{"id":28521,"date":"2025-05-09T17:33:13","date_gmt":"2025-05-09T09:33:13","guid":{"rendered":"http:\/\/192.168.10.115\/?p=28521"},"modified":"2025-05-09T17:33:13","modified_gmt":"2025-05-09T09:33:13","slug":"2025-05-09-%e6%b8%85%e5%8d%8e%e6%9c%80%e6%96%b0%ef%bc%81rift%ef%bc%9a%e9%97%ad%e7%8e%afrl%e5%be%ae%e8%b0%83%ef%bc%8c%e7%94%a8%e4%ba%8e%e7%9c%9f%e5%ae%9e%e5%8f%af%e6%8e%a7%e7%9a%84%e4%ba%a4%e9%80%9a","status":"publish","type":"post","link":"http:\/\/222.128.65.89:18086\/index.php\/2025\/05\/09\/28521\/","title":{"rendered":"2025-05-09 \u6e05\u534e\u6700\u65b0\uff01RIFT\uff1a\u95ed\u73afRL\u5fae\u8c03\uff0c\u7528\u4e8e\u771f\u5b9e\u53ef\u63a7\u7684\u4ea4\u901a\u6a21\u62df"},"content":{"rendered":"\n<p>\u539f\u6587\u94fe\u63a5\uff1a<a href=\"https:\/\/zhuanlan.zhihu.com\/p\/1903725539879032282\">https:\/\/zhuanlan.zhihu.com\/p\/1903725539879032282<\/a><\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1903725539879032282_0\"><a href=\"https:\/\/zhida.zhihu.com\/search?content_id=257459868&amp;content_type=Article&amp;match_order=1&amp;q=Frenet+Corridor+Planner&amp;zhida_source=entity\" target=\"_blank\" rel=\"noreferrer noopener\">Frenet Corridor Planner<\/a><\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8bba\u6587\u6807\u9898\uff1aFrenet Corridor Planner: An Optimal Local Path Planning Framework for Autonomous Driving<\/li>\n\n\n\n<li>\u8bba\u6587\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2505.03695\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2505.03695<\/a><\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pica.zhimg.com\/v2-89557e96cb5b8006d86f4f11ba24c5ca_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p><strong>\u6838\u5fc3\u521b\u65b0\u70b9\uff1a<\/strong><\/p>\n\n\n\n<p><strong>1. Frenet\u7a7a\u95f4\u4e0b\u7684\u969c\u788d\u7269\u5efa\u6a21\u4e0e\u53ef\u884c\u9a76\u8d70\u5eca\u751f\u6210<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5b89\u5168\u589e\u5f3a\u5305\u56f4\u76d2\uff08Safety-Augmented Bounding Boxes\uff09 \uff1a\u5c06\u52a8\u6001\u8f66\u8f86\u5efa\u6a21\u4e3a\u5b89\u5168\u6269\u5c55\u7684\u77e9\u5f62\u5305\u56f4\u76d2\uff0c\u884c\u4eba\u805a\u7c7b\u4e3a\u51f8\u5305\uff08Convex Hulls\uff09\uff0c\u5728Frenet\u5750\u6807\u7cfb\u4e0b\u5b9e\u73b0\u9ad8\u6548\u969c\u788d\u7269\u8868\u5f81\u3002<\/li>\n\n\n\n<li>\u52a8\u6001\u504f\u79bb\u4fa7\u51b3\u7b56\uff08Deviation Side Determination\uff09 \uff1a\u901a\u8fc7\u51b3\u7b56\u6811\uff08Decision Tree\uff09\u786e\u5b9a\u9759\u6001\u969c\u788d\u7269\u7684\u4e0a\u4e0b\u8fb9\u754c\u5f52\u5c5e\uff08Upper\/Lower Bound\uff09\uff0c\u6784\u5efa\u53ef\u884c\u9a76\u8d70\u5eca\uff08Corridor\uff09\uff0c\u89e3\u51b3\u4f20\u7edf\u65b9\u6cd5\u5728\u590d\u6742\u573a\u666f\u4e2d\u7684\u8fb9\u754c\u51b2\u7a81\u95ee\u9898\u3002<\/li>\n\n\n\n<li>\u8fb9\u754c\u751f\u6210\u7b97\u6cd5\uff08Boundary Generation Algorithm\uff09 \uff1a\u57fa\u4e8e\u7eb5\u5411\u5206\u6bb5\uff08Space-Domain Sampling\uff09\u76f4\u63a5\u751f\u6210\u8d70\u5eca\u8fb9\u754c\uff0c\u663e\u8457\u63d0\u5347\u8ba1\u7b97\u6548\u7387\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>2. \u7a7a\u95f4\u57df\u81ea\u884c\u8f66\u8fd0\u52a8\u5b66\u6a21\u578b\uff08Space-Domain Bicycle Kinematics\uff09<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u65f6\u7a7a\u89e3\u8026\u5efa\u6a21 \uff1a\u5c06\u4f20\u7edf\u65f6\u95f4\u57df\u7684\u81ea\u884c\u8f66\u6a21\u578b\u8f6c\u6362\u4e3a\u7a7a\u95f4\u57df\uff08Longitudinal Distance \u0394s\u4e3a\u81ea\u53d8\u91cf\uff09\uff0c\u6d88\u9664\u901f\u5ea6\u6ce2\u52a8\u5bf9\u8def\u5f84\u89c4\u5212\u7684\u5f71\u54cd\uff0c\u5b9e\u73b0\u8def\u5f84\u4e0e\u901f\u5ea6\u7684\u89e3\u8026\u4f18\u5316\u3002<\/li>\n\n\n\n<li>\u66f2\u7387\u8865\u507f\u673a\u5236\uff08Curvature-Based Correction\uff09 \uff1a\u5f15\u5165\u53c2\u8003\u8def\u5f84\u7684\u66f2\u7387\u7ea6\u675f\uff0c\u786e\u4fdd\u8fd0\u52a8\u5b66\u53ef\u884c\u6027\uff0c\u907f\u514d\u56e0\u5ffd\u7565\u9053\u8def\u66f2\u7387\u5bfc\u81f4\u7684\u8def\u5f84\u8ddf\u8e2a\u5931\u6548\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>3. \u591a\u76ee\u6807\u975e\u7ebf\u6027\u4f18\u5316\u6846\u67b6<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u52a8\u6001\u969c\u788d\u7269\u98ce\u9669\u9879\uff08Dynamic Obstacle Risk Term\uff09 \uff1a\u5c06\u52a8\u6001\u969c\u788d\u7269\u8f68\u8ff9\u9884\u6d4b\u5d4c\u5165\u76ee\u6807\u51fd\u6570\uff0c\u907f\u514d\u4f20\u7edf\u78b0\u649e\u89c4\u907f\u7ea6\u675f\u5bfc\u81f4\u7684\u9012\u5f52\u4e0d\u53ef\u884c\u6027\u3002<\/li>\n\n\n\n<li>\u9c81\u68d2\u6027\u589e\u5f3a \uff1a\u5f15\u5165\u677e\u5f1b\u53d8\u91cf\uff08Slack Variables\uff09\u4e0e\u60e9\u7f5a\u9879\u5904\u7406\u611f\u77e5\u566a\u58f0\uff0c\u4fdd\u969c\u89c4\u5212\u53ef\u884c\u6027\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>4. \u5b9e\u9a8c\u9a8c\u8bc1\u4e0e\u90e8\u7f72<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4eff\u771f\u4e0e\u5b9e\u8f66\u6d4b\u8bd5 \uff1a\u5728CARLA\u4eff\u771f\u5668\u53ca1\/10\u6bd4\u4f8b\u7684MuSHR\u786c\u4ef6\u5e73\u53f0\u4e0a\u9a8c\u8bc1\u4e86FCP\u7684\u5b9e\u65f6\u6027\uff08\u5e73\u5747\u8ba1\u7b97\u65f6\u95f40.0424\u79d2\uff09\u4e0e\u9c81\u68d2\u6027\uff0c\u5c24\u5176\u5728\u9ad8\u566a\u58f0\u73af\u5883\u4e0b\u4ecd\u80fd\u4fdd\u6301\u8def\u5f84\u4e00\u81f4\u6027\u3002<\/li>\n\n\n\n<li>\u6027\u80fd\u5bf9\u6bd4\u4f18\u52bf \uff1a\u76f8\u6bd4A* \u3001RRT\u7b49\u57fa\u7ebf\u65b9\u6cd5\uff0cFCP\u5728\u8def\u5f84\u5e73\u6ed1\u6027\uff08\u6700\u5927\u504f\u822a\u89d2\u53d8\u53160.053 rad vs. RRT \u76840.816 rad\uff09\u3001\u5b89\u5168\u6027\uff08\u6700\u5c0f\u8f66\u8ddd3.18 m vs. A\u76842.68 m\uff09\u53ca\u5b9e\u65f6\u6027\uff080.035\u79d2 vs. A \u76840.628\u79d2\uff09\u4e0a\u5747\u663e\u8457\u63d0\u5347\u3002<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1903725539879032282_1\"><a href=\"https:\/\/zhida.zhihu.com\/search?content_id=257459868&amp;content_type=Article&amp;match_order=1&amp;q=CaRaFFusion&amp;zhida_source=entity\" target=\"_blank\" rel=\"noreferrer noopener\">CaRaFFusion<\/a><\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8bba\u6587\u6807\u9898\uff1aCaRaFFusion: Improving 2D Semantic Segmentation with Camera-Radar Point Cloud Fusion and Zero-Shot Image Inpainting<\/li>\n\n\n\n<li>\u8bba\u6587\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2505.03679\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2505.03679<\/a><\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic2.zhimg.com\/v2-356f15466c21752a52136c0ba4c1f423_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p><strong>\u6838\u5fc3\u521b\u65b0\u70b9\uff1a<\/strong><\/p>\n\n\n\n<p><strong>1. \u4e09\u9636\u6bb5\u76f8\u673a-\u96f7\u8fbe\u878d\u5408\u67b6\u6784<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8de8\u6a21\u6001\u7279\u5f81\u878d\u5408 \uff1a\u901a\u8fc7\u4ea4\u53c9\u6ce8\u610f\u529b\u673a\u5236\uff08Cross-Attention\uff09\u878d\u5408\u96f7\u8fbe\u70b9\u4e91\u7a7a\u95f4\u7279\u5f81\u4e0eRGB\u56fe\u50cf\u89c6\u89c9\u7279\u5f81\uff0c\u751f\u6210\u521d\u59cb\u5206\u5272\u63a9\u7801\uff08<code>M&lt;sub>init&lt;\/sub><\/code>\uff09\u3002<\/li>\n\n\n\n<li>\u96f7\u8fbe\u9a71\u52a8\u7684\u4f2a\u63a9\u7801\u751f\u6210 \uff1a\u5229\u7528Segment-Anything Model\uff08SAM\uff09\u5c06\u96f7\u8fbe\u70b9\u4e91\u6295\u5f71\u4f5c\u4e3a\u70b9\u63d0\u793a\uff08Point Prompts\uff09\uff0c\u751f\u6210\u9c81\u68d2\u7684\u4f2a\u63a9\u7801\uff08<code>M&lt;sub>sam&lt;\/sub><\/code>\uff09\uff0c\u7f13\u89e3\u6076\u52a3\u5929\u6c14\u4e0b\u89c6\u89c9\u7279\u5f81\u5931\u6548\u95ee\u9898\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>2. \u566a\u58f0\u51cf\u5c11\u5355\u5143\uff08Noise Reduction Unit, NRU\uff09<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u52a8\u6001\u566a\u58f0\u8fc7\u6ee4 \uff1a\u9488\u5bf9\u96f7\u8fbe\u70b9\u4e91\u5728\u6c34\u4f53\u8868\u9762\u7b49\u533a\u57df\u7684\u566a\u58f0\u5e72\u6270\uff0c\u63d0\u51fa\u901a\u9053\u7ea7\u53bb\u566a\u7b56\u7565\uff1a<\/li>\n<\/ul>\n\n\n\n<ol class=\"wp-block-list\">\n<li>\u566a\u58f0\u63a9\u7801\u6784\u5efa \uff1a\u901a\u8fc7\u80cc\u666f\uff08<code>M&lt;sub>background&lt;\/sub><\/code>\uff09\u4e0e\u6c34\u9762\uff08<code>M&lt;sub>water&lt;\/sub><\/code>\uff09\u63a9\u7801\u53e0\u52a0\u751f\u6210<code>M&lt;sub>noise&lt;\/sub><\/code>\u3002<\/li>\n\n\n\n<li>\u63a9\u7801\u4f18\u5316 \uff1a\u5bf9\u4f2a\u63a9\u7801<code>M&lt;sub>sam&lt;\/sub><\/code>\u8fdb\u884c\u9010\u901a\u9053\u51cf\u6cd5\u53bb\u566a\uff08<code>M&lt;sub>denoised&lt;\/sub> = M&lt;sub>sam&lt;\/sub><\/code> &#8211; <code>M&lt;sub>noise&lt;\/sub><\/code>\uff09\uff0c\u7ed3\u5408ReLU\u6fc0\u6d3b\u4e0e\u521d\u59cb\u63a9\u7801\u53e0\u52a0\uff0c\u8f93\u51fa\u964d\u566a\u63a9\u7801<code>M&lt;sub>nr&lt;\/sub><\/code>\u3002<\/li>\n<\/ol>\n\n\n\n<p><strong>3. \u96f6\u6837\u672c\u56fe\u50cf\u4fee\u590d\uff08Zero-Shot Image