00\u7a7a\u767d\uff0c\u6211\u4eec\u5f15\u5165\u4e86 DriveBench\uff0c\u8fd9\u4e00\u57fa\u51c6\u6570\u636e\u96c6\u65e8\u5728\u8bc4\u4f30 VLMs \u5728 17 \u79cd\u8bbe\u7f6e\uff08\u6e05\u6d01\u3001\u53d7\u635f\u548c\u7eaf\u6587\u672c\u8f93\u5165\uff09\u4e0b\u7684\u53ef\u9760\u6027\uff0c\u6db5\u76d6 19,200 \u5e27\u300120,498 \u4e2a\u95ee\u7b54\u5bf9\u3001\u4e09\u79cd\u95ee\u9898\u7c7b\u578b\u3001\u56db\u9879\u4e3b\u6d41\u9a7e\u9a76\u4efb\u52a1\uff0c\u4ee5\u53ca\u5171 12 \u4e2a\u4e3b\u6d41 VLMs\u3002\u6211\u4eec\u7684\u7814\u7a76\u53d1\u73b0\uff0cVLMs \u5f80\u5f80\u4f1a\u57fa\u4e8e\u5e38\u8bc6\u6216\u6587\u672c\u7ebf\u7d22\u751f\u6210\u770b\u4f3c\u5408\u7406\u7684\u54cd\u5e94\uff0c\u800c\u975e\u771f\u6b63\u57fa\u4e8e\u89c6\u89c9\uff0c\u5c24\u5176\u662f\u5728\u89c6\u89c9\u8f93\u5165\u9000\u5316\u6216\u7f3a\u5931\u7684\u60c5\u51b5\u4e0b\u3002\u8fd9\u79cd\u884c\u4e3a\u88ab\u6570\u636e\u96c6\u4e0d\u5e73\u8861\u548c\u8bc4\u4f30\u6307\u6807\u4e0d\u8db3\u6240\u63a9\u76d6\uff0c\u5728\u81ea\u52a8\u9a7e\u9a76\u7b49\u5b89\u5168\u5173\u952e\u573a\u666f\u4e2d\u6784\u6210\u91cd\u5927\u98ce\u9669\u3002\u6211\u4eec\u8fd8\u89c2\u5bdf\u5230\uff0cVLMs \u5728\u591a\u6a21\u6001\u63a8\u7406\u65b9\u9762\u5b58\u5728\u56f0\u96be\uff0c\u4e14\u5bf9\u8f93\u5165\u635f\u574f\u9ad8\u5ea6\u654f\u611f\uff0c\u5bfc\u81f4\u6027\u80fd\u4e0d\u4e00\u81f4\u3002\u4e3a\u5e94\u5bf9\u8fd9\u4e9b\u6311\u6218\uff0c\u6211\u4eec\u63d0\u51fa\u4e86\u6539\u8fdb\u7684\u8bc4\u4f30\u6307\u6807\uff0c\u4f18\u5148\u8003\u8651\u7a33\u5065\u7684\u89c6\u89c9\u63a5\u5730\u548c\u591a\u6a21\u6001\u7406\u89e3\u3002\u6b64\u5916\uff0c\u6211\u4eec\u5f3a\u8c03\u4e86\u5229\u7528 VLMs \u5bf9\u635f\u574f\u7684\u611f\u77e5\u6765\u63d0\u9ad8\u5176\u53ef\u9760\u6027\u7684\u6f5c\u529b\uff0c\u4e3a\u5f00\u53d1\u66f4\u53ef\u4fe1\u3001\u53ef\u89e3\u91ca\u7684\u73b0\u5b9e\u4e16\u754c\u81ea\u52a8\u9a7e\u9a76\u51b3\u7b56\u7cfb\u7edf\u63d0\u4f9b\u4e86\u8def\u7ebf\u56fe\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/picx.zhimg.com\/v2-f6e098fcbc27672f9f24dfbf216835a9_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic2.zhimg.com\/v2-3b2a4e1c007a41a14de344b043fccb7d_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_23\">VisionLLM<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aSFF Rendering-Based Uncertainty Prediction using VisionLLM<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/openreview.net\/forum%3Fid%3Dq8ptjh1pDl\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/openreview.net\/forum?id=q8ptjh1pDl<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u97e9\u56fd\u5ef6\u4e16\u5927\u5b66<\/li>\n<\/ul>\n\n\n\n<p>\u672c\u7814\u7a76\u63d0\u51fa\u4e86\u4e00\u79cd\u57fa\u4e8eVisionLLM\u7684\u521b\u65b0\u6846\u67b6\uff0c\u7528\u4e8e\u81ea\u52a8\u9a7e\u9a76\u4e2d\u7684\u4e0d\u786e\u5b9a\u6027\u9884\u6d4b\u3002\u901a\u8fc7\u5229\u7528CARLA\u4eff\u771f\u5e73\u53f0\u91c7\u96c6\u7684\u9a7e\u9a76\u6570\u636e\uff0c\u6211\u4eec\u751f\u6210\u4e86\u4e0e\u540e\u7eed\u9a7e\u9a76\u52a8\u4f5c\u53ca\u4e0d\u786e\u5b9a\u6027\u5206\u503c\u914d\u5bf9\u7684BEV\u3002\u4e3a\u6a21\u62df\u771f\u5b9e\u9a7e\u9a76\u6311\u6218\uff0c\u6211\u4eec\u5728BEV\u4e2d\u5f15\u5165\u906e\u6321\u63a9\u819c\uff0c\u4ee5\u8868\u5f81\u4f20\u611f\u5668\u5c40\u9650\u5bfc\u81f4\u7684\u89c6\u91ce\u76f2\u533a\u3002\u8be5\u6a21\u578b\u901a\u8fc7\u878d\u5408\u989d\u5916\u56fe\u50cf\u8f93\u5165\uff0c\u5728\u906e\u6321\u4e25\u91cd\u7684\u590d\u6742\u573a\u666f\u4e2d\u540c\u6b65\u9884\u6d4b\u540e\u7eed\u9a7e\u9a76\u52a8\u4f5c\u4e0e\u4e0d\u786e\u5b9a\u6027\u5206\u503c\uff0c\u663e\u8457\u63d0\u5347\u4e86\u63a8\u7406\u80fd\u529b\u3002\u91c7\u7528\u53c2\u6570\u9ad8\u6548\u5fae\u8c03\u6280\u672f\uff08PEFT\uff09\u5982LoRA\u5bf9VisionLLM\u8fdb\u884c\u5fae\u8c03\uff0c\u5b9e\u9a8c\u7ed3\u679c\u8bc1\u5b9e\u4e86\u8be5\u65b9\u6cd5\u5728\u89e3\u51b3\u906e\u6321\u76f8\u5173\u4e0d\u786e\u5b9a\u6027\u95ee\u9898\u7684\u6548\u80fd\uff0c\u4e3a\u9ad8\u9636\u9a7e\u9a76\u81ea\u52a8\u5316\u7cfb\u7edf\u5b9e\u73b0\u66f4\u5b89\u5168\u53ef\u9760\u7684\u51b3\u7b56\u673a\u5236\u5960\u5b9a\u4e86\u6280\u672f\u57fa\u7840\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/picx.zhimg.com\/v2-aa04a01153531f96c00ee32adcae971d_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic1.zhimg.com\/v2-6e68b2c237ae87f5381de82750d894cc_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_24\">CODA-LM<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aAutomated Evaluation of Large Vision-Language Models on Self-driving Corner Cases<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2404.10595\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2404.10595<\/a><\/li>\n\n\n\n<li>\u9879\u76ee\u4e3b\u9875\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/coda-dataset.github.io\/coda-lm\/\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/coda-dataset.github.io\/coda-lm\/<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u9999\u6e2f\u79d1\u6280\u5927\u5b66\uff0c\u5927\u8fde\u7406\u5de5\u5927\u5b66\uff0c\u9999\u6e2f\u4e2d\u6587\u5927\u5b66\uff0c\u534e\u4e3a\u8bfa\u4e9a\u65b9\u821f\u5b9e\u9a8c\u5ba4<\/li>\n<\/ul>\n\n\n\n<p>WACV 2025\u4e2d\u7a3f\u7684\u5de5\u4f5c\uff0c\u5927\u578b\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08LVLMs\uff09\u5728\u63a8\u52a8\u53ef\u89e3\u91ca\u81ea\u52a8\u9a7e\u9a76\u65b9\u9762\u53d7\u5230\u4e86\u5e7f\u6cdb\u5173\u6ce8\u3002\u73b0\u6709\u7684 LVLMs \u8bc4\u4f30\u4e3b\u8981\u96c6\u4e2d\u4e8e\u81ea\u7136\u573a\u666f\u4e0b\u7684\u591a\u65b9\u9762\u80fd\u529b\uff0c\u7f3a\u4e4f\u9488\u5bf9\u81ea\u52a8\u9a7e\u9a76\u7684\u81ea\u52a8\u5316\u3001\u53ef\u91cf\u5316\u8bc4\u4f30\uff0c\u66f4\u4e0d\u7528\u8bf4\u5bf9\u4e25\u5cfb\u7684\u9053\u8def\u6781\u7aef\u573a\u666f\u7684\u8bc4\u4f30\u4e86\u3002\u5728\u672c\u7814\u7a76\u4e2d\uff0c\u6211\u4eec\u63d0\u51fa\u4e86 CODA-LM\uff0c\u8fd9\u662f\u9996\u4e2a\u7528\u4e8e\u81ea\u52a8\u9a7e\u9a76\u6781\u7aef\u573a\u666f\u4e0b LVLMs \u81ea\u52a8\u8bc4\u4f30\u7684\u57fa\u51c6\u3002\u6211\u4eec\u91c7\u7528\u5206\u5c42\u6570\u636e\u7ed3\u6784\uff0c\u63d0\u793a\u5f3a\u5927\u7684 LVLMs \u5206\u6790\u590d\u6742\u9a7e\u9a76\u573a\u666f\uff0c\u5e76\u4e3a\u4eba\u5de5\u6807\u6ce8\u8005\u751f\u6210\u9ad8\u8d28\u91cf\u7684\u9884\u6807\u6ce8\uff1b\u800c\u5728 LVLM \u8bc4\u4f30\u65b9\u9762\uff0c\u6211\u4eec\u53d1\u73b0\u4f7f\u7528\u7eaf\u6587\u672c\u5927\u8bed\u8a00\u6a21\u578b\uff08LLMs\uff09\u4f5c\u4e3a\u8bc4\u5224\u8005\uff0c\u4e0e\u4eba\u7c7b\u504f\u597d\u7684\u4e00\u81f4\u6027\u751a\u81f3\u4f18\u4e8e LVLM \u8bc4\u5224\u8005\u3002\u6b64\u5916\uff0c\u501f\u52a9\u6211\u4eec\u7684 CODA-LM\uff0c\u6211\u4eec\u6784\u5efa\u4e86 CODA-VLM\u2014\u2014 \u4e00\u79cd\u65b0\u7684\u9a7e\u9a76 LVLM\uff0c\u5176\u5728 CODA-LM \u4e0a\u7684\u8868\u73b0\u8d85\u8fc7\u4e86\u6240\u6709\u5f00\u6e90\u540c\u7c7b\u6a21\u578b\u3002CODA-VLM \u7684\u6027\u80fd\u4e0e GPT-4V \u76f8\u5f53\uff0c\u5728\u533a\u57df\u611f\u77e5\u4efb\u52a1\u4e0a\u751a\u81f3\u8d85\u8fc7 GPT-4V \u8fbe 21.42%\u3002\u6211\u4eec\u5e0c\u671b CODA-LM \u80fd\u6210\u4e3a\u63a8\u52a8 LVLMs \u8d4b\u80fd\u7684\u53ef\u89e3\u91ca\u81ea\u52a8\u9a7e\u9a76\u53d1\u5c55\u7684\u50ac\u5316\u5242\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pica.zhimg.com\/v2-7b613c854ddc5fc66e6add15f85c5086_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic4.zhimg.com\/v2-f4210fbce3679f37519faf34181ac24b_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_25\">Think-Driver<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aThink-Driver: From Driving-Scene Understanding to Decision-Making with Vision Language Models<\/li>\n\n\n\n<li>PDF\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/mllmav.github.io\/papers\/Think-Driver%3A%2520From%2520Driving-Scene%2520Understanding%2520to%2520Decision-Making%2520with%2520Vision%2520Language%2520Models.pdf\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/mllmav.github.io\/papers\/Think-Driver:%20From%20Driving-Scene%20Understanding%20to%20Decision-Making%20with%20Vision%20Language%20Models.pdf<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u9999\u6e2f\u79d1\u6280\u5927\u5b66\uff08\u5e7f\u5dde\uff09\uff0c\u7ea6\u7ff0\u65af\u30fb\u970d\u666e\u91d1\u65af\u5927\u5b66<\/li>\n<\/ul>\n\n\n\n<p>\u81ea\u52a8\u9a7e\u9a76\u8fd1\u671f\u5728\u4eff\u771f\u73af\u5883\u4e0e\u73b0\u5b9e\u573a\u666f\u4e2d\u7684\u8868\u73b0\u5747\u53d6\u5f97\u4e86\u663e\u8457\u7a81\u7834\uff0c\u7aef\u5230\u7aef\u65b9\u6cd5\u5c24\u5176\u4ee4\u4eba\u77a9\u76ee\u3002\u7136\u800c\uff0c\u8fd9\u7c7b\u6a21\u578b\u5f80\u5f80\u5982\u540c\u9ed1\u76d2\u822c\u8fd0\u4f5c\uff0c\u7f3a\u4e4f\u53ef\u89e3\u91ca\u6027\u3002\u5927\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u7684\u51fa\u73b0\u4e3a\u89e3\u51b3\u6b64\u95ee\u9898\u63d0\u4f9b\u4e86\u53ef\u80fd\uff0c\u5b83\u5c06\u6a21\u5757\u5316\u81ea\u52a8\u9a7e\u9a76\u4e0e\u8bed\u8a00\u89e3\u91ca\u80fd\u529b\u76f8\u7ed3\u5408\u3002\u76ee\u524d\u4e3b\u6d41\u7684LLM\u65b9\u6848\u5c06\u9a7e\u9a76\u8f93\u5165\u4fe1\u606f\u8f6c\u5316\u4e3a\u8bed\u8a00\u63cf\u8ff0\uff0c\u4f46\u8fd9\u901a\u5e38\u4f9d\u8d56\u4eba\u5de5\u8bbe\u8ba1\u7684\u63d0\u793a\u8bed\uff0c\u4e14\u53ef\u80fd\u5bfc\u81f4\u4fe1\u606f\u6548\u7387\u4e0d\u5c3d\u7406\u60f3\u3002\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08VLM\uff09\u867d\u53ef\u76f4\u63a5\u4ece\u56fe\u50cf\u4e2d\u63d0\u53d6\u4fe1\u606f\uff0c\u4f46\u5728\u6d89\u53ca\u8fde\u7eed\u9a7e\u9a76\u573a\u666f\u7406\u89e3\u4e0e\u4e0a\u4e0b\u6587\u63a8\u7406\u7684\u4efb\u52a1\u4e2d\u6709\u65f6\u8868\u73b0\u6b20\u4f73\u3002<\/p>\n\n\n\n<p>\u672c\u6587\u63d0\u51faThinkDriver\u6a21\u578b\uff0c\u8fd9\u662f\u4e00\u79cd\u5229\u7528\u591a\u89c6\u89d2\u56fe\u50cf\u751f\u6210\u7406\u6027\u9a7e\u9a76\u51b3\u7b56\u53ca\u63a8\u7406\u8fc7\u7a0b\u7684\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u3002\u8be5\u6a21\u578b\u80fd\u591f\u8bc4\u4f30\u611f\u77e5\u5230\u7684\u4ea4\u901a\u72b6\u51b5\uff0c\u5e76\u5bf9\u5f53\u524d\u9a7e\u9a76\u884c\u4e3a\u7684\u98ce\u9669\u8fdb\u884c\u7814\u5224\uff0c\u4ece\u800c\u52a9\u529b\u5f62\u6210\u7406\u6027\u51b3\u7b56\u3002\u901a\u8fc7\u95ed\u73af\u5b9e\u9a8c\u9a8c\u8bc1\uff0cThinkDriver\u7684\u8868\u73b0\u4f18\u4e8e\u5176\u4ed6\u57fa\u4e8e\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u7684\u57fa\u7ebf\u65b9\u6848\uff0c\u5176\u751f\u6210\u7684\u9a7e\u9a76\u51b3\u7b56\u5177\u5907\u53ef\u89e3\u91ca\u6027\uff0c\u5145\u5206\u8bc1\u660e\u4e86\u8be5\u6a21\u578b\u7684\u6709\u6548\u6027\u53ca\u5176\u5728\u672a\u6765\u5e94\u7528\u4e2d\u7684\u6f5c\u529b\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic3.zhimg.com\/v2-ebb65e9598c3225965b4f596f4d35a86_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic3.zhimg.com\/v2-4a19dd6e52647ace707688d0c7ec0872_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_26\">LLM-Augmented-MTR<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aLarge Language Models Powered Context-aware Motion Prediction in Autonomous Driving<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2403.11057\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2403.11057<\/a><\/li>\n\n\n\n<li>\u9879\u76ee\u4e3b\u9875\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/github.com\/AIR-DISCOVER\/LLM-Augmented-MTR\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/github.com\/AIR-DISCOVER\/LLM-Augmented-MTR<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u6e05\u534e\u5927\u5b66<\/li>\n<\/ul>\n\n\n\n<p>\u8fd0\u52a8\u9884\u6d4b\u662f\u81ea\u52a8\u9a7e\u9a76\u4e2d\u6700\u57fa\u7840\u7684\u4efb\u52a1\u4e4b\u4e00\u3002\u4f20\u7edf\u7684\u8fd0\u52a8\u9884\u6d4b\u65b9\u6cd5\u4e3b\u8981\u5bf9\u5730\u56fe\u7684\u5411\u91cf\u4fe1\u606f\u548c\u4ea4\u901a\u53c2\u4e0e\u8005\u7684\u5386\u53f2\u8f68\u8ff9\u6570\u636e\u8fdb\u884c\u7f16\u7801\uff0c\u7f3a\u4e4f\u5bf9\u6574\u4f53\u4ea4\u901a\u8bed\u4e49\u7684\u5168\u9762\u7406\u89e3\uff0c\u8fdb\u800c\u5f71\u54cd\u9884\u6d4b\u4efb\u52a1\u7684\u6027\u80fd\u3002\u5728\u672c\u6587\u4e2d\uff0c\u6211\u4eec\u5229\u7528\u5927\u8bed\u8a00\u6a21\u578b\uff08LLMs\uff09\u6765\u589e\u5f3a\u8fd0\u52a8\u9884\u6d4b\u4efb\u52a1\u4e2d\u7684\u5168\u5c40\u4ea4\u901a\u4e0a\u4e0b\u6587\u7406\u89e3\u3002\u6211\u4eec\u9996\u5148\u8fdb\u884c\u4e86\u7cfb\u7edf\u7684\u63d0\u793a\u8bcd\u5de5\u7a0b\uff0c\u5c06\u590d\u6742\u7684\u4ea4\u901a\u73af\u5883\u548c\u4ea4\u901a\u53c2\u4e0e\u8005\u7684\u5386\u53f2\u8f68\u8ff9\u4fe1\u606f\u53ef\u89c6\u5316\u5230\u56fe\u50cf\u63d0\u793a \u2014\u2014 \u4ea4\u901a\u573a\u666f\u56fe\uff08TC-Map\uff09\u4e2d\uff0c\u5e76\u8f85\u4ee5\u76f8\u5e94\u7684\u6587\u672c\u63d0\u793a\u3002\u901a\u8fc7\u8fd9\u79cd\u65b9\u6cd5\uff0c\u6211\u4eec\u4ece\u5927\u8bed\u8a00\u6a21\u578b\u4e2d\u83b7\u53d6\u4e86\u4e30\u5bcc\u7684\u4ea4\u901a\u4e0a\u4e0b\u6587\u4fe1\u606f\u3002\u5c06\u8fd9\u4e9b\u4fe1\u606f\u96c6\u6210\u5230\u8fd0\u52a8\u9884\u6d4b\u6a21\u578b\u4e2d\u540e\uff0c\u6211\u4eec\u8bc1\u660e\u6b64\u7c7b\u4e0a\u4e0b\u6587\u80fd\u591f\u63d0\u9ad8\u8fd0\u52a8\u9884\u6d4b\u7684\u51c6\u786e\u6027\u3002\u6b64\u5916\uff0c\u8003\u8651\u5230\u5927\u8bed\u8a00\u6a21\u578b\u7684\u6210\u672c\uff0c\u6211\u4eec\u63d0\u51fa\u4e86\u4e00\u79cd\u7ecf\u6d4e\u9ad8\u6548\u7684\u90e8\u7f72\u7b56\u7565\uff1a\u4f7f\u7528 0.7% \u7684\u5927\u8bed\u8a00\u6a21\u578b\u589e\u5f3a\u6570\u636e\u96c6\uff0c\u5927\u89c4\u6a21\u63d0\u5347\u8fd0\u52a8\u9884\u6d4b\u4efb\u52a1\u7684\u51c6\u786e\u6027\u3002\u6211\u4eec\u7684\u7814\u7a76\u4e3a\u589e\u5f3a\u5927\u8bed\u8a00\u6a21\u578b\u5bf9\u4ea4\u901a\u573a\u666f\u7684\u7406\u89e3\u4ee5\u53ca\u81ea\u52a8\u9a7e\u9a76\u7684\u8fd0\u52a8\u9884\u6d4b\u6027\u80fd\u63d0\u4f9b\u4e86\u6709\u4ef7\u503c\u7684\u89c1\u89e3\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic3.zhimg.com\/v2-356589ff0403a87790171fc0de9508be_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic4.zhimg.com\/v2-7292351610e88ea317b4325031cae4ad_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_27\">Reason2Drive<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aReason2Drive: Towards Interpretable and Chain-based Reasoning for Autonomous Driving<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2312.03661\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2312.03661<\/a><\/li>\n\n\n\n<li>\u9879\u76ee\u4e3b\u9875\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/github.com\/fudan-zvg\/reason2drive\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/github.com\/fudan-zvg\/reason2drive<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u590d\u65e6\u5927\u5b66\u3001\u534e\u4e3a\u8bfa\u4e9a\u65b9\u821f\u5b9e\u9a8c\u5ba4<\/li>\n<\/ul>\n\n\n\n<p>ECCV 2024\u4e2d\u7a3f\u7684\u5de5\u4f5c\uff0c\u5927\u578b\u89c6\u89c9 &#8211; \u8bed\u8a00\u6a21\u578b\uff08VLMs\uff09\u5728\u81ea\u52a8\u9a7e\u9a76\u9886\u57df\u5f15\u8d77\u4e86\u8d8a\u6765\u8d8a\u591a\u7684\u5173\u6ce8\uff0c\u8fd9\u5f97\u76ca\u4e8e\u5176\u5728\u9ad8\u5ea6\u81ea\u4e3b\u8f66\u8f86\u884c\u4e3a\u6240\u5fc5\u9700\u7684\u590d\u6742\u63a8\u7406\u4efb\u52a1\u4e2d\u5c55\u73b0\u51fa\u7684\u5148\u8fdb\u80fd\u529b\u3002\u5c3d\u7ba1\u6f5c\u529b\u5de8\u5927\uff0c\u4f46\u81ea\u4e3b\u7cfb\u7edf\u9886\u57df\u7684\u7814\u7a76\u56e0\u7f3a\u4e4f\u5e26\u6709\u6807\u6ce8\u63a8\u7406\u94fe\uff08\u7528\u4e8e\u89e3\u91ca\u9a7e\u9a76\u51b3\u7b56\u8fc7\u7a0b\uff09\u7684\u6570\u636e\u96c6\u800c\u53d7\u963b\u3002\u4e3a\u586b\u8865\u8fd9\u4e00\u7a7a\u767d\uff0c\u6211\u4eec\u63d0\u51fa\u4e86 Reason2Drive\u2014\u2014 \u4e00\u4e2a\u5305\u542b\u8d85\u8fc7 60 \u4e07\u5bf9\u89c6\u9891 &#8211; \u6587\u672c\u5bf9\u7684\u57fa\u51c6\u6570\u636e\u96c6\uff0c\u65e8\u5728\u63a8\u52a8\u590d\u6742\u9a7e\u9a76\u73af\u5883\u4e2d\u53ef\u89e3\u91ca\u63a8\u7406\u7684\u7814\u7a76\u3002\u6211\u4eec\u5c06\u81ea\u52a8\u9a7e\u9a76\u8fc7\u7a0b\u660e\u786e\u63cf\u8ff0\u4e3a\u611f\u77e5\u3001\u9884\u6d4b\u548c\u63a8\u7406\u6b65\u9aa4\u7684\u987a\u5e8f\u7ec4\u5408\uff0c\u5176\u95ee\u7b54\u5bf9\u81ea\u52a8\u4ece\u591a\u79cd\u5f00\u6e90\u6237\u5916\u9a7e\u9a76\u6570\u636e\u96c6\uff08\u5305\u62ec nuScenes\u3001Waymo \u548c ONCE\uff09\u4e2d\u6536\u96c6\u800c\u6765\u3002\u6b64\u5916\uff0c\u6211\u4eec\u5f15\u5165\u4e86\u4e00\u79cd\u65b0\u9896\u7684\u805a\u5408\u8bc4\u4f30\u6307\u6807\uff0c\u7528\u4e8e\u8bc4\u4f30\u81ea\u4e3b\u7cfb\u7edf\u4e2d\u57fa\u4e8e\u94fe\u7684\u63a8\u7406\u6027\u80fd\uff0c\u4ee5\u89e3\u51b3\u73b0\u6709\u6307\u6807\uff08\u5982 BLEU \u548c CIDEr\uff09\u5b58\u5728\u7684\u63a8\u7406\u6a21\u7cca\u6027\u95ee\u9898\u3002\u57fa\u4e8e\u6240\u63d0\u51fa\u7684\u57fa\u51c6\uff0c\u6211\u4eec\u5f00\u5c55\u5b9e\u9a8c\u8bc4\u4f30\u4e86\u591a\u79cd\u73b0\u6709 VLMs\uff0c\u63ed\u793a\u4e86\u5b83\u4eec\u5728\u63a8\u7406\u80fd\u529b\u65b9\u9762\u7684\u89c1\u89e3\u3002\u53e6\u5916\uff0c\u6211\u4eec\u5f00\u53d1\u4e86\u4e00\u79cd\u9ad8\u6548\u65b9\u6cd5\uff0c\u4f7f VLMs \u80fd\u591f\u5728\u7279\u5f81\u63d0\u53d6\u548c\u9884\u6d4b\u4e2d\u5229\u7528\u76ee\u6807\u7ea7\u611f\u77e5\u5143\u7d20\uff0c\u8fdb\u4e00\u6b65\u63d0\u9ad8\u4e86\u5176\u63a8\u7406\u51c6\u786e\u6027\u3002\u53ef\u6269\u5c55\u5b9e\u9a8c\u8868\u660e\uff0cReason2Drive \u5bf9\u89c6\u89c9\u63a8\u7406\u548c\u4e0b\u6e38\u89c4\u5212\u4efb\u52a1\u5177\u6709\u652f\u6301\u4f5c\u7528\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic3.zhimg.com\/v2-ccd6507c659d34b10ee81f9b5f9153a4_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic4.zhimg.com\/v2-41b2ac1e16fbcf5360820c25a027ccb3_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_28\">OmniDrive<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aOmniDrive: A Holistic Vision-Language Dataset for Autonomous Driving with Counterfactual