{"id":20697,"date":"2024-06-21T10:45:57","date_gmt":"2024-06-21T02:45:57","guid":{"rendered":"https:\/\/aif.amtbbs.org\/?p=20697"},"modified":"2024-06-21T10:45:57","modified_gmt":"2024-06-21T02:45:57","slug":"%e5%8c%97%e5%a4%a7%e6%8e%a8%e5%87%ba%e5%85%a8%e6%96%b0%e6%9c%ba%e5%99%a8%e4%ba%ba%e5%a4%9a%e6%a8%a1%e6%80%81%e5%a4%a7%e6%a8%a1%e5%9e%8b%ef%bc%81%e9%9d%a2%e5%90%91%e9%80%9a%e7%94%a8%e5%92%8c%e6%9c%ba","status":"publish","type":"post","link":"https:\/\/aif.amtbbs.org\/index.php\/2024\/06\/21\/20697\/","title":{"rendered":"\u5317\u5927\u63a8\u51fa\u5168\u65b0\u673a\u5668\u4eba\u591a\u6a21\u6001\u5927\u6a21\u578b\uff01\u9762\u5411\u901a\u7528\u548c\u673a\u5668\u4eba\u573a\u666f\u7684\u9ad8\u6548\u63a8\u7406\u548c\u64cd\u4f5c"},"content":{"rendered":"<div><img data-dominant-color=\"746c65\" data-has-transparency=\"false\" style=\"--dominant-color: #746c65;\" loading=\"lazy\" decoding=\"async\" class=\"not-transparent alignnone size-full wp-image-20698\" src=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2024\/06\/542578467c254a6946b989bb098d604f4b5731.jpg\" width=\"416\" height=\"232\" srcset=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2024\/06\/542578467c254a6946b989bb098d604f4b5731.jpg 416w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2024\/06\/542578467c254a6946b989bb098d604f4b5731-300x167.jpg 300w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2024\/06\/542578467c254a6946b989bb098d604f4b5731-150x84.jpg 150w\" sizes=\"auto, (max-width: 416px) 100vw, 416px\" \/><\/div>\n<div class=\"article-desc\">\u4e3a\u4e86\u8d4b\u4e88\u673a\u5668\u4eba\u7aef\u5230\u7aef\u7684\u63a8\u7406\u548c\u64cd\u7eb5\u80fd\u529b\uff0c\u672c\u6587\u521b\u65b0\u6027\u5730\u5c06\u89c6\u89c9\u7f16\u7801\u5668\u4e0e\u9ad8\u6548\u7684\u72b6\u6001\u7a7a\u95f4\u8bed\u8a00\u6a21\u578b\u96c6\u6210\uff0c\u6784\u5efa\u4e86\u5168\u65b0\u7684 RoboMamba \u591a\u6a21\u6001\u5927\u6a21\u578b\uff0c\u4f7f\u5176\u5177\u5907\u89c6\u89c9\u5e38\u8bc6\u4efb\u52a1\u548c\u673a\u5668\u4eba\u76f8\u5173\u4efb\u52a1\u7684\u63a8\u7406\u80fd\u529b\uff0c\u5e76\u90fd\u53d6\u5f97\u4e86\u5148\u8fdb\u7684\u6027\u80fd\u8868\u73b0\u3002<\/div>\n<div id=\"postspictures\" class=\"article-content\">\n<div id=\"container\" class=\"container am-engine\" data-v-1d7a5742=\"\" data-element=\"root\">\n<p><strong>\u672c\u6587\u7531 HMI Lab \u5b8c\u6210\u3002HMI Lab\u4f9d\u6258\u5317\u4eac\u5927\u5b66\u89c6\u9891\u4e0e\u89c6\u89c9\u6280\u672f\u56fd\u5bb6\u5de5\u7a0b\u7814\u7a76\u4e2d\u5fc3\u548c\u591a\u5a92\u4f53\u4fe1\u606f\u5904\u7406\u5168\u56fd\u91cd\u70b9\u5b9e\u9a8c\u5ba4\u4e24\u5927\u5e73\u53f0\uff0c\u957f\u671f\u4ece\u4e8b\u673a\u5668\u5b66\u4e60\u3001\u591a\u6a21\u6001\u5b66\u4e60\u548c\u5177\u8eab\u667a\u80fd\u65b9\u5411\u7684\u7814\u7a76\u3002\u672c\u5de5\u4f5c\u7b2c\u4e00\u4f5c\u8005\u4e3a\u5218\u5bb6\u94ed\u535a\u58eb\uff0c\u7814\u7a76\u65b9\u5411\u4e3a\u9762\u5411\u5f00\u653e\u4e16\u754c\u7684\u591a\u6a21\u6001\u5177\u8eab\u5927\u6a21\u578b\u4e0e\u6301\u7eed\u6027\u5b66\u4e60\u6280\u672f\u3002\u672c\u5de5\u4f5c\u7b2c\u4e8c\u4f5c\u8005\u4e3a\u5218\u68a6\u771f\uff0c\u7814\u7a76\u65b9\u5411\u4e3a\u89c6\u89c9\u57fa\u7840\u6a21\u578b\u4e0e\u673a\u5668\u4eba\u64cd\u7eb5\u3002\u6307\u5bfc\u8001\u5e08\u4e3a\u4ec9\u5c1a\u822a\uff0c\u5317\u4eac\u5927\u5b66\u8ba1\u7b97\u673a\u5b66\u9662\u7814\u7a76\u5458\u3001\u535a\u58eb\u751f\u5bfc\u5e08\u3001\u535a\u96c5\u9752\u5e74\u5b66\u8005\u3002\u4ece\u4e8b\u591a\u6a21\u6001\u5927\u6a21\u578b\u4e0e\u5177\u8eab\u667a\u80fd\u7814\u7a76\uff0c\u53d6\u5f97\u4e86\u4e00\u7cfb\u5217\u91cd\u8981\u7814\u7a76\u6210\u679c\uff0c\u5728\u4eba\u5de5\u667a\u80fd\u9876\u7ea7\u671f\u520a\u548c\u4f1a\u8bae\u4e0a\u53d1\u8868\u8bba\u6587 80 \u4f59\u7bc7\uff0c\u8c37\u6b4c\u5f15\u7528 9700 \u4f59\u6b21\u3002\u8363\u83b7\u4e16\u754c\u4eba\u5de5\u667a\u80fd\u9876\u4f1a AAAI \u6700\u4f73\u8bba\u6587\u5956\uff0c\u4f4d\u5217\u4e16\u754c\u6700\u5927\u5b66\u672f\u6e90\u4ee3\u7801\u4ed3\u5e93 Trending Research \u7b2c\u4e00\u4f4d\u3002<\/strong><\/p>\n<p>\u4e3a\u4e86\u8d4b\u4e88\u673a\u5668\u4eba\u7aef\u5230\u7aef\u7684\u63a8\u7406\u548c\u64cd\u7eb5\u80fd\u529b\uff0c\u672c\u6587\u521b\u65b0\u6027\u5730\u5c06\u89c6\u89c9\u7f16\u7801\u5668\u4e0e\u9ad8\u6548\u7684\u72b6\u6001\u7a7a\u95f4\u8bed\u8a00\u6a21\u578b\u96c6\u6210\uff0c\u6784\u5efa\u4e86\u5168\u65b0\u7684 RoboMamba \u591a\u6a21\u6001\u5927\u6a21\u578b\uff0c\u4f7f\u5176\u5177\u5907\u89c6\u89c9\u5e38\u8bc6\u4efb\u52a1\u548c\u673a\u5668\u4eba\u76f8\u5173\u4efb\u52a1\u7684\u63a8\u7406\u80fd\u529b\uff0c\u5e76\u90fd\u53d6\u5f97\u4e86\u5148\u8fdb\u7684\u6027\u80fd\u8868\u73b0\u3002\u540c\u65f6\uff0c\u672c\u6587\u53d1\u73b0\u5f53 RoboMamba \u5177\u5907\u5f3a\u5927\u7684\u63a8\u7406\u80fd\u529b\u540e\uff0c\u6211\u4eec\u53ef\u4ee5\u901a\u8fc7\u6781\u4f4e\u7684\u8bad\u7ec3\u6210\u672c\u4f7f\u5f97 RoboMamba \u638c\u63e1\u591a\u79cd\u64cd\u7eb5\u4f4d\u59ff\u9884\u6d4b\u80fd\u529b\u3002<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s9.51cto.com\/oss\/202406\/20\/f150653456a2181b870932a8ee6fd02b61bb6d.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<ul data-id=\"ua73dd4b-7gWH1gFT\">\n<li data-id=\"ld70c578-QZ7DZSF4\">\u8bba\u6587\uff1aRoboMamba: Multimodal State Space Model for Efficient Robot Reasoning and Manipulation<\/li>\n<li data-id=\"ld70c578-AOEwAeOm\">\u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/arxiv.org\/abs\/2406.04339<\/li>\n<li data-id=\"ld70c578-YZKoeVWt\">\u9879\u76ee\u4e3b\u9875\uff1ahttps:\/\/sites.google.com\/view\/robomamba-web<\/li>\n<li data-id=\"ld70c578-MWhyW08C\">Github\uff1ahttps:\/\/github.com\/lmzpai\/roboMamba<\/li>\n<\/ul>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s5.51cto.com\/oss\/202406\/20\/7615565989c09dfd6d51784a81091f13ae3f28.