Inpainting\uff09<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6269\u6563\u6a21\u578b\u9a71\u52a8\u7684\u7f3a\u5931\u4fe1\u606f\u91cd\u5efa \uff1a\u57fa\u4e8eStable Diffusion\u6a21\u578b\uff0c\u4ee5<code>M&lt;sub>nr&lt;\/sub><\/code>\u4e3a\u5f15\u5bfc\u4fee\u590d\u56fe\u50cf\u6a21\u7cca\/\u906e\u6321\u533a\u57df\uff0c\u751f\u6210\u8865\u5145\u4fe1\u606f\u7684\u4fee\u590d\u56fe\u50cf\uff08<code>I&lt;sub>inp&lt;\/sub><\/code>\uff09\u3002<\/li>\n\n\n\n<li>\u53cc\u7f16\u7801\u5668\u7279\u5f81\u878d\u5408 \uff1a\u901a\u8fc7\u72ec\u7acbSegformer\u7f16\u7801\u5668\u5206\u522b\u5904\u7406\u539f\u59cb\u56fe\u50cf\u4e0e\u4fee\u590d\u56fe\u50cf\uff0c\u878d\u5408\u7279\u5f81\u540e\u8f93\u51fa\u6700\u7ec8\u5206\u5272\u63a9\u7801\uff0c\u589e\u5f3a\u5bf9\u6076\u52a3\u5929\u6c14\u573a\u666f\u7684\u9002\u5e94\u6027\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>4. \u591a\u6a21\u6001\u534f\u540c\u4f18\u5316\u7b56\u7565<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u7aef\u5230\u7aef\u8bad\u7ec3\u673a\u5236 \uff1a\u8054\u5408\u4f18\u5316\u96f7\u8fbe\u70b9\u4e91\u5206\u7c7b\u4efb\u52a1\u4e0e\u56fe\u50cf\u5206\u5272\u4efb\u52a1\uff0c\u901a\u8fc7Focal Loss\u4e0eDice Loss\u5e73\u8861\u7c7b\u522b\u5206\u5e03\uff0c\u63d0\u5347\u591a\u6a21\u6001\u7279\u5f81\u8868\u5f81\u80fd\u529b\u3002<\/li>\n\n\n\n<li>\u6e10\u8fdb\u5f0f\u6846\u67b6\u8bbe\u8ba1 \uff1a\u5206\u9636\u6bb5\u8bad\u7ec3\u7b56\u7565\uff08Stage 1\u2192Stage 3\uff09\u964d\u4f4e\u8ba1\u7b97\u590d\u6742\u5ea6\uff0c\u540c\u65f6\u901a\u8fc7\u6d88\u878d\u5b9e\u9a8c\u8bc1\u660e\u5404\u6a21\u5757\u5bf9mIoU\u63d0\u5347\u7684\u8d21\u732e\uff08\u5982NRU\u63d0\u53471.48%\uff0c\u4fee\u590d\u6a21\u5757\u63d0\u53470.96%\uff09\u3002<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1903725539879032282_2\">RIFT<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8bba\u6587\u6807\u9898\uff1aRIFT: Closed-Loop RL Fine-Tuning for Realistic and Controllable Traffic Simulation<\/li>\n\n\n\n<li>\u8bba\u6587\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2505.03344\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2505.03344<\/a><\/li>\n\n\n\n<li>\u9879\u76ee\u4e3b\u9875\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/currychen77.github.io\/RIFT\/\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/currychen77.github.io\/RIFT\/<\/a><\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic3.zhimg.com\/v2-10d68842ca71caefa94420b471c0448a_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p><strong>\u6838\u5fc3\u521b\u65b0\u70b9\uff1a<\/strong><\/p>\n\n\n\n<p><strong>1. \u53cc\u9636\u6bb5\u4eff\u771f\u6846\u67b6<\/strong><\/p>\n\n\n\n<p>\u63d0\u51fa<strong>\u5f00\u73af\u6a21\u4eff\u5b66\u4e60\uff08IL\uff09\u9884\u8bad\u7ec3 + \u95ed\u73af\u5f3a\u5316\u5b66\u4e60\uff08RL\uff09\u5fae\u8c03<\/strong>\u7684\u53cc\u9636\u6bb5\u6846\u67b6\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6570\u636e\u9a71\u52a8\u4eff\u771f\u5668\u4e2d\u901a\u8fc7IL\u9884\u8bad\u7ec3\u6355\u6349\u8f68\u8ff9\u7ea7\u771f\u5b9e\u6027\u4e0e\u591a\u6a21\u6001\u6027\uff08\u5982\u901f\u5ea6\u3001\u52a0\u901f\u5ea6\u5206\u5e03\uff09\uff1b<\/li>\n\n\n\n<li>\u7269\u7406\u4eff\u771f\u5668\u4e2d\u901a\u8fc7\u95ed\u73afRL\u5fae\u8c03\u7f13\u89e3\u534f\u53d8\u91cf\u504f\u79fb\uff08covariate shift\uff09\uff0c\u589e\u5f3a\u4ea4\u4e92\u53ef\u63a7\u6027\uff0c\u7ed3\u5408\u4e24\u7c7b\u5e73\u53f0\u4f18\u52bf\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>2. RIFT\u95ed\u73af\u5fae\u8c03\u7b56\u7565<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u7ec4\u76f8\u5bf9\u4f18\u52bf<\/strong>\uff08GRPO-style Group-Relative Advantage\uff09\uff1a\u5c06\u591a\u5019\u9009\u8f68\u8ff9\u89c6\u4e3a\u7ec4\u8f93\u51fa\uff0c\u8ba1\u7b97\u7ec4\u5185\u76f8\u5bf9\u5956\u52b1\u4e0e\u4f18\u52bf\uff0c\u4fdd\u7559\u8f68\u8ff9\u591a\u6a21\u6001\u6027\uff0c\u907f\u514d\u4f20\u7edfRL\u4ec5\u4f18\u5316\u5355\u4e00\u6700\u4f18\u8f68\u8ff9\u7684\u5c40\u9650\u3002<\/li>\n\n\n\n<li><strong>\u53cc\u526a\u88c1\u673a\u5236<\/strong>\uff08Dual-Clip