Reasoning<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2405.01533\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2405.01533<\/a><\/li>\n\n\n\n<li>\u9879\u76ee\u4e3b\u9875\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/github.com\/NVlabs\/OmniDrive\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/github.com\/NVlabs\/OmniDrive<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1aNVIDIA\uff0c\u9999\u6e2f\u7406\u5de5\u5927\u5b66\uff0c\u5317\u4eac\u7406\u5de5\u5927\u5b66<\/li>\n<\/ul>\n\n\n\n<p>\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08VLMs\uff09\u7684\u8fdb\u6b65\u5f15\u53d1\u4e86\u4eba\u4eec\u5728\u81ea\u52a8\u9a7e\u9a76\u9886\u57df\u5bf9\u5176\u5f3a\u5927\u63a8\u7406\u80fd\u529b\u7684\u5173\u6ce8\u3002\u7136\u800c\uff0c\u5c06\u8fd9\u4e9b\u80fd\u529b\u4ece 2D \u6269\u5c55\u5230\u5168\u9762\u7684 3D \u7406\u89e3\u5bf9\u4e8e\u5b9e\u9645\u5e94\u7528\u81f3\u5173\u91cd\u8981\u3002\u4e3a\u5e94\u5bf9\u8fd9\u4e00\u6311\u6218\uff0c\u6211\u4eec\u63d0\u51fa\u4e86 OmniDrive\uff0c\u8fd9\u662f\u4e00\u4e2a\u5168\u9762\u7684\u89c6\u89c9\u8bed\u8a00\u6570\u636e\u96c6\uff0c\u901a\u8fc7\u53cd\u4e8b\u5b9e\u63a8\u7406\u4f7f\u667a\u80fd\u4f53\u6a21\u578b\u4e0e 3D \u9a7e\u9a76\u4efb\u52a1\u76f8\u5339\u914d\u3002\u8fd9\u79cd\u65b9\u6cd5\u901a\u8fc7\u8bc4\u4f30\u6f5c\u5728\u573a\u666f\u53ca\u5176\u7ed3\u679c\u6765\u589e\u5f3a\u51b3\u7b56\u80fd\u529b\uff0c\u7c7b\u4f3c\u4e8e\u4eba\u7c7b\u9a7e\u9a76\u5458\u8003\u8651\u66ff\u4ee3\u884c\u52a8\u7684\u65b9\u5f0f\u3002\u6211\u4eec\u57fa\u4e8e\u53cd\u4e8b\u5b9e\u7684\u5408\u6210\u6570\u636e\u6807\u6ce8\u8fc7\u7a0b\u751f\u6210\u4e86\u5927\u89c4\u6a21\u3001\u9ad8\u8d28\u91cf\u7684\u6570\u636e\u96c6\uff0c\u63d0\u4f9b\u4e86\u66f4\u5bc6\u96c6\u7684\u76d1\u7763\u4fe1\u53f7\uff0c\u8fde\u63a5\u4e86\u89c4\u5212\u8f68\u8ff9\u4e0e\u57fa\u4e8e\u8bed\u8a00\u7684\u63a8\u7406\u3002\u6b64\u5916\uff0c\u6211\u4eec\u63a2\u7d22\u4e86\u4e24\u4e2a\u5148\u8fdb\u7684 OmniDrive-Agent \u6846\u67b6\uff0c\u5373 Omni-L \u548c Omni-Q\uff0c\u4ee5\u8bc4\u4f30\u89c6\u89c9 &#8211; \u8bed\u8a00\u5bf9\u9f50\u4e0e 3D \u611f\u77e5\u7684\u91cd\u8981\u6027\uff0c\u63ed\u793a\u4e86\u8bbe\u8ba1\u9ad8\u6548 LLM \u667a\u80fd\u4f53\u7684\u5173\u952e\u89c1\u89e3\u3002\u5728 DriveLM \u95ee\u7b54\u57fa\u51c6\u548c nuScenes \u5f00\u73af\u89c4\u5212\u4efb\u52a1\u4e0a\u7684\u663e\u8457\u6539\u8fdb\uff0c\u8bc1\u660e\u4e86\u6211\u4eec\u7684\u6570\u636e\u96c6\u548c\u65b9\u6cd5\u7684\u6709\u6548\u6027\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pica.zhimg.com\/v2-35e2242199279605eac7234bf479585a_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic4.zhimg.com\/v2-e47c2def2c93ee5a33fd9b2b58b5fa45_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_29\">LATTE<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aLATTE: A Real-time Lightweight Attention-based Traffic Accident Anticipation Engine<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2504.04103\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2504.04103<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u6fb3\u95e8\u5927\u5b66<\/li>\n<\/ul>\n\n\n\n<p>\u51c6\u786e\u9884\u6d4b\u4ea4\u901a\u4e8b\u6545\u662f\u81ea\u52a8\u9a7e\u9a76\u9886\u57df\u7684\u4e00\u9879\u5173\u952e\u6311\u6218\uff0c\u5c24\u5176\u5728\u8d44\u6e90\u53d7\u9650\u73af\u5883\u4e2d\u3002\u73b0\u6709\u89e3\u51b3\u65b9\u6848\u5f80\u5f80\u5b58\u5728\u8ba1\u7b97\u5f00\u9500\u5927\u7684\u95ee\u9898\uff0c\u6216\u65e0\u6cd5\u5145\u5206\u5e94\u5bf9\u4ea4\u901a\u573a\u666f\u6f14\u5316\u7684\u4e0d\u786e\u5b9a\u6027\u3002\u672c\u6587\u63d0\u51fa\u4e86 LATTE\uff08\u8f7b\u91cf\u7ea7\u6ce8\u610f\u529b\u57fa\u4ea4\u901a\u4e8b\u6545\u9884\u6d4b\u5f15\u64ce\uff09\uff0c\u8be5\u5f15\u64ce\u878d\u5408\u4e86\u8ba1\u7b97\u6548\u7387\u4e0e\u6700\u5148\u8fdb\u6027\u80fd\u3002LATTE \u91c7\u7528\u9ad8\u6548\u591a\u5c3a\u5ea6\u7a7a\u95f4\u805a\u5408\uff08EMSA\uff09\u6355\u6349\u8de8\u5c3a\u5ea6\u7a7a\u95f4\u7279\u5f81\uff0c\u8bb0\u5fc6\u6ce8\u610f\u529b\u805a\u5408\uff08MAA\uff09\u589e\u5f3a\u65f6\u5e8f\u5efa\u6a21\uff0c\u8f85\u52a9\u81ea\u6ce8\u610f\u529b\u805a\u5408\uff08AAA\uff09\u63d0\u53d6\u957f\u5e8f\u5217\u6f5c\u5728\u4f9d\u8d56\u5173\u7cfb\u3002\u6b64\u5916\uff0cLATTE \u96c6\u6210\u4e86\u706b\u70c8\u9e1f\u8b66\u62a5\u8f85\u52a9\u7cfb\u7edf\uff08FAA\uff09\uff0c\u5229\u7528\u89c6\u89c9 &#8211; \u8bed\u8a00\u6a21\u578b\u63d0\u4f9b\u5b9e\u65f6\u3001\u6613\u4e8e\u7406\u89e3\u7684\u8bed\u97f3\u5371\u9669\u8b66\u62a5\uff0c\u63d0\u5347\u4e58\u5ba2\u60c5\u5883\u611f\u77e5\u80fd\u529b\u3002\u5728\u57fa\u51c6\u6570\u636e\u96c6\uff08DAD\u3001CCD\u3001A3D\uff09\u4e0a\u7684\u8bc4\u4f30\u8868\u660e\uff0cLATTE \u5177\u6709\u5353\u8d8a\u7684\u9884\u6d4b\u80fd\u529b\u548c\u8ba1\u7b97\u6548\u7387\u3002\u5728 DAD \u57fa\u51c6\u4e0a\uff0cLATTE \u5b9e\u73b0\u4e86 89.74% \u7684\u5e73\u5747\u7cbe\u5ea6\uff08AP\uff09\uff0c\u5e73\u5747\u4e8b\u6545\u63d0\u524d\u65f6\u95f4\uff08mTTA\uff09\u6bd4\u7b2c\u4e8c\u597d\u7684\u6a21\u578b\u9ad8 5.4%\uff0c\u5728\u53ec\u56de\u7387\u4e3a 80% \u65f6\u4fdd\u6301\u6709\u7ade\u4e89\u529b\u7684 mTTA\uff08TTA@R80 \u4e3a 4.04 \u79d2\uff09\uff0c\u4e14\u5728\u4e0d\u540c\u9a7e\u9a76\u6761\u4ef6\u4e0b\u5747\u80fd\u7a33\u5065\u9884\u6d4b\u4e8b\u6545\u3002\u5176\u8f7b\u91cf\u7ea7\u8bbe\u8ba1\u4f7f\u6d6e\u70b9\u8fd0\u7b97\uff08FLOPs\uff09\u51cf\u5c11 93.14%\uff0c\u53c2\u6570\u6570\u91cf\uff08Params\uff09\u51cf\u5c11 31.58%\uff0c\u53ef\u5728\u8d44\u6e90\u6709\u9650\u7684\u786c\u4ef6\u4e0a\u5b9e\u73b0\u5b9e\u65f6\u8fd0\u884c\u4e14\u4e0d\u635f\u5931\u6027\u80fd\u3002\u6d88\u878d\u7814\u7a76\u8bc1\u5b9e\u4e86 LATTE \u5404\u67b6\u6784\u7ec4\u4ef6\u7684\u6709\u6548\u6027\uff0c\u53ef\u89c6\u5316\u548c\u5931\u8d25\u6848\u4f8b\u5206\u6790\u7a81\u663e\u4e86\u5176\u5b9e\u7528\u6027\u53ca\u9700\u6539\u8fdb\u7684\u65b9\u5411\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/picx.zhimg.com\/v2-4fb08d4b808365e7f77cf3aa27929b6d_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/picx.zhimg.com\/v2-1d946cde3335f831c1335f752127806b_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_30\">EM-VLM4AD<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aMulti-Frame, Lightweight &amp; Efficient Vision-Language Models for Question Answering in