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u56fe 1. RoboMamba \u5177\u5907\u7684\u673a\u5668\u4eba\u76f8\u5173\u80fd\u529b\uff0c\u5176\u4e2d\u5305\u62ec\u4efb\u52a1\u89c4\u5212\u3001\u63d0\u793a\u6027\u4efb\u52a1\u89c4\u5212\u3001\u957f\u7a0b\u4efb\u52a1\u89c4\u5212\u3001\u53ef\u64cd\u7eb5\u6027\u5224\u65ad\u3001\u53ef\u64cd\u7eb5\u6027\u751f\u6210\u3001\u672a\u6765\u4e0e\u8fc7\u53bb\u9884\u6d4b\u3001\u672b\u7aef\u6267\u884c\u5668\u4f4d\u59ff\u9884\u6d4b\u7b49\u3002<\/p>\n<h4>\u6458\u8981<\/h4>\n<p>\u673a\u5668\u4eba\u64cd\u7eb5\u7684\u4e00\u4e2a\u57fa\u672c\u76ee\u6807\u662f\u4f7f\u6a21\u578b\u80fd\u591f\u7406\u89e3\u89c6\u89c9\u573a\u666f\u5e76\u6267\u884c\u52a8\u4f5c\u3002\u5c3d\u7ba1\u73b0\u6709\u7684\u673a\u5668\u4eba\u591a\u6a21\u6001\u5927\u6a21\u578b\uff08MLLM\uff09\u53ef\u4ee5\u5904\u7406\u4e00\u7cfb\u5217\u57fa\u672c\u4efb\u52a1\uff0c\u4f46\u5b83\u4eec\u4ecd\u7136\u9762\u4e34\u4e24\u4e2a\u65b9\u9762\u7684\u6311\u6218\uff1a1) \u5904\u7406\u590d\u6742\u4efb\u52a1\u7684\u63a8\u7406\u80fd\u529b\u4e0d\u8db3\uff1b2) MLLM \u5fae\u8c03\u548c\u63a8\u7406\u7684\u8ba1\u7b97\u6210\u672c\u8f83\u9ad8\u3002\u6700\u8fd1\u63d0\u51fa\u7684\u72b6\u6001\u7a7a\u95f4\u6a21\u578b\uff08SSM\uff09\uff0c\u5373 Mamba\uff0c\u5176\u5177\u5907\u7ebf\u6027\u63a8\u7406\u590d\u6742\u5ea6\u540c\u65f6\u5728\u5e8f\u5217\u5efa\u6a21\u4e2d\u5c55\u793a\u4e86\u4ee4\u4eba\u671f\u5f85\u7684\u80fd\u529b\u3002\u53d7\u6b64\u542f\u53d1\uff0c\u6211\u4eec\u63a8\u51fa\u4e86\u7aef\u5230\u7aef\u673a\u5668\u4eba MLLM\u2014RoboMamba\uff0c\u5b83\u5229\u7528 Mamba \u6a21\u578b\u63d0\u4f9b\u673a\u5668\u4eba\u63a8\u7406\u548c\u884c\u52a8\u80fd\u529b\uff0c\u540c\u65f6\u4fdd\u6301\u9ad8\u6548\u7684\u5fae\u8c03\u548c\u63a8\u7406\u80fd\u529b\u3002<\/p>\n<p>\u5177\u4f53\u6765\u8bf4\uff0c\u6211\u4eec\u9996\u5148\u5c06\u89c6\u89c9\u7f16\u7801\u5668\u4e0e Mamba \u96c6\u6210\u5728\u4e00\u8d77\uff0c\u901a\u8fc7\u5171\u540c\u8bad\u7ec3\u5c06\u89c6\u89c9\u6570\u636e\u4e0e\u8bed\u8a00\u5d4c\u5165\u5bf9\u9f50\uff0c\u4f7f\u6211\u4eec\u7684\u6a21\u578b\u5177\u6709\u89c6\u89c9\u5e38\u8bc6\u548c\u4e0e\u673a\u5668\u4eba\u76f8\u5173\u7684\u63a8\u7406\u80fd\u529b\u3002\u4e3a\u4e86\u8fdb\u4e00\u6b65\u589e\u5f3a RoboMamba \u7684\u64cd\u7eb5\u4f4d\u59ff\u9884\u6d4b\u80fd\u529b\uff0c\u6211\u4eec\u63a2\u7d22\u4e86\u4e00\u79cd\u4ec5\u4f7f\u7528\u7b80\u5355 Policy Head \u7684\u9ad8\u6548\u5fae\u8c03\u7b56\u7565\u3002\u6211\u4eec\u53d1\u73b0\uff0c\u4e00\u65e6 RoboMamba \u62e5\u6709\u8db3\u591f\u7684\u63a8\u7406\u80fd\u529b\uff0c\u5b83\u53ef\u4ee5\u901a\u8fc7\u6781\u5c11\u7684\u5fae\u8c03\u53c2\u6570\uff08\u6a21\u578b\u7684 0.1%\uff09\u548c\u5fae\u8c03\u65f6\u95f4\uff0820 \u5206\u949f\uff09\u6765\u638c\u63e1\u591a\u79cd\u64cd\u4f5c\u6280\u80fd\u3002\u5728\u5b9e\u9a8c\u4e2d\uff0cRoboMamba \u5728\u901a\u7528\u548c\u673a\u5668\u4eba\u8bc4\u4f30\u57fa\u51c6\u4e0a\u5c55\u793a\u4e86\u51fa\u8272\u7684\u63a8\u7406\u80fd\u529b\uff0c\u5982\u56fe 2 \u6240\u793a\u3002\u540c\u65f6\uff0c\u6211\u4eec\u7684\u6a21\u578b\u5728\u6a21\u62df\u548c\u73b0\u5b9e\u4e16\u754c\u5b9e\u9a8c\u4e2d\u5c55\u793a\u4e86\u4ee4\u4eba\u5370\u8c61\u6df1\u523b\u7684\u64cd\u7eb5\u4f4d\u59ff\u9884\u6d4b\u80fd\u529b\uff0c\u5176\u63a8\u7406\u901f\u5ea6\u6bd4\u73b0\u6709\u7684\u673a\u5668\u4eba MLLMs \u5feb 7 \u500d\u3002<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s5.51cto.com\/oss\/202406\/20\/573411b7162f052f64f897a0396a47ebc1f1b7.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u56fe 2. \u6982\u8ff0\uff1aRobomamba \u662f\u4e00\u79cd\u9ad8\u6548\u7684\u673a\u5668\u4eba\u591a\u6a21\u6001\u5927\u6a21\u578b\uff0c\u540c\u65f6\u5177\u5907\u5f3a\u5927\u7684\u63a8\u7406\u548c\u64cd\u4f5c\u80fd\u529b\u3002RoboMamba-2.8B \u5728\u901a\u7528 MLLM \u57fa\u51c6\u4e0a\u5b9e\u73b0\u4e86\u4e0e\u5176\u4ed6 7B MLLM \u53ef\u7ade\u4e89\u7684\u63a8\u7406\u6027\u80fd\uff0c\u540c\u65f6\u5728\u673a\u5668\u4eba\u4efb\u52a1\u4e2d\u5c55\u793a\u4e86\u957f\u7a0b\u63a8\u7406\u80fd\u529b\u3002\u968f\u540e\uff0c\u6211\u4eec\u5f15\u5165\u4e86\u4e00\u79cd\u6781\u5176\u9ad8\u6548\u7684\u5fae\u8c03\u7b56\u7565\uff0c\u4f7f RoboMamba \u5177\u5907\u64cd\u7eb5\u4f4d\u59ff\u9884\u6d4b\u80fd\u529b\uff0c\u53ea\u9700 20 \u5206\u949f\u5373\u53ef\u5fae\u8c03\u4e00\u4e2a\u7b80\u5355\u7684\u7b56\u7565\u5934\u3002<\/p>\n<p>\u672c\u6587\u4e3b\u8981\u8d21\u732e\u603b\u7ed3\u5982\u4e0b\uff1a<\/p>\n<ul data-id=\"ua73dd4b-Rol42B9O\">\n<li data-id=\"ld70c578-8NNEY1Vc\">\u6211\u4eec\u521b\u65b0\u5730\u5c06\u89c6\u89c9\u7f16\u7801\u5668\u4e0e\u9ad8\u6548\u7684 Mamba \u8bed\u8a00\u6a21\u578b\u96c6\u6210\uff0c\u6784\u5efa\u4e86\u5168\u65b0\u7684\u7aef\u5230\u7aef\u673a\u5668\u4eba\u591a\u6a21\u6001\u5927\u6a21\u578b\uff0cRoboMamba\uff0c\u5176\u5177\u5907\u89c6\u89c9\u5e38\u8bc6\u548c\u673a\u5668\u4eba\u76f8\u5173\u7684\u5168\u9762\u63a8\u7406\u80fd\u529b\u3002<\/li>\n<li data-id=\"ld70c578-M3XoNx8j\">\u4e3a\u4e86\u4f7f RoboMamba \u5177\u5907\u672b\u7aef\u6267\u884c\u5668\u64cd\u7eb5\u4f4d\u59ff\u9884\u6d4b\u80fd\u529b\uff0c\u6211\u4eec\u63a2\u7d22\u4e86\u4e00\u79cd\u4f7f\u7528\u7b80\u5355 Policy Head \u7684\u9ad8\u6548\u5fae\u8c03\u7b56\u7565\u3002\u6211\u4eec\u53d1\u73b0\uff0c\u4e00\u65e6 RoboMamba \u8fbe\u5230\u8db3\u591f\u7684\u63a8\u7406\u80fd\u529b\uff0c\u5b83\u53ef\u4ee5\u4ee5\u6781\u4f4e\u7684\u6210\u672c\u638c\u63e1\u64cd\u7eb5\u4f4d\u59ff\u9884\u6d4b\u6280\u80fd\u3002<\/li>\n<li data-id=\"ld70c578-fHWLq9qs\">\u5728\u6211\u4eec\u7684\u5927\u91cf\u5b9e\u9a8c\u4e2d\uff0cRoboMamba \u5728\u901a\u7528\u548c\u673a\u5668\u4eba\u63a8\u7406\u8bc4\u4f30\u57fa\u51c6\u4e0a\u8868\u73b0\u51fa\u8272\uff0c\u5e76\u5728\u6a21\u62df\u5668\u548c\u771f\u5b9e\u4e16\u754c\u5b9e\u9a8c\u4e2d\u5c55\u793a\u4e86\u4ee4\u4eba\u5370\u8c61\u6df1\u523b\u7684\u4f4d\u59ff\u9884\u6d4b\u7ed3\u679c\u3002<\/li>\n<\/ul>\n<h4>\u7814\u7a76\u80cc\u666f<\/h4>\n<p>\u6570\u636e\u7684 scaling up \u663e\u8457\u63a8\u52a8\u4e86\u5927\u8bed\u8a00\u6a21\u578b\uff08LLMs\uff09\u7814\u7a76\u7684\u53d1\u5c55\uff0c\u5c55\u793a\u4e86\u5728\u81ea\u7136\u8bed\u8a00\u5904\u7406\uff08NLP\uff09\u4e2d\u63a8\u7406\u548c\u6cdb\u5316\u80fd\u529b\u7684\u663e\u8457\u8fdb\u6b65\u3002\u4e3a\u4e86\u7406\u89e3\u591a\u6a21\u6001\u4fe1\u606f\uff0c\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b\uff08MLLMs\uff09\u5e94\u8fd0\u800c\u751f\uff0c\u8d4b\u4e88 LLMs \u89c6\u89c9\u6307\u4ee4\u8ddf\u968f\u548c\u573a\u666f\u7406\u89e3\u7684\u80fd\u529b\u3002\u53d7 MLLMs \u5728\u901a\u7528\u73af\u5883\u4e2d\u5f3a\u5927\u80fd\u529b\u7684\u542f\u53d1\uff0c\u8fd1\u671f\u7814\u7a76\u65e8\u5728\u5c06 MLLMs \u5e94\u7528\u4e8e\u673a\u5668\u4eba\u64cd\u4f5c\u9886\u57df\u3002\u4e00\u4e9b\u7814\u7a76\u5de5\u4f5c\u4f7f\u673a\u5668\u4eba\u80fd\u591f\u7406\u89e3\u81ea\u7136\u8bed\u8a00\u548c\u89c6\u89c9\u573a\u666f\uff0c\u81ea\u52a8\u751f\u6210\u4efb\u52a1\u8ba1\u5212\u3002\u53e6\u4e00\u4e9b\u7814\u7a76\u5de5\u4f5c\u5219\u662f\u5229\u7528 MLLMs \u7684\u56fa\u6709\u80fd\u529b\uff0c\u4f7f\u5176\u5177\u5907\u9884\u6d4b\u64cd\u4f5c\u4f4d\u59ff\u7684\u80fd\u529b\u3002<\/p>\n<p>\u673a\u5668\u4eba\u64cd\u4f5c\u6d89\u53ca\u5728\u52a8\u6001\u73af\u5883\u4e2d\u4e0e\u7269\u4f53\u4ea4\u4e92\uff0c\u9700\u8981\u7c7b\u4eba\u63a8\u7406\u80fd\u529b\u4ee5\u7406\u89e3\u573a\u666f\u7684\u8bed\u4e49\u4fe1\u606f\uff0c\u4ee5\u53ca\u5f3a\u5927\u7684\u64cd\u7eb5\u4f4d\u59ff\u9884\u6d4b\u80fd\u529b\u3002\u867d\u7136\u73b0\u6709\u57fa\u4e8e\u673a\u5668\u4eba