Surrogate\uff09\uff1a\u4ee5\u5e38\u6570\u4e0b\u754cc>1\u7ea6\u675f\u8d1f\u4f18\u52bf\u503c\uff0c\u66ff\u4ee3KL\u6b63\u5219\u5316\uff0c\u89e3\u51b3\u4f20\u7edfPPO\u5728\u7ec4\u4f18\u5316\u4e2d\u56e0\u91cd\u8981\u6027\u91c7\u6837\u6bd4\u7387\u8fc7\u5927\u5bfc\u81f4\u7684\u68af\u5ea6\u7206\u70b8\u95ee\u9898\uff0c\u63d0\u5347\u8bad\u7ec3\u7a33\u5b9a\u6027\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>3. \u5173\u952e\u80cc\u666f\u8f66\u8f86\uff08CBV\uff09\u52a8\u6001\u8bc6\u522b<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u57fa\u4e8e\u8def\u5f84\u7ea7\u4ea4\u4e92\u6982\u7387\u5206\u6790\uff0c\u7b5b\u9009\u4e0e\u81ea\u52a8\u9a7e\u9a76\u8f66\u8f86\uff08AV\uff09\u4ea4\u4e92\u6982\u7387\u9ad8\u7684\u80cc\u666f\u8f66\u8f86\uff0c\u4ec5\u5bf9CBV\u8fdb\u884c\u8f68\u8ff9\u751f\u6210\u4e0e\u5fae\u8c03\uff0c\u964d\u4f4e\u8ba1\u7b97\u5f00\u9500\uff0c\u540c\u65f6\u805a\u7126\u5173\u952e\u4ea4\u4e92\u573a\u666f\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>4. \u6a21\u5757\u5316\u5fae\u8c03\u8bbe\u8ba1<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4ec5\u5fae\u8c03\u8f68\u8ff9\u8bc4\u5206\u5934\uff08Trajectory Scoring Head\uff09\uff0c\u51bb\u7ed3\u9884\u8bad\u7ec3\u7f51\u7edc\u5176\u4ed6\u90e8\u5206\uff0c\u907f\u514d\u7834\u574fIL\u9636\u6bb5\u5b66\u4e60\u7684\u8f68\u8ff9\u7ea7\u771f\u5b9e\u6027\u4e0e\u591a\u6a21\u6001\u8868\u5f81\u3002<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1903725539879032282_3\">Automated Data Curation Using GPS<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8bba\u6587\u6807\u9898\uff1aAutomated Data Curation Using GPS &amp; NLP to Generate Instruction-Action Pairs for Autonomous Vehicle Vision-Language Navigation Datasets<\/li>\n\n\n\n<li>\u8bba\u6587\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2505.03174\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2505.03174<\/a><\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic4.zhimg.com\/v2-66510b7e6fe2c8ec35aa31f5ced728b5_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p><strong>\u6838\u5fc3\u521b\u65b0\u70b9\uff1a<\/strong><\/p>\n\n\n\n<p><strong>1. \u57fa\u4e8eGPS\u4e0eNLP\u7684\u6307\u4ee4-\u52a8\u4f5c\u5bf9\u81ea\u52a8\u751f\u6210\u6846\u67b6<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u63d0\u51fa\u4e86\u4e00\u79cd\u65e0\u9700\u4eba\u5de5\u6807\u6ce8\u7684\u81ea\u52a8\u5316\u6570\u636e\u6574\u7406\u65b9\u6cd5\uff0c\u901a\u8fc7\u89e3\u6790\u79fb\u52a8\u8bbe\u5907GPS\u5bfc\u822a\u5e94\u7528\uff08\u5982Apple Maps\u3001Google Maps\u3001Waze\uff09\u7684\u8bed\u97f3\u6307\u4ee4\u8f93\u51fa\uff0c\u7ed3\u5408\u81ea\u7136\u8bed\u8a00\u5904\u7406\uff08NLP\uff09\u6280\u672f\uff0c\u6279\u91cf\u751f\u6210\u9ad8\u8d28\u91cf\u7684\u6307\u4ee4-\u52a8\u4f5c\u5bf9\uff08Instruction-Action Pairs, IA Pairs\uff09 \u3002\u8be5\u6846\u67b6\u663e\u8457\u964d\u4f4e\u4e86\u4f20\u7edf\u4f9d\u8d56\u4eba\u5de5\u6807\u6ce8\u7684\u6570\u636e\u91c7\u96c6\u6210\u672c\u4e0e\u65f6\u8017\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>2. \u591a\u6a21\u6001\u89c6\u89c9-\u8bed\u8a00-\u52a8\u4f5c\uff08VLA\uff09\u4e09\u5143\u7ec4\u6784\u5efa<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5c06GPS\u8f68\u8ff9\uff08\u52a8\u4f5c\u7a7a\u95f4\uff09\u3001\u8f66\u8f7d\u6444\u50cf\u5934\u89c6\u9891\uff08\u89c6\u89c9\u7a7a\u95f4\uff09\u4e0e\u8bed\u97f3\u6307\u4ee4\u8f6c\u5f55\u6587\u672c\uff08\u8bed\u8a00\u7a7a\u95f4\uff09\u8fdb\u884c\u65f6\u7a7a\u540c\u6b65\uff0c\u6784\u5efa\u89c6\u89c9-\u8bed\u8a00-\u52a8\u4f5c\u4e09\u5143\u7ec4\uff08Vision-Language-Action Triads\uff09 \uff0c\u4e3a\u7aef\u5230\u7aef\u8bad\u7ec3\u89c6\u89c9-\u8bed\u8a00\u5bfc\u822a\u6a21\u578b\uff08Vision-Language Navigation, VLN\uff09\u63d0\u4f9b\u7ed3\u6784\u5316\u6570\u636e\u652f\u6301\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>3. \u5bfc\u822a\u6307\u4ee4\u7684\u7ec6\u7c92\u5ea6\u8bed\u4e49\u5206\u7c7b\u4f53\u7cfb<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u63d0\u51fa\u5305\u542b8\u4e2a\u8bed\u4e49\u7ef4\u5ea6\u7684\u6307\u4ee4\u5206\u7c7b\u6cd5\uff08Road Name, Distance, Static Object, Turn, Cardinal, Location Name, Lane Information, Light Information\uff09\uff0c\u7cfb\u7edf\u523b\u753b\u4e86\u5bfc\u822a\u6307\u4ee4\u4e2d\u7a7a\u95f4\u53c2\u7167\u7cfb\uff08spatial referentiality\uff09\u7684\u591a\u6837\u6027\uff0c\u63ed\u793a\u4e86\u4e0d\u540c\u5bfc\u822a\u5e94\u7528\u5728\u6307\u4ee3\u8868\u5f81\u4e0a\u7684\u5f02\u6784\u7279\u6027\uff08\u5982\u8ddd\u79bb\u53c2\u7167vs.