Autonomous Driving<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2403.19838\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2403.19838<\/a><\/li>\n\n\n\n<li>\u9879\u76ee\u4e3b\u9875\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/github.com\/akshaygopalkr\/EM-VLM4AD\/tree\/main\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/github.com\/akshaygopalkr\/EM-VLM4AD\/tree\/main<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u52a0\u5229\u798f\u5c3c\u4e9a\u5927\u5b66\u5723\u8fed\u6208\u5206\u6821<\/li>\n<\/ul>\n\n\n\n<p>\u89c6\u89c9 &#8211; \u8bed\u8a00\u6a21\u578b\uff08VLMs\uff09\u548c\u591a\u6a21\u6001\u8bed\u8a00\u6a21\u578b\uff08MMLMs\uff09\u5728\u81ea\u52a8\u9a7e\u9a76\u7814\u7a76\u4e2d\u5df2\u5360\u636e\u91cd\u8981\u5730\u4f4d\uff0c\u8fd9\u4e9b\u6a21\u578b\u80fd\u591f\u5229\u7528\u4ea4\u901a\u573a\u666f\u56fe\u50cf\u548c\u5176\u4ed6\u6570\u636e\u6a21\u6001\uff0c\u4e3a\u7aef\u5230\u7aef\u81ea\u52a8\u9a7e\u9a76\u5b89\u5168\u4efb\u52a1\u63d0\u4f9b\u53ef\u89e3\u91ca\u7684\u6587\u672c\u63a8\u7406\u548c\u54cd\u5e94\u3002\u7136\u800c\uff0c\u5f53\u524d\u8fd9\u4e9b\u7cfb\u7edf\u7684\u5b9e\u73b0\u65b9\u6848\u91c7\u7528\u4e86\u6602\u8d35\u7684\u5927\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u9aa8\u5e72\u7f51\u7edc\u548c\u56fe\u50cf\u7f16\u7801\u5668\uff0c\u4f7f\u5f97\u8fd9\u7c7b\u7cfb\u7edf\u4e0d\u9002\u7528\u4e8e\u5b58\u5728\u4e25\u683c\u5185\u5b58\u9650\u5236\u4e14\u9700\u8981\u5feb\u901f\u63a8\u7406\u65f6\u95f4\u7684\u5b9e\u65f6\u81ea\u52a8\u9a7e\u9a76\u7cfb\u7edf\u3002\u4e3a\u89e3\u51b3\u4e0a\u8ff0\u95ee\u9898\uff0c\u6211\u4eec\u5f00\u53d1\u4e86 EM-VLM4AD\uff0c\u8fd9\u662f\u4e00\u79cd\u9ad8\u6548\u3001\u8f7b\u91cf\u7684\u591a\u5e27\u89c6\u89c9 &#8211; \u8bed\u8a00\u6a21\u578b\uff0c\u7528\u4e8e\u6267\u884c\u81ea\u52a8\u9a7e\u9a76\u9886\u57df\u7684\u89c6\u89c9\u95ee\u7b54\u4efb\u52a1\u3002\u4e0e\u73b0\u6709\u65b9\u6cd5\u76f8\u6bd4\uff0cEM-VLM4AD \u6240\u9700\u5185\u5b58\u548c\u6d6e\u70b9\u8fd0\u7b97\u81f3\u5c11\u51cf\u5c11 10 \u500d\uff0c\u540c\u65f6\u5728 DriveLM \u6570\u636e\u96c6\u4e0a\u7684 CIDEr \u548c ROUGE \u8bc4\u5206\u4e5f\u9ad8\u4e8e\u73b0\u6709\u57fa\u7ebf\u3002\u6b64\u5916\uff0cEM-VLM4AD \u8fd8\u80fd\u591f\u4ece\u4e0e\u63d0\u793a\u76f8\u5173\u7684\u4ea4\u901a\u89c6\u56fe\u4e2d\u63d0\u53d6\u5173\u952e\u4fe1\u606f\uff0c\u5e76\u80fd\u56de\u7b54\u5404\u79cd\u81ea\u52a8\u9a7e\u9a76\u5b50\u4efb\u52a1\u7684\u76f8\u5173\u95ee\u9898\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic3.zhimg.com\/v2-b6374ee3df1436389057a9aae682148e_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic3.zhimg.com\/v2-dadb5686c8846eb2ccf316665be69a9a_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_31\">Is it safe to cross<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aIs it safe to cross? Interpretable Risk Assessment with GPT-4V for Safety-Aware Street Crossing<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/ieeexplore.ieee.org\/stamp\/stamp.jsp%3Farnumber%3D10597464\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/ieeexplore.ieee.org\/stamp\/stamp.jsp?arnumber=10597464<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u9a6c\u8428\u8bf8\u585e\u5927\u5b66\u963f\u9ed8\u65af\u7279\u5206\u6821<\/li>\n<\/ul>\n\n\n\n<p>\u5b89\u5168\u7a7f\u8d8a\u8857\u9053\u4ea4\u53c9\u53e3\u5bf9\u76f2\u4eba\u800c\u8a00\u662f\u4e00\u9879\u590d\u6742\u6311\u6218\uff0c\u8fd9\u8981\u6c42\u5bf9\u5468\u8fb9\u73af\u5883\u8fdb\u884c\u7cbe\u7ec6\u611f\u77e5\u2014\u2014\u800c\u8be5\u4efb\u52a1\u9ad8\u5ea6\u4f9d\u8d56\u89c6\u89c9\u7ebf\u7d22\u3002\u4f20\u7edf\u7684\u51b3\u7b56\u8f85\u52a9\u65b9\u6cd5\u5f80\u5f80\u5b58\u5728\u5c40\u9650\uff0c\u65e0\u6cd5\u63d0\u4f9b\u5168\u9762\u7684\u573a\u666f\u5206\u6790\u4e0e\u5b89\u5168\u7b49\u7ea7\u8bc4\u4f30\u3002\u672c\u6587\u63d0\u51fa\u4e00\u79cd\u521b\u65b0\u65b9\u6848\uff0c\u5229\u7528\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u89e3\u6790\u590d\u6742\u7684\u8fc7\u8857\u573a\u666f\uff0c\u8f83\u4e4b\u4f20\u7edf\u4ea4\u901a\u4fe1\u53f7\u8bc6\u522b\u6280\u672f\u5177\u6709\u663e\u8457\u8fdb\u6b65\u3002\u901a\u8fc7\u751f\u6210\u81ea\u7136\u8bed\u8a00\u63cf\u8ff0\u7684\u5b89\u5168\u8bc4\u5206\u4e0e\u573a\u666f\u89e3\u6790\uff0c\u672c\u65b9\u6cd5\u53ef\u4e3a\u89c6\u969c\u7fa4\u4f53\u63d0\u4f9b\u5b89\u5168\u51b3\u7b56\u652f\u6301\u3002\u6211\u4eec\u91c7\u96c6\u4e86\u5305\u542b\u56db\u8db3\u673a\u5668\u4eba\u591a\u89c6\u89d2\u7b2c\u4e00\u4eba\u79f0\u56fe\u50cf\u7684\u6591\u9a6c\u7ebf\u4ea4\u53c9\u53e3\u6570\u636e\u96c6\uff0c\u5e76\u4f9d\u636e\u9884\u8bbe\u7684\u5b89\u5168\u8bc4\u5206\u5206\u7ea7\u6807\u51c6\u5bf9\u56fe\u50cf\u8fdb\u884c\u6807\u6ce8\u8bc4\u5206\u3002\u57fa\u4e8e\u4ece\u56fe\u50cf\u548c\u6587\u672c\u63d0\u793a\u4e2d\u63d0\u53d6\u7684\u89c6\u89c9\u77e5\u8bc6\uff0c\u6211\u4eec\u8bc4\u4f30\u4e86\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u5728\u5b89\u5168\u8bc4\u5206\u9884\u6d4b\u4e0e\u573a\u666f\u63cf\u8ff0\u65b9\u9762\u7684\u6027\u80fd\u3002\u7814\u7a76\u8868\u660e\uff0c\u8be5\u6a21\u578b\u901a\u8fc7\u591a\u6837\u5316\u7684\u63d0\u793a\u8bcd\u6fc0\u6d3b\u7684\u63a8\u7406\u80fd\u529b\u4e0e\u5b89\u5168\u8bc4\u5206\u9884\u6d4b\u673a\u5236\uff0c\u4e3a\u5f00\u53d1\u53ef\u4fe1\u8d56\u51b3\u7b56\u652f\u6301\u7cfb\u7edf\u63d0\u4f9b\u4e86\u5173\u952e\u8def\u5f84\uff0c\u8fd9\u5bf9\u9700\u8981\u9ad8\u53ef\u9760\u6027\u51b3\u7b56\u8f85\u52a9\u7684\u5e94\u7528\u573a\u666f\u81f3\u5173\u91cd\u8981\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_32\">NuScenes-MQA<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aNuScenes-MQA: Integrated Evaluation of Captions and QA for Autonomous Driving Datasets using Markup