MLLM \u53ef\u4ee5\u5904\u7406\u4e00\u7cfb\u5217\u57fa\u7840\u4efb\u52a1\uff0c\u4f46\u5b83\u4eec\u5728\u4e24\u4e2a\u65b9\u9762\u4ecd\u7136\u9762\u4e34\u6311\u6218\u3002<\/p>\n<p>1\uff09\u9996\u5148\uff0c\u9884\u8bad\u7ec3\u7684 MLLMs \u5728\u673a\u5668\u4eba\u573a\u666f\u4e2d\u7684\u63a8\u7406\u80fd\u529b\u88ab\u53d1\u73b0\u662f\u4e0d\u8db3\u7684\u3002\u6b63\u5982\u56fe 2<\/p>\n<p>\u6240\u793a\uff0c\u5f53\u5fae\u8c03\u540e\u7684\u673a\u5668\u4eba MLLMs \u9047\u5230\u590d\u6742\u63a8\u7406\u4efb\u52a1\u65f6\uff0c\u8fd9\u79cd\u7f3a\u9677\u4f1a\u5e26\u6765\u6311\u6218\u3002<\/p>\n<p>2\uff09\u5176\u6b21\uff0c\u7531\u4e8e\u73b0\u6709 MLLM \u6ce8\u610f\u529b\u673a\u5236\u7684\u8ba1\u7b97\u590d\u6742\u5ea6\u8f83\u9ad8\uff0c\u5fae\u8c03 MLLMs \u5e76\u4f7f\u7528\u5b83\u4eec\u751f\u6210\u673a\u5668\u4eba\u64cd\u4f5c\u52a8\u4f5c\u4f1a\u4ea7\u751f\u66f4\u9ad8\u7684\u8ba1\u7b97\u6210\u672c\u3002<\/p>\n<p>\u4e3a\u4e86\u5e73\u8861\u63a8\u7406\u80fd\u529b\u548c\u6548\u7387\uff0cNLP \u9886\u57df\u51fa\u73b0\u4e86\u51e0\u9879\u7814\u7a76\u3002\u5c24\u5176\u662f\uff0cMamba \u5f15\u5165\u4e86\u521b\u65b0\u7684\u9009\u62e9\u6027\u72b6\u6001\u7a7a\u95f4\u6a21\u578b\uff08SSM\uff09\uff0c\u5728\u4fdd\u6301\u7ebf\u6027\u590d\u6742\u5ea6\u7684\u540c\u65f6\uff0c\u4fc3\u8fdb\u4e86\u4e0a\u4e0b\u6587\u611f\u77e5\u7684\u63a8\u7406\u3002<\/p>\n<p>\u53d7\u6b64\u542f\u53d1\uff0c\u6211\u4eec\u63d0\u51fa\u4e00\u4e2a\u95ee\u9898\uff1a\u201c\u6211\u4eec\u80fd\u5426\u5f00\u53d1\u51fa\u4e00\u79cd\u9ad8\u6548\u7684\u673a\u5668\u4eba MLLM\uff0c\u65e2\u5177\u5907\u5f3a\u5927\u7684\u63a8\u7406\u80fd\u529b\uff0c\u53c8\u80fd\u4ee5\u975e\u5e38\u7ecf\u6d4e\u7684\u65b9\u5f0f\u83b7\u5f97\u673a\u5668\u4eba\u64cd\u4f5c\u6280\u80fd\uff1f\u201d<\/p>\n<h4>RoboMamba \u65b9\u6cd5<\/h4>\n<p><strong>1. \u80cc\u666f\u77e5\u8bc6<\/strong><\/p>\n<ul data-id=\"ua73dd4b-OaUaU59W\">\n<li data-id=\"ld70c578-Eu1p5DWn\">\u95ee\u9898\u9648\u8ff0<\/li>\n<\/ul>\n<p>\u5bf9\u4e8e\u673a\u5668\u4eba\u89c6\u89c9\u63a8\u7406\uff0c\u6211\u4eec\u7684 RoboMamba \u57fa\u4e8e\u56fe\u50cf<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s7.51cto.com\/oss\/202406\/20\/82800f342d79d47a7733912326d2d535c9e73b.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>\u548c\u8bed\u8a00\u95ee\u9898<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s4.51cto.com\/oss\/202406\/20\/35fc8d447c02c6741d964362770de167bbd792.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>\u751f\u6210\u8bed\u8a00\u7b54\u6848<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s5.51cto.com\/oss\/202406\/20\/310b89b71a9b8b558ff738dc91a54452b417df.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>\uff0c\u8868\u793a\u4e3a<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s5.51cto.com\/oss\/202406\/20\/e9732f143e1d998b13d874377df120227faca4.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>\u3002\u63a8\u7406\u7b54\u6848\u901a\u5e38\u5305\u542b\u5355\u72ec\u7684\u5b50\u4efb\u52a1<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s6.51cto.com\/oss\/202406\/20\/55787c145eaadb2cd435872d366092af1e4ace.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>\u5bf9\u4e8e\u4e00\u4e2a\u95ee\u9898<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s2.51cto.com\/oss\/202406\/20\/18d424806171d060356342411e2ca467026e1e.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>\u3002\u4f8b\u5982\uff0c\u5f53\u9762\u5bf9\u4e00\u4e2a\u8ba1\u5212\u95ee\u9898\uff0c\u5982 \u201c\u5982\u4f55\u6536\u62fe\u684c\u5b50\uff1f\u201d\uff0c\u53cd\u5e94\u901a\u5e38\u5305\u62ec \u201c\u7b2c\u4e00\u6b65\uff1a\u6361\u8d77\u7269\u4f53\u201d \u548c \u201c\u7b2c\u4e8c\u6b65\uff1a\u628a\u7269\u4f53\u653e\u5165\u76d2\u5b50\u201d \u7b49\u6b65\u9aa4\u3002\u5bf9\u4e8e\u52a8\u4f5c\u9884\u6d4b\uff0c\u6211\u4eec\u5229\u7528\u4e00\u4e2a\u9ad8\u6548\u7b80\u5355\u7684\u7b56\u7565\u5934 \u03c0 \u6765\u9884\u6d4b\u52a8\u4f5c<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s6.51cto.com\/oss\/202406\/20\/a4e923a04cd1e9f6d42019645c5c450c06f324.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>\u3002\u6839\u636e\u4e4b\u524d\u7684\u5de5\u4f5c\uff0c\u6211\u4eec\u4f7f\u7528 6-DoF \u6765\u8868\u8fbe Franka Emika Panda \u673a\u68b0\u81c2\u7684\u672b\u7aef\u6267\u884c\u5668\u4f4d\u59ff\u30026 \u81ea\u7531\u5ea6\u5305\u62ec\u672b\u7aef\u6267\u884c\u5668\u4f4d\u7f6e<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s9.51cto.com\/oss\/202406\/20\/94d213321b1a45715971551745c944e0c2257d.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>\u8868\u793a\u4e09\u7ef4\u5750\u6807\uff0c\u65b9\u5411<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s6.51cto.com\/oss\/202406\/20\/67e8775390cbc9f5e26537fe0d261056347bc2.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>\u8868\u793a\u65cb\u8f6c\u77e9\u9635\u3002\u5982\u679c\u8bad\u7ec3\u6293\u53d6\u4efb\u52a1\uff0c\u6211\u4eec\u5c06\u6293\u5939\u72b6\u6001\u6dfb\u52a0\u5230\u4f4d\u59ff\u9884\u6d4b\u4e2d\uff0c\u4ece\u800c\u5b9e\u73b0 7-DoF \u63a7\u5236\u3002<\/p>\n<ul data-id=\"ua73dd4b-5G8QfFhP\">\n<li data-id=\"ld70c578-vPZJ7ziV\">\u72b6\u6001\u7a7a\u95f4\u6a21\u578b (SSM)<\/li>\n<\/ul>\n<p>\u672c\u6587\u9009\u62e9 Mamba \u4f5c\u4e3a\u5927\u8bed\u8a00\u6a21\u578b\u3002Mamba \u7531\u8bb8\u591a Mamba block \u7ec4\u6210\uff0c\u6700\u5173\u952e\u7684\u7ec4\u6210\u90e8\u5206\u662f SSM\u3002SSM \u662f\u57fa\u4e8e\u8fde\u7eed\u7cfb\u7edf\u8bbe\u8ba1\u7684\uff0c\u901a\u8fc7\u9690\u85cf\u72b6\u6001<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s2.51cto.com\/oss\/202406\/20\/15d7c4d01113625deca460df6f678691c449a6.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>\uff0c\u5c06 1D \u8f93\u5165\u5e8f\u5217<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s6.51cto.com\/oss\/202406\/20\/32f347912489e840cc50591efa8d2da52f6772.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>\u6295\u5f71\u5230 1D \u8f93\u51fa\u5e8f\u5217<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s3.51cto.com\/oss\/202406\/20\/657298d6607737484fd493664988e48ed1e523.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>\u3002SSM \u7531\u4e09\u4e2a\u5173\u952e\u53c2\u6570\u7ec4\u6210\uff1a\u72b6\u6001\u77e9\u9635<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s4.51cto.com\/oss\/202406\/20\/026b67373f28b7d9b21726b88daeaddbd165fd.