\u9759\u6001\u7269\u4f53\u53c2\u7167\uff09\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>4. ADVLAT-Engine\u539f\u578b\u7cfb\u7edf<\/strong><\/p>\n\n\n\n<p>\u5b9e\u73b0\u4e86\u5b8c\u6574\u7684\u81ea\u52a8\u5316\u6570\u636e\u91c7\u96c6\u6d41\u6c34\u7ebf\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u591a\u4f20\u611f\u5668\u878d\u5408 \uff1a\u540c\u6b65\u8bb0\u5f55\u89c6\u9891\u6d41\u3001GPS\u5750\u6807\u4e0e\u8bed\u97f3\u6307\u4ee4<\/li>\n\n\n\n<li>\u8bed\u97f3\u8f6c\u5f55\u5bf9\u9f50 \uff1a\u91c7\u7528Whisper\u6a21\u578b\u5b9e\u73b0\u6307\u4ee4\u65f6\u5e8f\u5b9a\u4f4d<\/li>\n\n\n\n<li>\u8f68\u8ff9-\u8bed\u4e49\u6620\u5c04 \uff1a\u5efa\u7acb\u8f66\u8f86\u8f68\u8ff9\u5173\u952e\u70b9\u4e0e\u8bed\u8a00\u6307\u4ee4\u7684\u8de8\u6a21\u6001\u5173\u8054<\/li>\n<\/ul>\n\n\n\n<p>\u8be5\u7cfb\u7edf\u652f\u6301\u6269\u5c55\u81f3LiDAR\u3001CAN\u603b\u7ebf\u7b49\u4f20\u611f\u5668\u6570\u636e\uff0c\u6ee1\u8db3\u4e0d\u540cVLA\u6a21\u578b\u7684\u8f93\u5165\u9700\u6c42\u3002<\/p>\n\n\n\n<p><strong>5. \u5927\u89c4\u6a21\u6307\u4ee4\u53d8\u4f53\u6570\u636e\u5e93\u6784\u5efa<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u901a\u8fc7\u5206\u67903\u4e2a\u4e3b\u6d41\u5bfc\u822a\u5e94\u7528\u572871\/82\/80\u6761\u8def\u5f84\u4e0a\u7684\u6307\u4ee4\u8f93\u51fa\uff0c\u6784\u5efa\u4e86\u5305\u542b\u591a\u5c5e\u6027\u7ec4\u5408\uff08\u5982&#8221;Distance+Turn+Road Name&#8221;\uff09\u7684\u6307\u4ee4\u53d8\u4f53\u5e93\uff08\u603b\u8ba178\u79cd\u7ec4\u5408\u6a21\u5f0f\uff09\uff0c\u63ed\u793a\u4e86\u5b9e\u9645\u5bfc\u822a\u573a\u666f\u4e2d\u8bed\u8a00\u6307\u4ee4\u7684\u590d\u6742\u6027\u4e0e\u591a\u6837\u6027\u3002<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1903725539879032282_4\">Spotting the Unexpected (STU)<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8bba\u6587\u6807\u9898\uff1aSpotting the Unexpected (STU): A 3D LiDAR Dataset for Anomaly Segmentation in Autonomous Driving<\/li>\n\n\n\n<li>\u8bba\u6587\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2505.02148\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2505.02148<\/a><\/li>\n\n\n\n<li>\u9879\u76ee\u4e3b\u9875\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/www.vision.rwth-aachen.de\/stu-dataset\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/www.vision.rwth-aachen.de\/stu-dataset<\/a><\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic2.zhimg.com\/v2-7674b29ce5f1cdc88217f5a76bbb3173_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p><strong>\u6838\u5fc3\u521b\u65b0\u70b9\uff1a<\/strong><\/p>\n\n\n\n<p><strong>1. \u9996\u4e2a\u9762\u5411\u81ea\u52a8\u9a7e\u9a76\u76843D LiDAR\u5f02\u5e38\u5206\u5272\u516c\u5f00\u6570\u636e\u96c6\uff08STU\uff09<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u586b\u8865\u9886\u57df\u7a7a\u767d \uff1a\u63d0\u51fa\u9996\u4e2a\u4e13\u6ce8\u4e8e\u9053\u8def\u5f02\u5e38\u7269\u4f53\uff08Out-of-Distribution, OOD\uff09\u7684\u5bc6\u96c63D\u8bed\u4e49\u6807\u6ce8 \u6570\u636e\u96c6\uff0c\u7ed3\u5408\u9ad8\u5206\u8fa8\u7387LiDAR\u70b9\u4e91 \u4e0e\u540c\u6b65\u591a\u89c6\u89d2\u76f8\u673a\u6570\u636e \uff0c\u652f\u6301\u8de8\u6a21\u6001\u4e0e\u65f6\u95f4\u5e8f\u5217\u5206\u6790\u3002<\/li>\n\n\n\n<li>\u6570\u636e\u89c4\u6a21\u4e0e\u8d28\u91cf \uff1a\u5305\u542b70\u4e2a\u5f02\u5e38\u5e8f\u5217\uff0868\u4e2a\u53d7\u63a7\u573a\u666f+2\u4e2a\u771f\u5b9e\u9053\u8def\u573a\u666f\uff09\uff0c\u8986\u76d6\u57ce\u5e02\u4e0e\u4e61\u6751\u73af\u5883\uff0c\u6807\u6ce8\u8303\u56f4\u8fbe150\u7c73\uff08\u8bad\u7ec3\u9636\u6bb5\u9650\u5236\u4e3a50\u7c73\uff09\u3002\u63d0\u4f9b\u9010\u70b9\u8bed\u4e49\u6807\u7b7e \uff08inlier\/outlier\/unlabeled\uff09\u53ca\u5b9e\u4f8b\u7ea7\u6807\u6ce8 \uff0c\u652f\u6301\u5168\u666f\u5206\u5272\u8bc4\u4f30\u3002<\/li>\n\n\n\n<li>\u73b0\u5b9e\u573a\u666f\u9002\u914d \uff1a\u901a\u8fc7\u5206\u6790SemanticKITTI\u4e0enuScenes\u6807\u7b7e\u4f53\u7cfb\uff0c\u786e\u4fdd\u5f02\u5e38\u7269\u4f53\uff08\u5982\u8def\u969c\u3001\u788e\u7247\uff09\u4e0e\u8bad\u7ec3\u96c6\u65e0\u91cd\u53e0\uff0c\u907f\u514d\u5f00\u653e\u96c6\u5206\u5272\u7684\u6807\u7b7e\u6c61\u67d3\u95ee\u9898\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>2. \u591a\u6a21\u6001\u6570\u636e\u91c7\u96c6\u4e0e\u6807\u6ce8\u6846\u67b6<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u786c\u4ef6\u914d\u7f6e \uff1a\u91c7\u7528128\u7ebfOuster LiDAR \uff08\u5782\u76f4\u5206\u8fa8\u738745\u00b0\uff0c\u91cf\u7a0b200\u7c73\uff09\u4e0e8\u4e2a\u786c\u4ef6\u89e6\u53d1\u76f8\u673a\uff08\u8986\u76d6360\u00b0\u89c6\u573a\uff09\uff0c\u5b9e\u73b0\u540c\u6b65\u591a\u6a21\u6001\u6570\u636e\u91c7\u96c6 \u3002<\/li>\n\n\n\n<li>\u6807\u6ce8\u7b56\u7565 \uff1a\u57fa\u4e8eSemanticKITTI\u6807\u7b7e\u4f53\u7cfb\u6269\u5c55\uff0c\u5b9a\u4e49\u4e09\u7c7b\u6807\u7b7e\uff1a\n<ul