Annotations<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/ieeexplore.ieee.org\/abstract\/document\/10495633\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/ieeexplore.ieee.org\/abstract\/document\/10495633<\/a><\/li>\n\n\n\n<li>\u6570\u636e\u96c6\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/github.com\/turingmotors\/NuScenes-MQA\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/github.com\/turingmotors\/NuScenes-MQA<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u56fe\u7075<\/li>\n<\/ul>\n\n\n\n<p>\u89c6\u89c9\u95ee\u7b54\uff08VQA\uff09\u4f5c\u4e3a\u81ea\u52a8\u9a7e\u9a76\u9886\u57df\u7684\u5173\u952e\u6280\u672f\uff0c\u9700\u5b9e\u73b0\u7cbe\u51c6\u73af\u5883\u8bc6\u522b\u4e0e\u590d\u6742\u573a\u666f\u8bc4\u4f30\u3002\u7136\u800c\uff0c\u76ee\u524d\u5c1a\u7f3a\u4e4f\u57fa\u4e8e\u884c\u8f66\u573a\u666f\u6784\u5efa\u7684\u95ee\u7b54\u683c\u5f0f\u6807\u6ce8\u6570\u636e\u96c6\uff0c\u6b64\u7c7b\u6570\u636e\u96c6\u5bf9\u4fdd\u8bc1\u8bed\u8a00\u751f\u6210\u7cbe\u786e\u6027\u4e0e\u573a\u666f\u8bc6\u522b\u80fd\u529b\u81f3\u5173\u91cd\u8981\u3002\u672c\u7814\u7a76\u63d0\u51fa\u521b\u65b0\u6027\u6807\u6ce8\u6280\u672fMarkup-QA\u2014\u2014\u901a\u8fc7\u6807\u8bb0\u8bed\u8a00\u5c01\u88c5\u95ee\u7b54\u5bf9\uff0c\u8be5\u6846\u67b6\u53ef\u540c\u6b65\u8bc4\u4f30\u6a21\u578b\u5728\u8bed\u53e5\u751f\u6210\u4e0e\u89c6\u89c9\u95ee\u7b54\u7684\u53cc\u91cd\u80fd\u529b\u3002\u57fa\u4e8e\u6b64\u6807\u6ce8\u65b9\u6cd5\uff0c\u6211\u4eec\u6784\u5efa\u4e86NuScenes-MQA\u6570\u636e\u96c6\uff0c\u805a\u7126\u63cf\u8ff0\u80fd\u529b\u4e0e\u7cbe\u51c6\u95ee\u7b54\u7684\u53cc\u91cd\u7279\u6027\uff0c\u4e3a\u81ea\u52a8\u9a7e\u9a76\u4efb\u52a1\u7684\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u5f00\u53d1\u63d0\u4f9b\u65b0\u8303\u672c\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_33\">LLM Multimodal Traffic Accident Forecasting<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aLLM Multimodal Traffic Accident Forecasting<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/www.mdpi.com\/1424-8220\/23\/22\/9225\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/www.mdpi.com\/1424-8220\/23\/22\/9225<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u6b4c\u5fb7\u5927\u5b66\uff0c\u74e6\u4f26\u897f\u4e9a\u7406\u5de5\u5927\u5b66\u7b49<\/li>\n<\/ul>\n\n\n\n<p>\u968f\u7740\u57ce\u5e02\u4e2d\u5fc3\u4ea4\u901a\u62e5\u5835\u65e5\u76ca\u52a0\u5267\uff0c\u4ea4\u901a\u4e8b\u6545\u9884\u6d4b\u5bf9\u4e8e\u57ce\u5e02\u89c4\u5212\u548c\u516c\u5171\u5b89\u5168\u5df2\u53d8\u5f97\u81f3\u5173\u91cd\u8981\u3002\u672c\u6587\u7cfb\u7edf\u8bc4\u4f30\u4e86\u73b0\u4ee3\u6df1\u5ea6\u5b66\u4e60\uff08DL\uff09\u65b9\u6cd5\u5728\u9884\u6d4b\u4ea4\u901a\u4e8b\u6545\u4ee5\u53ca\u901a\u8fc7\u53ef\u64cd\u4f5c\u7684\u89c6\u542c\u63d0\u793a\u589e\u5f3aL4\/L5\u7ea7\u9a7e\u9a76\u8f85\u52a9\u7cfb\u7edf\u65b9\u9762\u7684\u6548\u80fd\u3002\u57fa\u4e8e\u8be6\u7ec6\u8bb0\u5f55\u4ea4\u901a\u4e8b\u6545\u53d1\u751f\u60c5\u51b5\u7684\u4e30\u5bcc\u6570\u636e\u96c6\uff0c\u6211\u4eec\u5bf9\u6bd4\u9a8c\u8bc1\u4e86Transformer\u6a21\u578b\u4e0e\u4f20\u7edf\u65f6\u95f4\u5e8f\u5217\u6a21\u578b\uff08\u5982ARIMA\uff09\u53ca\u8f83\u65b0\u7684Prophet\u6a21\u578b\u7684\u6027\u80fd\u3002\u6b64\u5916\uff0c\u901a\u8fc7\u7ec6\u81f4\u7684\u5206\u6790\uff0c\u6211\u4eec\u8fd0\u7528\u4e3b\u6210\u5206\u5206\u6790\uff08PCA\uff09\u8f7d\u8377\u6df1\u5165\u63a2\u7a76\u4e86\u7279\u5f81\u91cd\u8981\u6027\uff0c\u63ed\u793a\u4e86\u5bfc\u81f4\u4e8b\u6545\u7684\u5173\u952e\u56e0\u7d20\u3002\u6211\u4eec\u521b\u65b0\u6027\u5730\u63d0\u51fa\u5728\u81ea\u52a8\u9a7e\u9a76\u4e2d\u5229\u7528\u8f7b\u91cf\u7ea7\u7d27\u51d1\u578b\u5927\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\uff08\u5982LLaMA-2\u548cZephyr-7b- \uff09\u8fdb\u884c\u5b9e\u65f6\u5e72\u9884\u7684\u601d\u8def\u3002\u6211\u4eec\u7684\u63a2\u7d22\u8fd8\u5ef6\u4f38\u81f3\u591a\u6a21\u6001\u9886\u57df\uff1a\u901a\u8fc7\u7ed3\u5408\u5927\u578b\u8bed\u8a00\u89c6\u89c9\u52a9\u624b\uff08LLaVA\uff09\u2014\u2014\u4e00\u79cd\u5229\u7528\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08VLM\uff09\u6865\u63a5\u89c6\u89c9\u4e0e\u8bed\u8a00\u63d0\u793a\u7684\u6280\u672f\u2014\u2014\u4e0e\u6df1\u5ea6\u6982\u7387\u63a8\u7406\uff0c\u63d0\u5347\u4e86\u81ea\u52a8\u9a7e\u9a76\u7cfb\u7edf\u7684\u5b9e\u65f6\u54cd\u5e94\u80fd\u529b\u3002\u672c\u7814\u7a76\u9610\u660e\u4e86\u5728\u6df1\u5ea6\u5b66\u4e60\u548c\u6df1\u5ea6\u6982\u7387\u7f16\u7a0b\u4e2d\u8fd0\u7528\u5927\u578b\u591a\u6a21\u6001\u6a21\u578b\uff0c\u5bf9\u4e8e\u63d0\u5347\u65f6\u95f4\u5e8f\u5217\u9884\u6d4b\u53ca\u7279\u5f81\u6743\u91cd\u91cd\u8981\u6027\u7684\u6027\u80fd\u548c\u5b9e\u7528\u6027\u7684\u4f18\u52bf\uff0c\u5c24\u5176\u5728\u81ea\u52a8\u9a7e\u9a76\u573a\u666f\u4e2d\u3002\u6b64\u9879\u5de5\u4f5c\u4e3a\u6784\u5efa\u4f9d\u6258\u6570\u636e\u9a71\u52a8\u51b3\u7b56\u7684\u66f4\u5b89\u5168\u3001\u66f4\u667a\u80fd\u7684\u57ce\u5e02\u5960\u5b9a\u4e86\u91cd\u8981\u57fa\u7840\u3002<\/p>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic1.zhimg.com\/v2-c98db49a1a7856f990fb1c749c5678c6_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic1.zhimg.com\/v2-870d3c6c57c91b25030fde0e4a314f04_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_34\">Talk2BEV<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aTalk2BEV: Language-enhanced Bird&#8217;s-eye View Maps for Autonomous Driving<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2310.02251\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2310.02251<\/a><\/li>\n\n\n\n<li>\u9879\u76ee\u4e3b\u9875\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/llmbev.github.io\/talk2bev\/\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/llmbev.github.io\/talk2bev\/<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u9ebb\u7701\u7406\u5de5\u5b66\u9662\uff0c\u4e0d\u5217\u98a0\u54e5\u4f26\u6bd4\u4e9a\u5927\u5b66\u3001\u5854\u5c14\u56fe\u5927\u5b66\u7b49<\/li>\n<\/ul>\n\n\n\n<p>\u672c\u6587\u4ecb\u7ecd\u4e86 Talk2BEV\uff0c\u8fd9\u662f\u4e00\u79cd\u7528\u4e8e\u81ea\u52a8\u9a7e\u9a76\u573a\u666f\u4e0bBEV\u5730\u56fe\u7684\u5927\u578b\u89c6\u89c9 &#8211; \u8bed\u8a00\u6a21\u578b\uff08LVLM\uff09\u63a5\u53e3\u3002\u73b0\u6709\u81ea\u52a8\u9a7e\u9a76\u611f\u77e5\u7cfb\u7edf\u4e3b\u8981\u805a\u7126\u4e8e\u9884\u5b9a\u4e49\uff08\u5c01\u95ed\uff09\u7684\u5bf9\u8c61\u7c7b\u522b\u548c\u9a7e\u9a76\u573a\u666f\uff0c\u800c Talk2BEV \u5c06\u901a\u7528\u8bed\u8a00\u548c\u89c6\u89c9\u6a21\u578b\u7684\u6700\u65b0\u8fdb\u5c55\u4e0e BEV \u7ed3\u6784\u7684\u5730\u56fe\u8868\u793a\u76f8\u7ed3\u5408\uff0c\u65e0\u9700\u7279\u5b9a\u4efb\u52a1\u6a21\u578b\u3002\u8fd9\u4f7f\u5f97\u5355\u4e00\u7cfb\u7edf\u80fd\u591f\u5e94\u5bf9\u591a\u79cd\u81ea\u52a8\u9a7e\u9a76\u4efb\u52a1\uff0c\u5305\u62ec\u89c6\u89c9\u4e0e\u7a7a\u95f4\u63a8\u7406\u3001\u9884\u6d4b\u4ea4\u901a\u53c2\u4e0e\u8005\u610f\u56fe\u4ee5\u53ca\u57fa\u4e8e\u89c6\u89c9\u7ebf\u7d22\u7684\u51b3\u7b56\u3002\u6211\u4eec\u5728\u5927\u91cf\u573a\u666f\u7406\u89e3\u4efb\u52a1\u4e0a\u5bf9 Talk2BEV \u8fdb\u884c\u4e86\u5e7f\u6cdb\u8bc4\u4f30\uff0c\u8fd9\u4e9b\u4efb\u52a1\u65e2\u4f9d\u8d56\u4e8e\u89e3\u91ca\u81ea\u7531\u5f62\u5f0f\u81ea\u7136\u8bed\u8a00\u67e5\u8be2\u7684\u80fd\u529b\uff0c\u4e5f\u4f9d\u8d56\u4e8e\u5c06\u8fd9\u4e9b\u67e5\u8be2\u4e0e\u5d4c\u5165\u8bed\u8a00\u589e\u5f3a BEV \u5730\u56fe\u4e2d\u7684\u89c6\u89c9\u4e0a\u4e0b\u6587\u76f8\u5173\u8054\u7684\u80fd\u529b\u3002\u4e3a\u4e86\u63a8\u52a8\u5927\u578b\u89c6\u89c9 &#8211; \u8bed\u8a00\u6a21\u578b\u5728\u81ea\u52a8\u9a7e\u9a76\u573a\u666f\u4e2d\u7684\u8fdb\u4e00\u6b65\u7814\u7a76\uff0c\u6211\u4eec\u5f00\u53d1\u5e76\u53d1\u5e03\u4e86 Talk2BEV-Bench \u57fa\u51c6\uff0c\u8be5\u57fa\u51c6\u5305\u542b 1000 \u4e2a\u4eba\u5de5\u6807\u6ce8\u7684 BEV \u573a\u666f\uff0c\u4ee5\u53ca\u6765\u81ea NuScenes \u6570\u636e\u96c6\u7684 20000 \u591a\u4e2a\u95ee\u9898\u548c\u771f\u5b9e\u6807\u6ce8\u7b54\u6848\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic3.zhimg.com\/v2-814c7ccbbcf6fdd23336c79865255fbe_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pica.zhimg.com\/v2-49e990c8f2a0c391d5cc96f67bd77026_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_35\">On the Road with GPT-4V<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aOn the Road with GPT-4V(ision): Early Explorations of Visual-Language Model on Autonomous Driving<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2311.05332\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2311.05332<\/a><\/li>\n\n\n\n<li>\u9879\u76ee\u4e3b\u9875\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/github.com\/PJLab-ADG\/GPT4V-AD-Exploration\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/github.com\/PJLab-ADG\/GPT4V-AD-Exploration<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u4e0a\u6d77\u4eba\u5de5\u667a\u80fd\u5b9e\u9a8c\u5ba4\u3001\u534e\u4e1c\u5e08\u8303\u5927\u5b66\u3001\u9999\u6e2f\u4e2d\u6587\u5927\u5b66\u7b49<\/li>\n<\/ul>\n\n\n\n<p>\u81ea\u52a8\u9a7e\u9a76\u6280\u672f\u7684\u8ffd\u6c42\u53d6\u51b3\u4e8e\u611f\u77e5\u3001\u51b3\u7b56\u548c\u63a7\u5236\u7cfb\u7edf\u7684\u590d\u6742\u6574\u5408\u3002\u65e0\u8bba\u662f\u6570\u636e\u9a71\u52a8\u8fd8\u662f\u57fa\u4e8e\u89c4\u5219\u7684\u4f20\u7edf\u65b9\u6cd5\uff0c\u90fd\u56e0\u65e0\u6cd5\u628a\u63e1\u590d\u6742\u9a7e\u9a76\u73af\u5883\u7684\u7ec6\u5fae\u5dee\u522b\u4ee5\u53ca\u5176\u4ed6\u9053\u8def\u4f7f\u7528\u8005\u7684\u610f\u56fe\u800c\u53d7\u963b\u3002\u8fd9\u4e00\u70b9\u5df2\u6210\u4e3a\u663e\u8457\u74f6\u9888\uff0c\u5c24\u5176\u4f53\u73b0\u5728\u5f00\u53d1\u5b89\u5168\u53ef\u9760\u7684\u81ea\u52a8\u9a7e\u9a76\u6240\u9700\u7684\u5e38\u8bc6\u63a8\u7406\u548c\u7cbe\u7ec6\u573a\u666f\u7406\u89e3\u65b9\u9762\u3002\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08VLM\uff09\u7684\u51fa\u73b0\u4e3a\u5b9e\u73b0\u5b8c\u5168\u81ea\u52a8\u9a7e\u9a76\u5f00\u8f9f\u4e86\u65b0\u9886\u57df\u3002\u672c\u62a5\u544a\u5bf9\u6700\u5148\u8fdb\u7684 VLM\u2014\u2014GPT-4V\uff08\u89c6\u89c9\u7248\uff09\u53ca\u5176\u5728\u81ea\u52a8\u9a7e\u9a76\u573a\u666f\u4e2d\u7684\u5e94\u7528\u8fdb\u884c\u4e86\u5168\u9762\u8bc4\u4f30\u3002\u6211\u4eec\u63a2\u7a76\u4e86\u8be5\u6a21\u578b\u7406\u89e3\u548c\u63a8\u7406\u9a7e\u9a76\u573a\u666f\u3001\u505a\u51fa\u51b3\u7b56\u5e76\u6700\u7ec8\u4ee5\u9a7e\u9a76\u5458\u8eab\u4efd\u91c7\u53d6\u884c\u52a8\u7684\u80fd\u529b\u3002\u6211\u4eec\u7684\u7efc\u5408\u6d4b\u8bd5\u6db5\u76d6\u4e86\u4ece\u57fa\u672c\u573a\u666f\u8bc6\u522b\u5230\u590d\u6742\u56e0\u679c\u63a8\u7406\uff0c\u4ee5\u53ca\u4e0d\u540c\u6761\u4ef6\u4e0b\u7684\u5b9e\u65f6\u51b3\u7b56\u3002\u7814\u7a76\u7ed3\u679c\u8868\u660e\uff0c\u4e0e\u73b0\u6709\u81ea\u52a8\u9a7e\u9a76\u7cfb\u7edf\u76f8\u6bd4\uff0cGPT-4V \u5728\u573a\u666f\u7406\u89e3\u548c\u56e0\u679c\u63a8\u7406\u65b9\u9762\u8868\u73b0\u66f4\u4f18\u3002\u5b83\u5c55\u793a\u51fa\u5904\u7406\u5206\u5e03\u5916\u573a\u666f\u3001\u8bc6\u522b\u610f\u56fe\u4ee5\u53ca\u5728\u771f\u5b9e\u9a7e\u9a76\u73af\u5883\u4e2d\u505a\u51fa\u660e\u667a\u51b3\u7b56\u7684\u6f5c\u529b\u3002\u7136\u800c\uff0c\u6311\u6218\u4f9d\u7136\u5b58\u5728\uff0c\u5c24\u5176\u5728\u65b9\u5411\u8fa8\u522b\u3001\u4ea4\u901a\u706f\u8bc6\u522b\u3001\u89c6\u89c9\u5b9a\u4f4d\u548c\u7a7a\u95f4\u63a8\u7406\u4efb\u52a1\u4e2d\u3002\u8fd9\u4e9b\u5c40\u9650\u6027\u51f8\u663e\u4e86\u8fdb\u4e00\u6b65\u7814\u53d1\u7684\u5fc5\u8981\u6027\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic3.zhimg.com\/v2-8bab991db1323c7248d753ee70735edc_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_36\">OpenAnnotate3D<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aOpenAnnotate3D: Open-Vocabulary Auto-Labeling System for Multi-modal 3D