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>\uff0c\u8f93\u5165\u77e9\u9635<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s8.51cto.com\/oss\/202406\/20\/e521beb35d1e64a3561275c4aa14bc2dff2a92.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>\uff0c\u8f93\u51fa\u77e9\u9635<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s5.51cto.com\/oss\/202406\/20\/c4a753d71f55872519a420a54ab2daecad1c28.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>\u3002SSM \u53ef\u4ee5\u8868\u793a\u4e3a:<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s7.51cto.com\/oss\/202406\/20\/a9db538758d5fd683ea7662f39863d56ae967f.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u6700\u8fd1\u7684 SSM (\u4f8b\u5982\uff0cMamba) \u88ab\u6784\u9020\u4e3a\u4f7f\u7528\u65f6\u95f4\u5c3a\u5ea6\u53c2\u6570\u2206\u7684\u79bb\u6563\u8fde\u7eed\u7cfb\u7edf\u3002\u8be5\u53c2\u6570\u5c06\u8fde\u7eed\u53c2\u6570 A \u548c B \u8f6c\u6362\u4e3a\u79bb\u6563\u53c2\u6570<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s9.51cto.com\/oss\/202406\/20\/54f6d2e73762b575455868d80a2173dfc4bdd6.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>\u548c<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s4.51cto.com\/oss\/202406\/20\/2762d2e18f186a5ccf95082e0c7569408ca513.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>\u3002\u79bb\u6563\u5316\u91c7\u7528\u96f6\u9636\u4fdd\u6301\u65b9\u6cd5\uff0c\u5b9a\u4e49\u5982\u4e0b:<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s9.51cto.com\/oss\/202406\/20\/f83e7ef58d68111d800048a891df55b4eea97f.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>Mamba \u5f15\u5165\u4e86\u9009\u62e9\u6027\u626b\u63cf\u673a\u5236 (S6)\uff0c\u5728\u6bcf\u4e2a Mamba block \u4e2d\u5f62\u6210\u5176 SSM \u64cd\u4f5c\u3002SSM \u53c2\u6570\u66f4\u65b0\u4e3a<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s8.51cto.com\/oss\/202406\/20\/38b9a73721e06985f43843f0d384b0c9cdf578.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>\uff0c\u5b9e\u73b0\u66f4\u597d\u7684\u5185\u5bb9\u611f\u77e5\u63a8\u7406\u3002\u4e0b\u56fe 3 \u4e2d\u5c55\u793a\u4e86 Mamba block \u7684\u8be6\u7ec6\u4fe1\u606f\u3002<\/p>\n<p><strong>2. RoboMamba \u6a21\u578b\u7ed3\u6784<\/strong><\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s2.51cto.com\/oss\/202406\/20\/181e3e373a22baa24325434190b8cd2a02b2d2.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u56fe 3.\u00a0<strong>Robomamba \u6574\u4f53\u6846\u67b6\u3002<\/strong>RoboMamba \u901a\u8fc7\u89c6\u89c9\u7f16\u7801\u5668\u548c\u6295\u5f71\u5c42\u5c06\u56fe\u50cf\u6295\u5f71\u5230 Mamba \u7684\u8bed\u8a00\u5d4c\u5165\u7a7a\u95f4\uff0c\u7136\u540e\u4e0e\u6587\u672c tokens \u8fde\u63a5\uff0c\u5e76\u8f93\u5165\u5230 Mamba \u6a21\u578b\u4e2d\u3002\u4e3a\u4e86\u9884\u6d4b\u672b\u7aef\u6267\u884c\u5668\u7684\u4f4d\u7f6e\u548c\u65b9\u5411\uff0c\u6211\u4eec\u5f15\u5165\u7b80\u5355\u7684 MLP \u7b56\u7565\u5934\uff0c\u5e76\u4f7f\u7528\u6c60\u5316\u64cd\u4f5c\u4ece\u8bed\u8a00\u8f93\u51fa tokens \u751f\u6210\u7684\u5168\u5c40 token \u4f5c\u4e3a\u8f93\u5165\u3002<strong>RoboMamba \u7684\u8bad\u7ec3\u7b56\u7565\u3002<\/strong>\u4e3a\u4e86\u8fdb\u884c\u6a21\u578b\u8bad\u7ec3\uff0c\u6211\u4eec\u5c06\u8bad\u7ec3\u6d41\u7a0b\u5206\u4e3a\u4e24\u4e2a\u9636\u6bb5\u3002\u5728 Stage 1\uff0c\u6211\u4eec\u5f15\u5165\u5bf9\u9f50\u9884\u8bad\u7ec3\uff08Stage 1.1\uff09\u548c\u6307\u4ee4\u5171\u540c\u8bad\u7ec3\uff08Stage 1.2\uff09\uff0c\u4ee5\u4f7f RoboMamba \u5177\u5907\u5e38\u8bc6\u63a8\u7406\u548c\u673a\u5668\u4eba\u76f8\u5173\u7684\u63a8\u7406\u80fd\u529b\u3002\u5728 Stage 2\uff0c\u6211\u4eec\u63d0\u51fa\u673a\u5668\u4eba\u64cd\u4f5c\u5fae\u8c03\uff0c\u4ee5\u9ad8\u6548\u5730\u8d4b\u4e88 RoboMamba Low-Level \u64cd\u4f5c\u6280\u80fd\u3002<\/p>\n<p>\u4e3a\u4e86\u4f7f RoboMamba \u5177\u5907\u89c6\u89c9\u63a8\u7406\u548c\u64cd\u4f5c\u80fd\u529b\uff0c\u6211\u4eec\u4ece\u9884\u8bad\u7ec3\u7684\u5927\u8bed\u8a00\u6a21\u578b\uff08LLMs\uff09\u548c\u89c6\u89c9\u6a21\u578b\u5f00\u59cb\uff0c\u6784\u5efa\u4e86\u4e00\u4e2a\u9ad8\u6548\u7684 MLLM \u67b6\u6784\u3002\u5982\u4e0a\u56fe 3 \u6240\u793a\uff0c\u6211\u4eec\u5229\u7528 CLIP \u89c6\u89c9\u7f16\u7801\u5668\u4ece\u8f93\u5165\u56fe\u50cf I \u4e2d\u63d0\u53d6\u89c6\u89c9\u7279\u5f81<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s5.51cto.com\/oss\/202406\/20\/350b5086001fe8f35611814f1c3bd9b1545bdb.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>\uff0c\u5176\u4e2d B \u548c N \u5206\u522b\u8868\u793a batch size \u548c token \u6570\u3002\u4e0e\u6700\u8fd1\u7684 MLLMs \u4e0d\u540c\uff0c\u6211\u4eec\u4e0d\u91c7\u7528\u89c6\u89c9\u7f16\u7801\u5668\u96c6\u6210\u6280\u672f\uff0c\u8fd9\u79cd\u6280\u672f\u4f7f\u7528\u4e86\u591a\u79cd\u9aa8\u5e72\u7f51\u7edc\uff08\u5373 DINOv2\u3001CLIP-ConvNeXt\u3001CLIP-ViT\uff09\u8fdb\u884c\u56fe\u50cf\u7279\u5f81\u63d0\u53d6\u3002\u96c6\u6210\u5f15\u5165\u4e86\u989d\u5916\u7684\u8ba1\u7b97\u6210\u672c\uff0c\u4e25\u91cd\u5f71\u54cd\u4e86\u673a\u5668\u4eba MLLM \u5728\u73b0\u5b9e\u4e16\u754c\u4e2d\u7684\u5b9e\u7528\u6027\u3002\u56e0\u6b64\uff0c\u6211\u4eec\u8bc1\u660e\u4e86\uff0c\u5f53\u9ad8\u8d28\u91cf\u6570\u636e\u548c\u9002\u5f53\u7684\u8bad\u7ec3\u7b56\u7565\u7ed3\u5408\u65f6\uff0c\u7b80\u5355\u4e14\u76f4\u63a5\u7684\u6a21\u578b\u8bbe\u8ba1\u4e5f\u80fd\u5b9e\u73b0\u5f3a\u5927\u7684\u63a8\u7406\u80fd\u529b\u3002\u4e3a\u4e86\u4f7f LLM \u7406\u89e3\u89c6\u89c9\u7279\u5f81\uff0c\u6211\u4eec\u4f7f\u7528\u591a\u5c42\u611f\u77e5\u5668\uff08MLP\uff09\u5c06\u89c6\u89c9\u7f16\u7801\u5668\u8fde\u63a5\u5230 LLM\u3002\u901a\u8fc7\u8fd9\u4e2a\u7b80\u5355\u7684\u8de8\u6a21\u6001\u8fde\u63a5\u5668\uff0cRoboMamba \u53ef\u4ee5\u5c06\u89c6\u89c9\u4fe1\u606f\u8f6c\u6362\u4e3a\u8bed\u8a00\u5d4c\u5165\u7a7a\u95f4<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s8.51cto.com\/oss\/202406\/20\/855531b1807556c974b5676b20f91b6341b43a.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>\u3002<\/p>\n<p>\u8bf7\u6ce8\u610f\uff0c\u6a21\u578b\u6548\u7387\u5728\u673a\u5668\u4eba\u9886\u57df\u81f3\u5173\u91cd\u8981\uff0c\u56e0\u4e3a\u673a\u5668\u4eba\u9700\u8981\u6839\u636e\u4eba\u7c7b\u6307\u4ee4\u5feb\u901f\u54cd\u5e94\u3002\u56e0\u6b64\uff0c\u6211\u4eec\u9009\u62e9 Mamba \u4f5c\u4e3a\u6211\u4eec\u7684\u5927\u8bed\u8a00\u6a21\u578b\uff0c\u56e0\u4e3a\u5b83\u5177\u6709\u4e0a\u4e0b\u6587\u611f\u77e5\u63a8\u7406\u80fd\u529b\u548c\u7ebf\u6027\u8ba1\u7b97\u590d\u6742\u5ea6\u3002\u6587\u672c\u63d0\u793a\u4f7f\u7528\u9884\u8bad\u7ec3\u7684\u5206\u8bcd\u5668\u7f16\u7801\u4e3a\u5d4c\u5165\u7a7a\u95f4<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s2.51cto.com\/oss\/202406\/20\/e7b2dd100c8972b52d3506bba52ddac669909d.