class=\"wp-block-list\">\n<li>Inlier \uff1a\u5e38\u89c4\u9053\u8def\u7269\u4f53\uff08\u8f66\u8f86\u3001\u884c\u4eba\u3001\u57fa\u7840\u8bbe\u65bd\uff09\uff1b<\/li>\n\n\n\n<li>Outlier \uff1a\u5f02\u5e38\u7269\u4f53\uff08\u975e\u9053\u8def\u56fa\u6709\u7269\u4f53\uff09\uff1b<\/li>\n\n\n\n<li>Unlabeled \uff1a\u8bad\u7ec3\u96c6\u4e2d\u5b58\u5728\u7684\u5f31\u76d1\u7763\u7c7b\u522b\uff08\u5982\u8def\u706f\u3001\u5783\u573e\u6876\uff09\uff0c\u907f\u514d\u6a21\u578b\u504f\u5dee\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u540e\u5904\u7406\u6280\u672f \uff1a\u5e94\u7528KISS ICP\u7b97\u6cd5\u8fdb\u884c\u70b9\u4e91\u914d\u51c6\uff0c\u7ed3\u5408Patchwork++\u5b9e\u73b0\u5730\u9762\u5206\u5272\uff1b\u91c7\u7528DashcamCleaner\u4e0eDeepPrivacy2\u5b8c\u6210\u56fe\u50cf\u9690\u79c1\u4fdd\u62a4\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>3. 3D\u5f02\u5e38\u5206\u5272\u57fa\u7ebf\u6a21\u578b\u4e0e\u8bc4\u4f30\u4f53\u7cfb<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u57fa\u7ebf\u8fc1\u79fb \uff1a\u5c062D\u5f02\u5e38\u5206\u5272\u65b9\u6cd5\uff08\u5982MaxLogit\u3001RbA\uff09\u9002\u914d\u81f33D\u9886\u57df\uff0c\u57fa\u4e8eMask4Former-3D\u67b6\u6784\u6784\u5efa\u9996\u4e2a3D LiDAR\u5f02\u5e38\u5206\u5272\u57fa\u7ebf \u3002<\/li>\n\n\n\n<li>\u6027\u80fd\u5206\u6790 \uff1a\u5b9e\u9a8c\u8868\u660e\uff0c\u73b0\u6709\u65b9\u6cd5\u57283D\u57df\u8868\u73b0\u663e\u8457\u4e0b\u964d\uff08\u5982Deep Ensemble\u572850\u7c73\u5185AUROC=86.74\uff0cAP=5.17\uff09\uff0c\u63ed\u793a\u70b9\u4e91\u7a00\u758f\u6027 \u4e0e\u957f\u5c3e\u5206\u5e03 \u5e26\u6765\u7684\u6311\u6218\u3002<\/li>\n\n\n\n<li>\u8bc4\u4f30\u6307\u6807 \uff1a\u878d\u5408\u70b9\u7ea7\uff08AUROC\/FPR@95\/AP\uff09\u4e0e\u5b9e\u4f8b\u7ea7\uff08Unknown Quality, UQ\uff09\u6307\u6807\uff0c\u91cf\u5316\u6a21\u578b\u5bf9\u5fae\u5c0f\u5f02\u5e38\u7269\u4f53\uff08\u5982&lt;50\u70b9\/\u5b9e\u4f8b\uff09\u7684\u68c0\u6d4b\u80fd\u529b\u3002<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1903725539879032282_5\">DriveNetBench<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8bba\u6587\u6807\u9898\uff1aDriveNetBench: An Affordable and Configurable Single-Camera Benchmarking System for Autonomous Driving Networks<\/li>\n\n\n\n<li>\u8bba\u6587\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2505.01893\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2505.01893<\/a><\/li>\n\n\n\n<li>\u4ee3\u7801\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/github.com\/alibustami\/DriveNetBench\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/github.com\/alibustami\/DriveNetBench<\/a><\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic3.zhimg.com\/v2-37a36c7fc97c0b837ed638b5f6392e9a_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p><strong>\u6838\u5fc3\u521b\u65b0\u70b9\uff1a<\/strong><\/p>\n\n\n\n<p><strong>1. \u4f4e\u6210\u672c\u5355\u6444\u50cf\u5934\u57fa\u51c6\u6d4b\u8bd5\u67b6\u6784<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u521b\u65b0\u6027 \uff1a\u9996\u6b21\u63d0\u51fa\u57fa\u4e8e\u5355\u6444\u50cf\u5934\uff08\u5355\u76ee\u89c6\u89c9\uff09\u7684\u6807\u51c6\u5316\u57fa\u51c6\u6d4b\u8bd5\u7cfb\u7edf\uff0c\u7a81\u7834\u4f20\u7edf\u591a\u4f20\u611f\u5668\uff08LiDAR\/\u96f7\u8fbe\uff09\u7cfb\u7edf\u7684\u9ad8\u6210\u672c\u9650\u5236\u3002<\/li>\n\n\n\n<li>\u6280\u672f\u5b9e\u73b0 \uff1a\u91c7\u7528\u73b0\u6210\u6d88\u8d39\u7ea7\u786c\u4ef6\uff08\u5982\u4f4e\u529f\u8017\u76f8\u673a\u3001\u5d4c\u5165\u5f0f\u8bbe\u5907\uff09\u4e0e\u5f00\u6e90\u8f6f\u4ef6\u6808\uff0c\u6784\u5efa\u53ef\u590d\u73b0\u7684\u95ed\u73af\u6d4b\u8bd5\u73af\u5883\u3002<\/li>\n\n\n\n<li>\u4f18\u52bf \uff1a\u5c06\u81ea\u52a8\u9a7e\u9a76\u7f51\u7edc\u9a8c\u8bc1\u6210\u672c\u964d\u4f4e\u81f3\u4f20\u7edf\u65b9\u6848\u76841\/10\u4ee5\u4e0b\uff08\u6587\u4e2d\u793a\u4f8b\u786c\u4ef6\u6e05\u5355\u603b\u4ef7\u7ea6200\u7f8e\u5143\uff09\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>2. \u6a21\u5757\u5316\u57fa\u51c6\u6d4b\u8bd5\u6d41\u6c34\u7ebf<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u67b6\u6784\u8bbe\u8ba1 \uff1a\n<ul class=\"wp-block-list\">\n<li>\u56db\u5c42\u89e3\u8026\u7ed3\u6784 \uff1a\u8f93\u5165\u6570\u636e\u6e90 \u2192 \u7f51\u7edc\u5f85\u6d4b\u6a21\u5757\uff08Network Under Test\uff09 \u2192 \u53d8\u6362\u6a21\u5757\uff08Homography\u6295\u5f71\uff09 \u2192 \u8bc4\u4f30\u6a21\u5757\u3002<\/li>\n\n\n\n<li>\u9ed1\u7bb1\u517c\u5bb9\u6027 \uff1a\u652f\u6301\u4efb\u610f\u89c6\u89c9\u6a21\u578b\uff08\u5982YOLO\u76ee\u6807\u68c0\u6d4b\u3001\u7aef\u5230\u7aef\u8f66\u9053\u4fdd\u6301\u6a21\u578b\uff09\u5373\u63d2\u5373\u7528\uff0c\u65e0\u9700\u4fee\u6539\u4ee3\u7801\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5173\u952e\u6a21\u5757 \uff1a\n<ul class=\"wp-block-list\">\n<li>\u6570\u5b57\u5b6a\u751f\u5bf9\u9f50 \uff1a\u901a\u8fc7Keypoints Definer\u4e0eView Transformer\u5b9e\u73b0\u7269\u7406\u8f68\u8ff9\u4e0e\u6570\u5b57\u8f68\u9053\u5750\u6807\u7cfb\u7684\u7cbe\u786e\u6620\u5c04\uff08\u5e73\u5747\u91cd\u6295\u5f71\u8bef\u5dee&lt;5\u50cf\u7d20\uff09\u3002<\/li>\n\n\n\n<li>\u52a8\u6001\u8def\u5f84\u751f\u6210 \uff1aTrackProcessor\u81ea\u52a8\u63d0\u53d6\u4e2d\u5fc3\u7ebf\u4f5c\u4e3a\u53c2\u8003\u8def\u5f84\uff0c\u652f\u6301\u4efb\u610f\u4e8c\u7ef4\u8f68\u9053\u8bbe\u8ba1\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n<p><strong>3. \u95ed\u73af\u573a\u666f\u4e0b\u7684\u6807\u51c6\u5316\u8bc4\u4f30\u6307\u6807<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6838\u5fc3\u6307\u6807 \uff1a\n<ul class=\"wp-block-list\">\n<li>\u8def\u5f84\u76f8\u4f3c\u5ea6\uff08Path Similarity\uff09 \uff1a\u91c7\u7528\u52a8\u6001\u65f6\u95f4\u89c4\u6574\uff08DTW\uff09\u4e0eFr\u00e9chet\u8ddd\u79bb\u53cc\u5ea6\u91cf\uff0c\u91cf\u5316\u8f68\u8ff9\u4e0e\u53c2\u8003\u8def\u5f84\u7684\u5339\u914d\u7a0b\u5ea6\uff08\u516c\u5f0f1\u5f52\u4e00\u5316\u5904\u7406\uff09\u3002<\/li>\n\n\n\n<li>\u5b8c\u6210\u65f6\u95f4\uff08Completion Time\uff09 \uff1a\u5f15\u5165\u5931\u8d25\u60e9\u7f5a\u673a\u5236\uff08\u5982\u8131\u8f68\u4e8b\u4ef6\uff09\uff0c\u53cd\u6620\u7b97\u6cd5\u9c81\u68d2\u6027\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8bef\u5dee\u8bca\u65ad \uff1a\u901a\u8fc7Homography\u5173\u952e\u70b9\u6b8b\u5dee\u5206\u6790\uff08\u56fe4\u663e\u793a5\u70b9\u6821\u51c6\u540e\u8bef\u5dee\u4e0b\u964d82%\uff09\uff0c\u4fdd\u969c\u8bc4\u4f30\u53ef\u4fe1\u5ea6\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>4. \u53ef\u914d\u7f6e\u5316\u5b9e\u9a8c\u6846\u67b6<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u53c2\u6570\u5316\u63a7\u5236 \uff1a\u901a\u8fc7YAML\u914d\u7f6e\u6587\u4ef6\u5b9e\u73b0\u8f68\u9053\u5e03\u5c40\u3001\u7f6e\u4fe1\u5ea6\u9608\u503c\u3001\u8bc4\u4f30\u6307\u6807\u7684\u52a8\u6001\u8c03\u6574\uff08\u65e0\u9700\u4ee3\u7801\u4fee\u6539\uff09\u3002<\/li>\n\n\n\n<li>\u8de8\u5e73\u53f0\u517c\u5bb9\u6027 \uff1a\u652f\u6301F1TENTH\u3001DeepRacer\u7b49\u5f02\u6784\u786c\u4ef6\u5e73\u53f0\u7684\u6a2a\u5411\u5bf9\u6bd4\u3002<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_1903725539879032282_6\"><a href=\"https:\/\/zhida.zhihu.com\/search?content_id=257459868&amp;content_type=Article&amp;match_order=1&amp;q=Learning+to+Drive&amp;zhida_source=entity\" target=\"_blank\" rel=\"noreferrer noopener\">Learning to Drive<\/a><\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8bba\u6587\u6807\u9898\uff1aLearning to Drive from a World Model<\/li>\n\n\n\n<li>\u8bba\u6587\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2504.19077\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2504.19077<\/a><\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic2.zhimg.com\/v2-07f44c1a9ec1c9593803cb4e76f13d7d_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p><strong>\u6838\u5fc3\u521b\u65b0\u70b9\uff1a<\/strong><\/p>\n\n\n\n<p><strong>1. \u7aef\u5230\u7aef\u9a7e\u9a76\u7b56\u7565\u67b6\u6784<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u63d0\u51fa\u65e0\u9700\u624b\u5de5\u7f16\u7801\u611f\u77e5\u6a21\u5757\u6216\u9a7e\u9a76\u89c4\u5219\u7684\u7aef\u5230\u7aef\uff08E2E\uff09\u8bad\u7ec3\u6846\u67b6\uff0c\u76f4\u63a5\u4ece\u4eba\u7c7b\u9a7e\u9a76\u6570\u636e\u5b66\u4e60\u63a7\u5236\u7b56\u7565\u3002\u7b56\u7565\u57fa\u4e8e\u539f\u59cb\u4f20\u611f\u5668\u8f93\u5165\uff08\u5982\u56fe\u50cf\uff09\u76f4\u63a5\u8f93\u51fa\u9a7e\u9a76\u52a8\u4f5c\uff0c\u7b80\u5316\u7cfb\u7edf\u67b6\u6784\u5e76\u63d0\u5347\u53ef\u6269\u5c55\u6027\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>2. \u53cc\u6570\u636e\u9a71\u52a8\u6a21\u62df\u5668\u65b9\u6cd5<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u91cd\u6295\u5f71\u6a21\u62df\u5668<\/strong>\uff08Reprojective Simulation\uff09\uff1a\u901a\u8fc7\u6df1\u5ea6\u56fe\u4e0e\u59ff\u6001\u4fe1\u606f\u751f\u6210\u65b0\u89c6\u89d2\u56fe\u50cf\uff0c\u4f46\u53d7\u9650\u4e8e\u9759\u6001\u573a\u666f\u5047\u8bbe\u3001\u6df1\u5ea6\u8bef\u5dee\u53ca\u5149\u7167\u4f2a\u5f71\u7b49\u95ee\u9898\u3002<\/li>\n\n\n\n<li><strong>\u4e16\u754c\u6a21\u578b\u6a21\u62df\u5668<\/strong>\uff08World Model Simulation\uff09\uff1a\u57fa\u4e8e\u6269\u6563\u53d8\u6362\u5668\uff08Diffusion Transformer, DiT\uff09\u548c\u672a\u6765\u951a\u5b9a\uff08Future Anchoring\uff09\u6280\u672f\uff0c\u9884\u6d4b\u52a8\u6001\u573a\u666f\u7684\u672a\u6765\u72b6\u6001\u4e0e\u8f68\u8ff9\uff0c\u652f\u6301\u7b56\u7565\u5728\u975e\u72ec\u7acb\u540c\u5206\u5e03\uff08non-i.i.d.