Data<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2310.13398\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2310.13398<\/a><\/li>\n\n\n\n<li>\u9879\u76ee\u4e3b\u9875\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/github.com\/Fudan-ProjectTitan\/OpenAnnotate3D\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/github.com\/Fudan-ProjectTitan\/OpenAnnotate3D<\/a><\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1a\u590d\u65e6\u5927\u5b66\uff0c\u591a\u4f26\u591a\u5927\u5b66<\/li>\n<\/ul>\n\n\n\n<p>\u5728\u5927\u6570\u636e\u548c\u5927\u6a21\u578b\u65f6\u4ee3\uff0c\u591a\u6a21\u6001\u6570\u636e\u7684\u81ea\u52a8\u6807\u6ce8\u529f\u80fd\u5bf9\u4e8e\u73b0\u5b9e\u4e16\u754c\u4e2d\u4eba\u5de5\u667a\u80fd\u9a71\u52a8\u7684\u5e94\u7528\uff08\u5982\u81ea\u52a8\u9a7e\u9a76\u548c\u5177\u8eab\u667a\u80fd\uff09\u5177\u6709\u91cd\u8981\u610f\u4e49\u3002\u4e0e\u4f20\u7edf\u7684\u95ed\u96c6\u6807\u6ce8\u4e0d\u540c\uff0c\u5f00\u653e\u8bcd\u6c47\u6807\u6ce8\u662f\u5b9e\u73b0\u4eba\u7c7b\u7ea7\u8ba4\u77e5\u80fd\u529b\u7684\u5173\u952e\u3002\u7136\u800c\uff0c\u9488\u5bf9\u591a\u6a21\u6001 3D \u6570\u636e\u7684\u5f00\u653e\u8bcd\u6c47\u81ea\u52a8\u6807\u6ce8\u7cfb\u7edf\u8f83\u5c11\u3002\u5728\u672c\u6587\u4e2d\uff0c\u6211\u4eec\u4ecb\u7ecd OpenAnnotate3D\uff0c\u8fd9\u662f\u4e00\u4e2a\u5f00\u6e90\u7684\u5f00\u653e\u8bcd\u6c47\u81ea\u52a8\u6807\u6ce8\u7cfb\u7edf\uff0c\u80fd\u591f\u4e3a\u89c6\u89c9\u548c\u70b9\u4e91\u6570\u636e\u81ea\u52a8\u751f\u6210 2D \u63a9\u7801\u30013D \u63a9\u7801\u548c 3D \u8fb9\u754c\u6846\u6807\u6ce8\u3002\u8be5\u7cfb\u7edf\u6574\u5408\u4e86\u5927\u8bed\u8a00\u6a21\u578b\uff08LLMs\uff09\u7684\u601d\u7ef4\u94fe\u80fd\u529b\u548c\u89c6\u89c9 &#8211; \u8bed\u8a00\u6a21\u578b\uff08VLMs\uff09\u7684\u8de8\u6a21\u6001\u80fd\u529b\u3002\u636e\u6211\u4eec\u6240\u77e5\uff0cOpenAnnotate3D \u662f\u5f00\u653e\u8bcd\u6c47\u591a\u6a21\u6001 3D \u81ea\u52a8\u6807\u6ce8\u9886\u57df\u7684\u5f00\u521b\u6027\u5de5\u4f5c\u4e4b\u4e00\u3002\u6211\u4eec\u5728\u516c\u5171\u6570\u636e\u96c6\u548c\u5185\u90e8\u771f\u5b9e\u4e16\u754c\u6570\u636e\u96c6\u4e0a\u8fdb\u884c\u4e86\u5168\u9762\u8bc4\u4f30\uff0c\u7ed3\u679c\u8868\u660e\uff0c\u4e0e\u4eba\u5de5\u6807\u6ce8\u76f8\u6bd4\uff0c\u8be5\u7cfb\u7edf\u663e\u8457\u63d0\u9ad8\u4e86\u6807\u6ce8\u6548\u7387\uff0c\u540c\u65f6\u63d0\u4f9b\u4e86\u51c6\u786e\u7684\u5f00\u653e\u8bcd\u6c47\u81ea\u52a8\u6807\u6ce8\u7ed3\u679c\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic1.zhimg.com\/v2-8a995f46f8323202759630429cc437a0_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pica.zhimg.com\/v2-6dc8e1260034a1306a9fd2ab5927f2a6_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1936719591561724387_37\">Unsupervised 3D Perception<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6807\u9898\uff1aUnsupervised 3D Perception with 2D Vision-Language Distillation for Autonomous Driving<\/li>\n\n\n\n<li>\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2309.14491\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2309.14491<\/a><\/li>\n\n\n\n<li>\u9879\u76ee\u4e3b\u9875\uff1a<\/li>\n\n\n\n<li>\u4f5c\u8005\u5355\u4f4d\uff1aWaymoLLC<\/li>\n<\/ul>\n\n\n\n<p>ICCV 2023\u4e2d\u7a3f\u7684\u5de5\u4f5c\uff0c\u95ed\u96c63D\u611f\u77e5\u6a21\u578b\u4ec5\u5728\u9884\u5b9a\u4e49\u7684\u7269\u4f53\u7c7b\u522b\u96c6\u4e0a\u8bad\u7ec3\uff0c\u5bf9\u4e8e\u81ea\u52a8\u9a7e\u9a76\u7b49\u5b89\u5168\u5173\u952e\u5e94\u7528\u800c\u8a00\u53ef\u80fd\u5b58\u5728\u4e0d\u8db3\uff0c\u56e0\u4e3a\u90e8\u7f72\u540e\u53ef\u80fd\u4f1a\u9047\u5230\u65b0\u7684\u7269\u4f53\u7c7b\u578b\u3002\u672c\u6587\u63d0\u51fa\u4e86\u4e00\u79cd\u591a\u6a21\u6001\u81ea\u52a8\u6807\u6ce8\u6d41\u7a0b\uff0c\u80fd\u591f\u751f\u6210\u65e0\u6a21\u6001 3D \u8fb9\u754c\u6846\u548c\u8f68\u8ff9\u7247\u6bb5\uff0c\u7528\u4e8e\u5728\u65e0 3D \u4eba\u5de5\u6807\u6ce8\u7684\u60c5\u51b5\u4e0b\u8bad\u7ec3\u5f00\u653e\u96c6\u7c7b\u522b\u7684\u6a21\u578b\u3002\u6211\u4eec\u7684\u6d41\u7a0b\u5229\u7528\u70b9\u4e91\u5e8f\u5217\u4e2d\u56fa\u6709\u7684\u8fd0\u52a8\u7ebf\u7d22\uff0c\u7ed3\u5408\u53ef\u514d\u8d39\u83b7\u53d6\u7684 2D \u56fe\u50cf &#8211; \u6587\u672c\u5bf9\uff0c\u6765\u8bc6\u522b\u548c\u8ddf\u8e2a\u6240\u6709\u4ea4\u901a\u53c2\u4e0e\u8005\u3002\u4e0e\u8be5\u9886\u57df\u8fd1\u671f\u7684\u7814\u7a76\u76f8\u6bd4\uff0c\u90a3\u4e9b\u7814\u7a76\u4ec5\u80fd\u63d0\u4f9b\u9650\u4e8e\u79fb\u52a8\u7269\u4f53\u7684\u65e0\u7c7b\u522b\u81ea\u52a8\u6807\u6ce8\uff0c\u800c\u6211\u4eec\u7684\u65b9\u6cd5\u80fd\u591f\u4ee5\u65e0\u76d1\u7763\u65b9\u5f0f\u5904\u7406\u9759\u6001\u548c\u52a8\u6001\u7269\u4f53\uff0c\u5e76\u4e14\u7531\u4e8e\u6240\u63d0\u51fa\u7684\u89c6\u89c9 &#8211; \u8bed\u8a00\u77e5\u8bc6\u84b8\u998f\uff0c\u80fd\u591f\u8f93\u51fa\u5f00\u653e\u8bcd\u6c47\u8bed\u4e49\u6807\u7b7e\u3002\u5728 Waymo \u5f00\u653e\u6570\u636e\u96c6\u4e0a\u7684\u5b9e\u9a8c\u8868\u660e\uff0c\u6211\u4eec\u7684\u65b9\u6cd5\u5728\u5404\u79cd\u65e0\u76d1\u7763 3D \u611f\u77e5\u4efb\u52a1\u4e0a\u663e\u8457\u4f18\u4e8e\u5148\u524d\u7684\u5de5\u4f5c\u3002<\/p>\n\n\n\n<p>\u7b97\u6cd5\u6982\u89c8\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic2.zhimg.com\/v2-4e1171b4f2487a12ecff165863c53701_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e3b\u8981\u5b9e\u9a8c\u7ed3\u679c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/picx.zhimg.com\/v2-4536f11c630309cdee7d60eab03d2821_1440w.jpg\" alt=\"\"\/><\/figure>\n","protected":false},"excerpt":{"rendered":"<p>\u539f\u6587\u94fe\u63a5\uff1ahttps:\/\/zhuanlan.z [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3,1],"tags":[],"class_list":["post-31883","post","type-post","status-publish","format-standard","hentry","category-technology-frontier","category-home"],"views":0,"_links":{"self":[{"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/posts\/31883"}],"collection":[{"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/comments?post=31883"}],"version-history":[{"count":1,"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/posts\/31883\/revisions"}],"predecessor-version":[{"id":31884,"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/posts\/31883\/revisions\/31884"}],"wp:attachment":[{"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/media?parent=31883"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/categories?post=31883"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/tags?post=31883"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}