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>\uff0c\u7136\u540e\u4e0e\u89c6\u89c9 token \u8fde\u63a5\uff08cat\uff09\u5e76\u8f93\u5165 Mamba\u3002\u6211\u4eec\u5229\u7528 Mamba \u5f3a\u5927\u7684\u5e8f\u5217\u5efa\u6a21\u6765\u7406\u89e3\u591a\u6a21\u6001\u4fe1\u606f\uff0c\u5e76\u4f7f\u7528\u6709\u6548\u7684\u8bad\u7ec3\u7b56\u7565\u6765\u5f00\u53d1\u89c6\u89c9\u63a8\u7406\u80fd\u529b\uff08\u5982\u4e0b\u4e00\u8282\u6240\u8ff0\uff09\u3002\u8f93\u51fa token (<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s6.51cto.com\/oss\/202406\/20\/27e12e424bb5f093aa9885fdd3ac3095aa5ec8.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>) \u7136\u540e\u88ab\u89e3\u7801\uff08det\uff09\uff0c\u751f\u6210\u81ea\u7136\u8bed\u8a00\u54cd\u5e94<img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s8.51cto.com\/oss\/202406\/20\/d416fed461ad394f09293807e72c0913334ec4.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/>\u3002\u6a21\u578b\u7684\u524d\u5411\u8fc7\u7a0b\u53ef\u4ee5\u8868\u793a\u5982\u4e0b\uff1a<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s3.51cto.com\/oss\/202406\/20\/98a79c911c4abaf332d046050318bb969d2210.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p><strong>3.RoboMamba \u901a\u7528\u89c6\u89c9\u548c\u673a\u5668\u4eba\u63a8\u7406\u80fd\u529b\u8bad\u7ec3<\/strong><\/p>\n<p>\u5728\u6784\u5efa\u4e86 RoboMamba \u67b6\u6784\u540e\uff0c\u63a5\u4e0b\u6765\u7684\u76ee\u6807\u662f\u8bad\u7ec3\u6211\u4eec\u7684\u6a21\u578b\u5b66\u4e60\u901a\u7528\u89c6\u89c9\u63a8\u7406\u548c\u673a\u5668\u4eba\u76f8\u5173\u7684\u63a8\u7406\u80fd\u529b\u3002\u5982\u56fe 3 \u6240\u793a\uff0c\u6211\u4eec\u5c06 Stage 1 \u7684\u8bad\u7ec3\u5206\u4e3a\u4e24\u4e2a\u5b50\u6b65\u9aa4\uff1a\u5bf9\u9f50\u9884\u8bad\u7ec3\uff08Stage 1.1\uff09\u548c\u6307\u4ee4\u5171\u540c\u8bad\u7ec3\uff08Stage 1.2\uff09\u3002\u5177\u4f53\u800c\u8a00\uff0c\u4e0e\u4ee5\u5f80\u7684 MLLM \u8bad\u7ec3\u65b9\u6cd5\u4e0d\u540c\uff0c\u6211\u4eec\u7684\u76ee\u6807\u662f\u4f7f RoboMamba \u80fd\u591f\u7406\u89e3\u901a\u7528\u89c6\u89c9\u548c\u673a\u5668\u4eba\u573a\u666f\u3002\u9274\u4e8e\u673a\u5668\u4eba\u9886\u57df\u6d89\u53ca\u8bb8\u591a\u590d\u6742\u4e14\u65b0\u9896\u7684\u4efb\u52a1\uff0cRoboMamba \u9700\u8981\u66f4\u5f3a\u7684\u6cdb\u5316\u80fd\u529b\u3002\u56e0\u6b64\uff0c\u6211\u4eec\u5728 Stage 1.2 \u9636\u6bb5\u91c7\u7528\u4e86\u5171\u540c\u8bad\u7ec3\u7b56\u7565\uff0c\u5c06\u9ad8\u5c42\u6b21\u7684\u673a\u5668\u4eba\u6570\u636e\uff08\u4f8b\u5982\u4efb\u52a1\u89c4\u5212\uff09\u4e0e\u901a\u7528\u6307\u4ee4\u6570\u636e\u7ed3\u5408\u8d77\u6765\u3002\u6211\u4eec\u53d1\u73b0\uff0c\u5171\u540c\u8bad\u7ec3\u4e0d\u4ec5\u53ef\u4ee5\u83b7\u5f97\u66f4\u5177\u6cdb\u5316\u80fd\u529b\u7684\u673a\u5668\u4eba\u7b56\u7565\uff0c\u8fd8\u7531\u4e8e\u673a\u5668\u4eba\u6570\u636e\u4e2d\u7684\u590d\u6742\u63a8\u7406\u4efb\u52a1\u800c\u5e26\u6765\u7684\u901a\u7528\u573a\u666f\u63a8\u7406\u80fd\u529b\u589e\u5f3a\u3002\u8bad\u7ec3\u7ec6\u8282\u5982\u4e0b\uff1a<\/p>\n<ul data-id=\"ua73dd4b-Y73F39Za\">\n<li data-id=\"ld70c578-LOaEyDBD\">Stage 1.1\uff1a\u5bf9\u9f50\u9884\u8bad\u7ec3\u3002<\/li>\n<\/ul>\n<p>\u6211\u4eec\u91c7\u7528 LLaVA \u8fc7\u6ee4\u7684 558k \u56fe\u50cf &#8211; \u6587\u672c\u914d\u5bf9\u6570\u636e\u96c6\u8fdb\u884c\u8de8\u6a21\u6001\u5bf9\u9f50\u3002\u5982\u56fe 3 \u6240\u793a\uff0c\u6211\u4eec\u51bb\u7ed3 CLIP \u7f16\u7801\u5668\u548c Mamba \u8bed\u8a00\u6a21\u578b\u7684\u53c2\u6570\uff0c\u4ec5\u66f4\u65b0\u6295\u5f71\u5c42\u3002\u901a\u8fc7\u8fd9\u79cd\u65b9\u5f0f\uff0c\u6211\u4eec\u53ef\u4ee5\u5c06\u56fe\u50cf\u7279\u5f81\u4e0e\u9884\u8bad\u7ec3\u7684 Mamba \u8bcd\u5d4c\u5165\u5bf9\u9f50\u3002<\/p>\n<ul data-id=\"ua73dd4b-X5kCX7j9\">\n<li data-id=\"ld70c578-vxK9HYua\">Stage 1.2\uff1a\u6307\u4ee4\u5171\u540c\u8bad\u7ec3\u3002<\/li>\n<\/ul>\n<p>\u5728\u8fd9\u4e00\u9636\u6bb5\uff0c\u6211\u4eec\u9996\u5148\u9075\u5faa\u5148\u524d MLLM \u7684\u5de5\u4f5c\u8fdb\u884c\u901a\u7528\u89c6\u89c9\u6307\u4ee4\u6570\u636e\u6536\u96c6\u3002\u6211\u4eec\u91c7\u7528\u4e86 655K LLaVA \u6df7\u5408\u6307\u4ee4\u6570\u636e\u96c6\u548c 400K LRV-Instruct \u6570\u636e\u96c6\uff0c\u5206\u522b\u7528\u4e8e\u5b66\u4e60\u89c6\u89c9\u6307\u4ee4\u8ddf\u968f\u548c\u51cf\u8f7b\u5e7b\u89c9\u3002\u9700\u8981\u6ce8\u610f\u7684\u662f\uff0c\u51cf\u8f7b\u5e7b\u89c9\u5728\u673a\u5668\u4eba\u573a\u666f\u4e2d\u8d77\u7740\u91cd\u8981\u4f5c\u7528\uff0c\u56e0\u4e3a\u673a\u5668\u4eba MLLM \u9700\u8981\u57fa\u4e8e\u771f\u5b9e\u573a\u666f\u751f\u6210\u4efb\u52a1\u89c4\u5212\uff0c\u800c\u4e0d\u662f\u60f3\u8c61\u4e2d\u7684\u573a\u666f\u3002\u4f8b\u5982\uff0c\u73b0\u6709\u7684 MLLMs \u53ef\u80fd\u516c\u5f0f\u5316\u5730\u56de\u7b54 \u201c\u6253\u5f00\u5fae\u6ce2\u7089\u201d \u65f6\u8bf4 \u201c\u6b65\u9aa4 1\uff1a\u627e\u5230\u628a\u624b\u201d\uff0c\u4f46\u8bb8\u591a\u5fae\u6ce2\u7089\u6ca1\u6709\u628a\u624b\u3002\u63a5\u4e0b\u6765\uff0c\u6211\u4eec\u7ed3\u5408\u4e86 800K RoboVQA \u6570\u636e\u96c6\uff0c\u4ee5\u5b66\u4e60\u9ad8\u5c42\u6b21\u7684\u673a\u5668\u4eba\u6280\u80fd\uff0c\u5982\u957f\u7a0b\u4efb\u52a1\u89c4\u5212\u3001\u53ef\u64cd\u7eb5\u6027\u5224\u65ad\u3001\u53ef\u64cd\u7eb5\u6027\u751f\u6210\u3001\u672a\u6765\u4e0e\u8fc7\u53bb\u9884\u6d4b\u7b49\u3002\u5728\u5171\u540c\u8bad\u7ec3\u671f\u95f4\uff0c\u5982\u56fe 3 \u6240\u793a\uff0c\u6211\u4eec\u51bb\u7ed3 CLIP \u7f16\u7801\u5668\u7684\u53c2\u6570\uff0c\u5e76\u5728 1.8m \u5408\u5e76\u6570\u636e\u96c6\u4e0a\u5fae\u8c03\u6295\u5f71\u5c42\u548c Mamba\u3002\u6240\u6709\u6765\u81ea Mamba \u8bed\u8a00\u6a21\u578b\u7684\u8f93\u51fa\u90fd\u4f7f\u7528\u4ea4\u53c9\u71b5\u635f\u5931\u8fdb\u884c\u76d1\u7763\u3002<\/p>\n<p><strong>4.RoboMamba \u64cd\u7eb5\u80fd\u529b\u5fae\u8c03\u8bad\u7ec3<\/strong><\/p>\n<p>\u5728 RoboMamba \u5f3a\u5927\u7684\u63a8\u7406\u80fd\u529b\u57fa\u7840\u4e0a\uff0c\u6211\u4eec\u5728\u672c\u8282\u4ecb\u7ecd\u4e86\u6211\u4eec\u7684\u673a\u5668\u4eba\u64cd\u4f5c\u5fae\u8c03\u7b56\u7565\uff0c\u5728\u56fe 3 \u4e2d\u79f0\u4e3a\u8bad\u7ec3 Stage 2\u3002\u73b0\u6709\u7684\u57fa\u4e8e MLLM \u7684\u673a\u5668\u4eba\u64cd\u4f5c\u65b9\u6cd5\u5728\u64cd\u4f5c\u5fae\u8c03\u9636\u6bb5\u9700\u8981\u66f4\u65b0\u6295\u5f71\u5c42\u548c\u6574\u4e2a LLM\u3002\u867d\u7136\u8fd9\u79cd\u8303\u5f0f\u53ef\u4ee5\u8d4b\u4e88\u6a21\u578b\u52a8\u4f5c\u4f4d\u59ff\u9884\u6d4b\u80fd\u529b\uff0c\u4f46\u5b83\u4e5f\u7834\u574f\u4e86 MLLM \u7684\u56fa\u6709\u80fd\u529b\uff0c\u5e76\u4e14\u9700\u8981\u5927\u91cf\u7684\u8bad\u7ec3\u8d44\u6e90\u3002\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e9b\u6311\u6218\uff0c\u6211\u4eec\u63d0\u51fa\u4e86\u4e00\u79cd\u9ad8\u6548\u7684\u5fae\u8c03\u7b56\u7565\uff0c\u5982\u56fe 3 \u6240\u793a\u3002\u6211\u4eec\u51bb\u7ed3 RoboMamba \u7684\u6240\u6709\u53c2\u6570\uff0c\u5e76\u5f15\u5165\u4e00\u4e2a\u7b80\u5355\u7684 Policy head \u6765\u5efa\u6a21 Mamba \u7684\u8f93\u51fa token\u3002Policy head \u5305\u542b\u4e24\u4e2a MLP \u5206\u522b\u5b66\u4e60\u672b\u7aef\u6267\u884c\u5668\u4f4d\u7f6e\u548c\u65b9\u5411\uff0c\u603b\u5171\u5360\u7528\u6574\u4e2a\u6a21\u578b\u53c2\u6570\u7684 0.1%\u3002\u6839\u636e\u524d\u671f\u5de5\u4f5c where2act\uff0c\u4f4d\u7f6e\u548c\u65b9\u5411\u7684\u635f\u5931\u516c\u5f0f\u5982\u4e0b\uff1a<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s8.51cto.com\/oss\/202406\/20\/450882401a47451cf1f7326cac32864e56dceb.