\uff09\u6570\u636e\u4e0b\u7684\u8bad\u7ec3\uff0c\u5e76\u751f\u6210\u7b26\u5408\u4eba\u7c7b\u9a7e\u9a76\u884c\u4e3a\u7684\u89c6\u9891\u5e8f\u5217\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>3. \u672a\u6765\u951a\u5b9a\u673a\u5236\uff08Future Anchoring\uff09<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5728World Model\u4e2d\u5f15\u5165\u672a\u6765\u76ee\u6807\u72b6\u6001F\u4f5c\u4e3a\u6761\u4ef6\uff0c\u901a\u8fc7\u89c4\u5212\u5934\uff08Plan Head\uff09\u751f\u6210\u6536\u655b\u4e8e\u76ee\u6807\u7684\u8f68\u8ff9\uff0c\u89e3\u51b3\u7b56\u7565\u5728\u9519\u8bef\u72b6\u6001\u4e0b\u7684\u6062\u590d\u95ee\u9898\uff08Recovery Pressure\uff09\uff0c\u63d0\u5347\u95ed\u73af\u6a21\u62df\u7684\u7a33\u5b9a\u6027\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>4. \u5728\u7b56\u7565\u8bad\u7ec3\uff08On-Policy Learning\uff09\u4e0e\u4fe1\u606f\u74f6\u9888<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u91c7\u7528\u7c7bIMPALA\u67b6\u6784\uff0c\u901a\u8fc7\u5e76\u884c\u6267\u884c\u5668\uff08Actors\uff09\u4e0e\u4e2d\u5fc3\u5b66\u4e60\u8005\uff08Learner\uff09\u4ea4\u4e92\uff0c\u4f7f\u7b56\u7565\u4ece\u81ea\u8eab\u4ea4\u4e92\u4e2d\u5b66\u4e60\u7ea0\u9519\u3002<\/li>\n\n\n\n<li>\u5f15\u5165\u4fe1\u606f\u74f6\u9888\uff08Information Bottleneck\uff09\uff0c\u901a\u8fc7\u6dfb\u52a0\u9ad8\u65af\u566a\u58f0\u9650\u5236\u7279\u5f81\u63d0\u53d6\u5668\u7684\u4fe1\u606f\u5bb9\u91cf\uff08\u7ea6700 bits\uff09\uff0c\u6291\u5236\u7b56\u7565\u5bf9\u6a21\u62df\u5668\u4f2a\u5f71\u7684\u8fc7\u62df\u5408\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>5. \u6269\u6563\u5efa\u6a21\u4e0e\u566a\u58f0\u589e\u5f3a\u6280\u672f<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u57fa\u4e8eRectified Flow\u76ee\u6807\u7684\u6269\u6563\u6a21\u578b\u751f\u6210\u73af\u5883\u72b6\u6001\uff0c\u7ed3\u5408\u591a\u5047\u8bbe\u89c4\u5212\u635f\u5931\uff08MHP Loss\uff09\u4f18\u5316\u8f68\u8ff9\u9884\u6d4b\u3002<\/li>\n\n\n\n<li>\u91c7\u7528\u566a\u58f0\u6c34\u5e73\u589e\u5f3a\uff08Noise Level Augmentation\uff09\uff0c\u7f13\u89e3\u81ea\u56de\u5f52\u6f02\u79fb\uff08Auto-regressive Drift\uff09\u95ee\u9898\uff0c\u63d0\u5347\u6a21\u578b\u5bf9\u7d2f\u79ef\u8bef\u5dee\u7684\u9c81\u68d2\u6027\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>6. \u5b9e\u9645\u90e8\u7f72\u9a8c\u8bc1<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5728\u5f00\u6e90ADAS\u7cfb\u7edf\uff08openpilot\uff09\u4e2d\u90e8\u7f72\u7b56\u7565\uff0c\u5b9e\u73b0\u771f\u5b9e\u8f66\u8f86\u7684\u6a2a\u5411\u63a7\u5236\uff08\u5982\u8f66\u9053\u4fdd\u6301\u3001\u53d8\u9053\uff09\u3002\u5b9e\u9a8c\u8868\u660e\uff0cWorld Model\u7b56\u7565\u5728\u95ed\u5faa\u73af\u6d4b\u8bd5\uff08MetaDrive\uff09\u4e2d\u901a\u8fc7\u7387\u8d8590%\uff0c\u5e76\u5728\u771f\u5b9e\u8def\u6d4b\u4e2d\u8fbe\u523052.49%\u7684\u8fde\u7eed\u4ecb\u5165\u91cc\u7a0b\u5360\u6bd4\uff0c\u9a8c\u8bc1\u4e86\u65b9\u6cd5\u7684\u5b9e\u7528\u6027\u4e0e\u53ef\u6269\u5c55\u6027\u3002<\/li>\n<\/ul>\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-28521","post","type-post","status-publish","format-standard","hentry","category-technology-frontier","category-home"],"views":3,"_links":{"self":[{"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/posts\/28521"}],"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=28521"}],"version-history":[{"count":1,"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/posts\/28521\/revisions"}],"predecessor-version":[{"id":28522,"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/posts\/28521\/revisions\/28522"}],"wp:attachment":[{"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/media?parent=28521"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/categories?post=28521"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/tags?post=28521"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}