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u5176\u4e2d\uff0cN \u8868\u793a\u8bad\u7ec3\u6837\u672c\u7684\u6570\u91cf\uff0cTr (A) \u8868\u793a\u77e9\u9635 A \u7684\u8ff9\u3002RoboMamba \u53ea\u9884\u6d4b\u56fe\u50cf\u4e2d\u63a5\u89e6\u50cf\u7d20\u7684\u4e8c\u7ef4\u4f4d\u7f6e\uff08x, y\uff09\uff0c\u7136\u540e\u4f7f\u7528\u6df1\u5ea6\u4fe1\u606f\u5c06\u5176\u8f6c\u6362\u4e3a\u4e09\u7ef4\u7a7a\u95f4\u3002\u4e3a\u4e86\u8bc4\u4f30\u8fd9\u4e00\u5fae\u8c03\u7b56\u7565\uff0c\u6211\u4eec\u4f7f\u7528 SAPIEN \u6a21\u62df\u751f\u6210\u4e86\u4e00\u4e2a\u5305\u542b 1 \u4e07\u6761\u672b\u7aef\u6267\u884c\u5668\u4f4d\u59ff\u9884\u6d4b\u7684\u6570\u636e\u96c6\u3002<\/p>\n<p>\u5728\u64cd\u4f5c\u5fae\u8c03\u4e4b\u540e\uff0c\u6211\u4eec\u53d1\u73b0\u4e00\u65e6 RoboMamba \u5177\u5907\u4e86\u8db3\u591f\u7684\u63a8\u7406\u80fd\u529b\uff0c\u5b83\u53ef\u4ee5\u901a\u8fc7\u6781\u9ad8\u6548\u7684\u5fae\u8c03\u6765\u83b7\u53d6\u4f4d\u59ff\u9884\u6d4b\u6280\u80fd\u3002\u7531\u4e8e\u5fae\u8c03\u53c2\u6570\uff087MB\uff09\u6781\u5c11\u4e14\u6a21\u578b\u8bbe\u8ba1\u9ad8\u6548\uff0c\u6211\u4eec\u53ea\u9700 20 \u5206\u949f\u5373\u53ef\u5b9e\u73b0\u65b0\u7684\u64cd\u4f5c\u6280\u80fd\u5b66\u4e60\u3002\u8fd9\u4e00\u53d1\u73b0\u7a81\u51fa\u4e86\u63a8\u7406\u80fd\u529b\u5bf9\u4e8e\u5b66\u4e60\u64cd\u4f5c\u6280\u80fd\u7684\u91cd\u8981\u6027\uff0c\u5e76\u63d0\u51fa\u4e86\u4e00\u4e2a\u65b0\u7684\u89c6\u89d2\uff1a\u6211\u4eec\u53ef\u4ee5\u5728\u4e0d\u5f71\u54cd MLLM \u56fa\u6709\u63a8\u7406\u80fd\u529b\u7684\u60c5\u51b5\u4e0b\uff0c\u9ad8\u6548\u5730\u8d4b\u4e88\u5176\u64cd\u4f5c\u80fd\u529b\u3002\u6700\u540e\uff0cRoboMamba \u53ef\u4ee5\u4f7f\u7528\u8bed\u8a00\u54cd\u5e94\u8fdb\u884c\u5e38\u8bc6\u548c\u4e0e\u673a\u5668\u4eba\u76f8\u5173\u7684\u63a8\u7406\uff0c\u5e76\u4f7f\u7528 Policy head \u8fdb\u884c\u52a8\u4f5c\u4f4d\u59ff\u9884\u6d4b\u3002<\/p>\n<h4>\u5b9a\u91cf\u5b9e\u9a8c<\/h4>\n<p><strong>1. \u901a\u7528\u63a8\u7406\u80fd\u529b\u8bc4\u4f30\uff08MLLM Benchmarks\uff09<\/strong><\/p>\n<p>\u4e3a\u4e86\u8bc4\u4f30\u63a8\u7406\u80fd\u529b\uff0c\u6211\u4eec\u4f7f\u7528\u4e86\u51e0\u4e2a\u6d41\u884c\u7684\u57fa\u51c6\uff0c\u5305\u62ec VQAv2\u3001OKVQA\u3001GQA\u3001OCRVQA\u3001VizWiz\u3001POPE\u3001MME\u3001MMBench \u548c MM-Vet\u3002\u9664\u6b64\u4e4b\u5916\uff0c\u6211\u4eec\u8fd8\u5728 RoboVQA \u7684 18k \u9a8c\u8bc1\u6570\u636e\u96c6\u4e0a\u76f4\u63a5\u8bc4\u4f30\u4e86 RoboMamba \u7684\u673a\u5668\u4eba\u76f8\u5173\u63a8\u7406\u80fd\u529b\uff0c\u6db5\u76d6\u4e86\u673a\u5668\u4eba\u4efb\u52a1\uff0c\u5982\u4efb\u52a1\u89c4\u5212\u3001\u63d0\u793a\u6027\u4efb\u52a1\u89c4\u5212\u3001\u957f\u7a0b\u4efb\u52a1\u89c4\u5212\u3001\u53ef\u64cd\u7eb5\u6027\u5224\u65ad\u3001\u53ef\u64cd\u7eb5\u6027\u751f\u6210\u3001\u8fc7\u53bb\u63cf\u8ff0\u548c\u672a\u6765\u9884\u6d4b\u7b49\u3002<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s6.51cto.com\/oss\/202406\/20\/37c932a08c89a6008e9581a3480eaff25d0210.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u8868 1. Robomamba \u4e0e\u73b0\u6709 MLLMs \u5728\u591a\u4e2a\u57fa\u51c6\u4e0a\u7684\u901a\u7528\u63a8\u7406\u80fd\u529b\u6bd4\u8f83\u3002<\/p>\n<p>\u5982\u8868 1 \u6240\u793a\uff0c\u6211\u4eec\u5c06 RoboMamba \u4e0e\u4ee5\u524d\u6700\u5148\u8fdb\u7684 (SOTA) MLLM \u5728\u901a\u7528\u7684 VQA \u548c\u6700\u8fd1\u7684 MLLM \u57fa\u51c6\u6d4b\u8bd5\u4e0a\u8fdb\u884c\u6bd4\u8f83\u3002\u9996\u5148\uff0c\u6211\u4eec\u53d1\u73b0 RoboMamba \u4ec5\u4f7f\u7528 2.7B \u8bed\u8a00\u6a21\u578b\uff0c\u5c31\u5728\u6240\u6709 VQA \u57fa\u51c6\u6d4b\u8bd5\u4e2d\u53d6\u5f97\u4e86\u4ee4\u4eba\u6ee1\u610f\u7684\u7ed3\u679c\u3002\u7ed3\u679c\u8868\u660e\uff0c\u7b80\u5355\u7684\u7ed3\u6784\u8bbe\u8ba1\u662f\u6709\u6548\u7684\u3002\u5bf9\u9f50\u9884\u8bad\u7ec3\u548c\u6307\u4ee4\u534f\u540c\u8bad\u7ec3\u663e\u8457\u63d0\u9ad8\u4e86 MLLM \u7684\u63a8\u7406\u80fd\u529b\u3002\u4f8b\u5982\uff0c\u7531\u4e8e\u5728\u534f\u540c\u8bad\u7ec3\u9636\u6bb5\u5f15\u5165\u4e86\u5927\u91cf\u7684\u673a\u5668\u4eba\u6570\u636e\uff0cRoboMamba \u5728 GQA \u57fa\u51c6\u4e0a\u7684\u7a7a\u95f4\u8bc6\u522b\u6027\u80fd\u5f97\u5230\u4e86\u63d0\u9ad8\u3002\u540c\u65f6\uff0c\u6211\u4eec\u8fd8\u5728\u6700\u8fd1\u63d0\u51fa\u7684 MLLM \u57fa\u51c6\u4e0a\u6d4b\u8bd5\u4e86\u6211\u4eec\u7684 RoboMamba\u3002\u4e0e\u4ee5\u524d\u7684 MLLMs \u76f8\u6bd4\uff0c\u6211\u4eec\u89c2\u5bdf\u5230\u6211\u4eec\u7684\u6a21\u578b\u5728\u6240\u6709\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u90fd\u53d6\u5f97\u4e86\u5177\u6709\u7ade\u4e89\u529b\u7684\u7ed3\u679c\u3002\u867d\u7136 RoboMamba \u7684\u4e00\u4e9b\u6027\u80fd\u4ecd\u7136\u4f4e\u4e8e\u6700\u5148\u8fdb\u7684 7B MLLM \uff08e.g., LLaVA1.5 \u548c SPHINX\uff09\uff0c\u4f46\u6211\u4eec\u4f18\u5148\u4f7f\u7528\u66f4\u5c0f\u66f4\u5feb\u7684 Mamba-2.7B \u6765\u5e73\u8861\u673a\u5668\u4eba\u6a21\u578b\u7684\u6548\u7387\u3002\u5728\u672a\u6765\uff0c\u6211\u4eec\u8ba1\u5212\u4e3a\u8d44\u6e90\u4e0d\u53d7\u9650\u5236\u7684\u573a\u666f\u5f00\u53d1 RoboMamba-7B\u3002<\/p>\n<p><strong>2. \u673a\u5668\u4eba\u63a8\u7406\u80fd\u529b\u8bc4\u4f30\uff08RoboVQA Benchmark\uff09<\/strong><\/p>\n<p>\u53e6\u5916\uff0c\u4e3a\u4e86\u5168\u9762\u6bd4\u8f83 RoboMamba \u4e0e\u673a\u5668\u4eba\u76f8\u5173\u7684\u63a8\u7406\u80fd\u529b\uff0c\u6211\u4eec\u5728 RoboVQA \u9a8c\u8bc1\u96c6\u4e0a\u4e0e LLaMA-AdapterV2 \u8fdb\u884c\u57fa\u51c6\u6d4b\u8bd5\u3002\u6211\u4eec\u9009\u62e9 LLaMA-AdapterV2 \u4f5c\u4e3a\u57fa\u51c6\uff0c\u56e0\u4e3a\u5b83\u662f\u5f53\u524d SOTA \u673a\u5668\u4eba MLLM (ManipLLM) \u7684\u57fa\u7840\u6a21\u578b\u3002\u4e3a\u4e86\u8fdb\u884c\u516c\u5e73\u7684\u6bd4\u8f83\uff0c\u6211\u4eec\u52a0\u8f7d\u4e86 LLaMA-AdapterV2 \u9884\u8bad\u7ec3\u53c2\u6570\uff0c\u5e76\u4f7f\u7528\u5176\u5b98\u65b9\u6307\u4ee4\u5fae\u8c03\u65b9\u6cd5\u5728 RoboVQA \u8bad\u7ec3\u96c6\u4e0a\u5bf9\u5176\u8fdb\u884c\u4e86\u4e24\u4e2a epoch \u7684\u5fae\u8c03\u3002\u5982\u56fe 4 a\uff09\u6240\u793a\uff0cRoboMamba \u5728 BLEU-1 \u5230 BLEU-4 \u4e4b\u95f4\u5b9e\u73b0\u4e86\u5353\u8d8a\u7684\u6027\u80fd\u3002\u7ed3\u679c\u8868\u660e\uff0c\u6211\u4eec\u7684\u6a21\u578b\u5177\u6709\u5148\u8fdb\u7684\u673a\u5668\u4eba\u76f8\u5173\u63a8\u7406\u80fd\u529b\uff0c\u5e76\u8bc1\u5b9e\u4e86\u6211\u4eec\u7684\u8bad\u7ec3\u7b56\u7565\u7684\u6709\u6548\u6027\u3002\u9664\u4e86\u66f4\u9ad8\u7684\u51c6\u786e\u7387\u5916\uff0c\u6211\u4eec\u7684\u6a21\u578b\u5b9e\u73b0\u7684\u63a8\u7406\u901f\u5ea6\u6bd4 LLaMA-AdapterV2 \u548c ManipLLM \u5feb 7 \u500d\uff0c\u8fd9\u53ef\u4ee5\u5f52\u56e0\u4e8e Mamba \u8bed\u8a00\u6a21\u578b\u7684\u5185\u5bb9\u611f\u77e5\u63a8\u7406\u80fd\u529b\u548c\u6548\u7387\u3002<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s2.51cto.com\/oss\/202406\/20\/33b395d10af8c735a22864b01e454d78dbd7f8.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u56fe 4. \u00a0RoboVQA \u4e0a\u4e0e\u673a\u5668\u4eba\u76f8\u5173\u7684\u63a8\u7406\u5bf9\u6bd4\u3002<\/p>\n<p><strong>3. \u673a\u5668\u4eba\u64cd\u7eb5\u80fd\u529b\u8bc4\u4f30\uff08SAPIEN\uff09<\/strong><\/p>\n<p>\u4e3a\u4e86\u8bc4\u4f30 RoboMamba \u7684\u64cd\u4f5c\u80fd\u529b\uff0c\u6211\u4eec\u5c06\u6211\u4eec\u7684\u6a21\u578b\u4e0e\u56db\u4e2a\u57fa\u7ebf\u8fdb\u884c\u4e86\u6bd4\u8f83\uff1aUMPNet\uff0c Flowbot3D\uff0c RoboFlamingo \u548c ManipLLM\u3002\u5728\u6bd4\u8f83\u4e4b\u524d\uff0c\u6211\u4eec\u590d\u73b0\u6240\u6709\u57fa\u7ebf\u5e76\u5728\u6211\u4eec\u6536\u96c6\u7684\u6570\u636e\u96c6\u4e0a\u8bad\u7ec3\u5b83\u4eec\u3002\u5bf9\u4e8e UMPNet\uff0c\u6211\u4eec\u5728\u9884\u6d4b\u7684\u63a5\u89e6\u70b9\u4e0a\u6267\u884c\u64cd\u4f5c\uff0c\u65b9\u5411\u5782\u76f4\u4e8e\u7269\u4f53\u8868\u9762\u3002Flowbot3D \u5728\u70b9\u4e91\u4e0a\u9884\u6d4b\u8fd0\u52a8\u65b9\u5411\uff0c\u9009\u62e9\u6700\u5927\u7684\u6d41\u4f5c\u4e3a\u4ea4\u4e92\u70b9\uff0c\u5e76\u4f7f\u7528\u6d41\u65b9\u5411\u8868\u793a\u672b\u7aef\u6267\u884c\u5668\u7684\u65b9\u5411\u3002RoboFlamingo \u548c ManipLLM \u5206\u522b\u52a0\u8f7d OpenFlamingo \u548c LLaMA-AdapterV2 \u9884\u8bad\u7ec3\u53c2\u6570\uff0c\u5e76\u9075\u5faa\u5404\u81ea\u7684\u5fae\u8c03\u548c\u6a21\u578b\u66f4\u65b0\u7b56\u7565\u3002\u5982\u8868 2 \u6240\u793a\uff0c\u4e0e\u4e4b\u524d\u7684 SOTA ManipLLM \u76f8\u6bd4\uff0c\u6211\u4eec\u7684 RoboMamba \u5728\u53ef\u89c1\u7c7b\u522b\u4e0a\u5b9e\u73b0\u4e86 7.0% \u7684\u6539\u8fdb\uff0c\u5728\u4e0d\u53ef\u89c1\u7c7b\u522b\u4e0a\u5b9e\u73b0\u4e86 2.0% \u7684\u6539\u8fdb\u3002\u5728\u6548\u7387\u65b9\u9762\uff0cRoboFlamingo \u66f4\u65b0\u4e86 35.5% (1.8B) \u7684\u6a21\u578b\u53c2\u6570\uff0cManipLLM \u66f4\u65b0\u4e86 LLM \u4e2d\u7684\u9002\u914d\u5668 (41.3M)\uff0c\u5305\u542b 0.5% \u7684\u6a21\u578b\u53c2\u6570\uff0c\u800c\u6211\u4eec\u7684\u5fae\u8c03 Policy head (3.7M) \u4ec5\u5360\u6a21\u578b\u53c2\u6570\u7684 0.1%\u3002RoboMamba \u6bd4\u4ee5\u524d\u57fa\u4e8e MLLM \u7684\u65b9\u6cd5\u66f4\u65b0\u7684\u53c2\u6570\u5c11\u4e86 10 \u500d\uff0c\u800c\u63a8\u7406\u901f\u5ea6\u63d0\u9ad8\u4e86 7 \u500d\u3002\u7ed3\u679c\u8868\u660e\uff0c\u6211\u4eec\u7684 RoboMamba \u4e0d\u4ec5\u5177\u6709\u5f3a\u5927\u7684\u63a8\u7406\u80fd\u529b\uff0c\u800c\u4e14\u80fd\u591f\u4ee5\u4f4e\u6210\u672c\u7684\u65b9\u5f0f\u83b7\u5f97\u64cd\u7eb5\u80fd\u529b\u3002<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s9.51cto.com\/oss\/202406\/20\/5914d0d67f4b39dee8f8687f348564773e2da3.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u8868 2. Robomamba \u4e0e\u5176\u4ed6 baseline \u7684\u6210\u529f\u7387\u6bd4\u8f83<\/p>\n<h4>\u5b9a\u6027\u7ed3\u679c<\/h4>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s5.51cto.com\/oss\/202406\/20\/131ad9c068035a6d51a2946c68c45a980a3eca.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u56fe 4. RoboMamba \u9762\u5bf9\u73b0\u5b9e\u4e16\u754c\u4e2d\u5404\u79cd\u673a\u5668\u4eba\u4e0b\u6e38\u4efb\u52a1\u7684\u53ef\u89c6\u5316\u3002<\/p>\n<p>\u5982\u56fe 4 \u6240\u793a\uff0c\u6211\u4eec\u53ef\u89c6\u5316\u4e86 RoboMamba \u5728\u5404\u79cd\u673a\u5668\u4eba\u4e0b\u6e38\u4efb\u52a1\u4e2d\u7684\u63a8\u7406\u7ed3\u679c\u3002\u5728\u4efb\u52a1\u89c4\u5212\u65b9\u9762\uff0c\u76f8\u8f83\u4e8e LLaMA-AdapterV2, RoboMamba \u51ed\u501f\u5176\u5f3a\u5927\u7684\u63a8\u7406\u80fd\u529b\uff0c\u5c55\u73b0\u51fa\u4e86\u66f4\u51c6\u786e\u3001\u66f4\u957f\u8fdc\u7684\u89c4\u5212\u80fd\u529b\u3002\u4e3a\u4e86\u8fdb\u884c\u516c\u5e73\u7684\u6bd4\u8f83\uff0c\u6211\u4eec\u8fd8\u5bf9 RoboVQA \u6570\u636e\u96c6\u4e0a\u7684\u57fa\u51c6 LLaMA-AdapterV2 \u8fdb\u884c\u4e86\u5fae\u8c03\u3002\u5bf9\u4e8e\u64cd\u7eb5\u4f4d\u59ff\u9884\u6d4b\uff0c\u6211\u4eec\u4f7f\u7528\u4e86 Franka Emika \u673a\u68b0\u81c2\u6765\u4e0e\u5404\u79cd\u5bb6\u5ead\u7269\u54c1\u8fdb\u884c\u4ea4\u4e92\u3002\u6211\u4eec\u5c06 RoboMamba \u9884\u6d4b\u7684 3D \u4f4d\u59ff\u6295\u5f71\u5230 2D \u56fe\u50cf\u4e0a\uff0c\u4f7f\u7528\u7ea2\u70b9\u8868\u793a\u63a5\u89e6\u70b9\uff0c\u672b\u7aef\u6267\u884c\u5668\u8868\u793a\u65b9\u5411\uff0c\u5982\u56fe\u53f3\u4e0b\u89d2\u6240\u793a\u3002<\/p>\n<p><img data-dominant-color=\"a49e92\" data-has-transparency=\"false\" style=\"--dominant-color: #a49e92;\" loading=\"lazy\" decoding=\"async\" class=\"not-transparent alignnone size-full wp-image-20699\" src=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2024\/06\/320292e19d5096b3efe1549781aca77881a877.gif\" width=\"400\" height=\"427\" alt=\"\u56fe\u7247\" title=\"\u56fe\u7247\" \/><\/p>\n<p><img data-dominant-color=\"b8b09a\" data-has-transparency=\"false\" style=\"--dominant-color: #b8b09a;\" loading=\"lazy\" decoding=\"async\" class=\"not-transparent alignnone size-full wp-image-20700\" src=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2024\/06\/e95ca57673591cecf927209533ec154c34de0f.gif\" width=\"400\" height=\"426\" alt=\"\u56fe\u7247\" title=\"\u56fe\u7247\" \/><\/p>\n<p><img data-dominant-color=\"bab095\" data-has-transparency=\"false\" style=\"--dominant-color: #bab095;\" loading=\"lazy\" decoding=\"async\" class=\"not-transparent alignnone size-full wp-image-20701\" src=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2024\/06\/242e2772199ec90e91d146f3c4146469055feb.gif\" width=\"400\" height=\"422\" alt=\"\u56fe\u7247\" title=\"\u56fe\u7247\" \/><\/p>\n<\/div>\n<\/div>\n<div class=\"pvc_clear\"><\/div>\n<p id=\"pvc_stats_20697\" class=\"pvc_stats total_only  \" data-element-id=\"20697\" style=\"\"><i class=\"pvc-stats-icon medium\" aria-hidden=\"true\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" version=\"1.0\" viewBox=\"0 0 502 315\" preserveAspectRatio=\"xMidYMid meet\"><g transform=\"translate(0,332) scale(0.1,-0.1)\" fill=\"\" stroke=\"none\"><path d=\"M2394 3279 l-29 -30 -3 -207 c-2 -182 0 -211 15 -242 39 -76 157 -76 196 0 15 31 17 60 15 243 l-3 209 -33 29 c-26 23 -41 29 -80 29 -41 0 -53 -5 -78 -31z\"\/><path d=\"M3085 3251 c-45 -19 -58 -50 -96 -229 -47 -217 -49 -260 -13 -295 52 -53 146 -42 177 20 16 31 87 366 87 410 0 70 -86 122 -155 94z\"\/><path d=\"M1751 3234 c-13 -9 -29 -31 -37 -50 -12 -29 -10 -49 21 -204 19 -94 39 -189 45 -210 14 -50 54 -80 110 -80 34 0 48 6 76 34 21 21 34 44 34 59 0 14 -18 113 -40 219 -37 178 -43 195 -70 221 -36 32 -101 37 -139 11z\"\/><path d=\"M1163 3073 c-36 -7 -73 -59 -73 -102 0 -56 133 -378 171 -413 34 -32 83 -37 129 -13 70 36 67 87 -16 290 -86 209 -89 214 -129 231 -35 14 -42 15 -82 7z\"\/><path d=\"M3689 3066 c-15 -9 -33 -30 -42 -48 -48 -103 -147 -355 -147 -375 0 -98 131 -148 192 -74 13 15 57 108 97 206 80 196 84 226 37 273 -30 30 -99 39 -137 18z\"\/><path d=\"M583 2784 c-38 -19 -67 -74 -58 -113 9 -42 211 -354 242 -373 16 -10 45 -18 66 -18 51 0 107 52 107 100 0 39 -1 41 -124 234 -80 126 -108 162 -133 173 -41 17 -61 16 -100 -3z\"\/><path d=\"M4250 2784 c-14 -9 -74 -91 -133 -183 -95 -150 -107 -173 -107 -213 0 -55 33 -94 87 -104 67 -13 90 8 211 198 130 202 137 225 78 284 -27 27 -42 34 -72 34 -22 0 -50 -8 -64 -16z\"\/><path d=\"M2275 2693 c-553 -48 -1095 -270 -1585 -649 -135 -104 -459 -423 -483 -476 -23 -49 -22 -139 2 -186 73 -142 361 -457 571 -626 285 -228 642 -407 990 -497 242 -63 336 -73 660 -74 310 0 370 5 595 52 535 111 1045 392 1455 803 122 121 250 273 275 326 19 41 19 137 0 174 -41 79 -309 363 -465 492 -447 370 -946 591 -1479 653 -113 14 -422 18 -536 8z m395 -428 c171 -34 330 -124 456 -258 112 -119 167 -219 211 -378 27 -96 24 -300 -5 -401 -72 -255 -236 -447 -474 -557 -132 -62 -201 -76 -368 -76 -167 0 -236 14 -368 76 -213 98 -373 271 -451 485 -162 444 86 934 547 1084 153 49 292 57 452 25z m909 -232 c222 -123 408 -262 593 -441 76 -74 138 -139 138 -144 0 -16 -233 -242 -330 -319 -155 -123 -309 -223 -461 -299 l-81 -41 32 46 c18 26 49 83 70 128 143 306 141 649 -6 957 -25 52 -61 116 -79 142 l-34 47 45 -20 c26 -10 76 -36 113 -56z m-2057 25 c-40 -58 -105 -190 -130 -263 -110 -324 -59 -707 132 -981 25 -35 42 -64 37 -64 -19 0 -241 119 -326 174 -188 122 -406 314 -532 468 l-58 71 108 103 c185 178 428 349 672 473 66 33 121 60 123 61 2 0 -10 -19 -26 -42z\"\/><path d=\"M2375 1950 c-198 -44 -350 -190 -395 -379 -18 -76 -8 -221 19 -290 114 -284 457 -406 731 -260 98 52 188 154 231 260 27 69 37 214 19 290 -38 163 -166 304 -326 360 -67 23 -215 33 -279 19z\"\/><\/g><\/svg><\/i> <img loading=\"lazy\" decoding=\"async\" width=\"16\" height=\"16\" alt=\"Loading\" src=\"https:\/\/aif.amtbbs.org\/wp-content\/plugins\/page-views-count\/ajax-loader-2x.gif\" border=0 \/><\/p>\n<div class=\"pvc_clear\"><\/div>\n","protected":false},"excerpt":{"rendered":"<p>\u4e3a\u4e86\u8d4b\u4e88\u673a\u5668\u4eba\u7aef\u5230\u7aef\u7684\u63a8\u7406\u548c\u64cd\u7eb5\u80fd\u529b\uff0c\u672c\u6587\u521b\u65b0\u6027\u5730\u5c06\u89c6\u89c9\u7f16\u7801\u5668\u4e0e\u9ad8\u6548\u7684\u72b6\u6001\u7a7a\u95f4\u8bed\u8a00\u6a21\u578b\u96c6\u6210\uff0c\u6784\u5efa\u4e86\u5168\u65b0\u7684 Ro [&hellip;]<\/p>\n<div class=\"pvc_clear\"><\/div>\n<p id=\"pvc_stats_20697\" class=\"pvc_stats total_only  \" data-element-id=\"20697\" style=\"\"><i class=\"pvc-stats-icon medium\" aria-hidden=\"true\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" version=\"1.0\" viewBox=\"0 0 502 315\" preserveAspectRatio=\"xMidYMid meet\"><g transform=\"translate(0,332) scale(0.1,-0.1)\" fill=\"\" stroke=\"none\"><path d=\"M2394 3279 l-29 -30 -3 -207 c-2 -182 0 -211 15 -242 39 -76 157 -76 196 0 15 31 17 60 15 243 l-3 209 -33 29 c-26 23 -41 29 -80 29 -41 0 -53 -5 -78 -31z\"\/><path d=\"M3085 3251 c-45 -19 -58 -50 -96 -229 -47 -217 -49 -260 -13 -295 52 -53 146 -42 177 20 16 31 87 366 87 410 0 70 -86 122 -155 94z\"\/><path d=\"M1751 3234 c-13 -9 -29 -31 -37 -50 -12 -29 -10 -49 21 -204 19 -94 39 -189 45 -210 14 -50 54 -80 110 -80 34 0 48 6 76 34 21 21 34 44 34 59 0 14 -18 113 -40 219 -37 178 -43 195 -70 221 -36 32 -101 37 -139 11z\"\/><path d=\"M1163 3073 c-36 -7 -73 -59 -73 -102 0 -56 133 -378 171 -413 34 -32 83 -37 129 -13 70 36 67 87 -16 290 -86 209 -89 214 -129 231 -35 14 -42 15 -82 7z\"\/><path d=\"M3689 3066 c-15 -9 -33 -30 -42 -48 -48 -103 -147 -355 -147 -375 0 -98 131 -148 192 -74 13 15 57 108 97 206 80 196 84 226 37 273 -30 30 -99 39 -137 18z\"\/><path d=\"M583 2784 c-38 -19 -67 -74 -58 -113 9 -42 211 -354 242 -373 16 -10 45 -18 66 -18 51 0 107 52 107 100 0 39 -1 41 -124 234 -80 126 -108 162 -133 173 -41 17 -61 16 -100 -3z\"\/><path d=\"M4250 2784 c-14 -9 -74 -91 -133 -183 -95 -150 -107 -173 -107 -213 0 -55 33 -94 87 -104 67 -13 90 8 211 198 130 202 137 225 78 284 -27 27 -42 34 -72 34 -22 0 -50 -8 -64 -16z\"\/><path d=\"M2275 2693 c-553 -48 -1095 -270 -1585 -649 -135 -104 -459 -423 -483 -476 -23 -49 -22 -139 2 -186 73 -142 361 -457 571 -626 285 -228 642 -407 990 -497 242 -63 336 -73 660 -74 310 0 370 5 595 52 535 111 1045 392 1455 803 122 121 250 273 275 326 19 41 19 137 0 174 -41 79 -309 363 -465 492 -447 370 -946 591 -1479 653 -113 14 -422 18 -536 8z m395 -428 c171 -34 330 -124 456 -258 112 -119 167 -219 211 -378 27 -96 24 -300 -5 -401 -72 -255 -236 -447 -474 -557 -132 -62 -201 -76 -368 -76 -167 0 -236 14 -368 76 -213 98 -373 271 -451 485 -162 444 86 934 547 1084 153 49 292 57 452 25z m909 -232 c222 -123 408 -262 593 -441 76 -74 138 -139 138 -144 0 -16 -233 -242 -330 -319 -155 -123 -309 -223 -461 -299 l-81 -41 32 46 c18 26 49 83 70 128 143 306 141 649 -6 957 -25 52 -61 116 -79 142 l-34 47 45 -20 c26 -10 76 -36 113 -56z m-2057 25 c-40 -58 -105 -190 -130 -263 -110 -324 -59 -707 132 -981 25 -35 42 -64 37 -64 -19 0 -241 119 -326 174 -188 122 -406 314 -532 468 l-58 71 108 103 c185 178 428 349 672 473 66 33 121 60 123 61 2 0 -10 -19 -26 -42z\"\/><path d=\"M2375 1950 c-198 -44 -350 -190 -395 -379 -18 -76 -8 -221 19 -290 114 -284 457 -406 731 -260 98 52 188 154 231 260 27 69 37 214 19 290 -38 163 -166 304 -326 360 -67 23 -215 33 -279 19z\"\/><\/g><\/svg><\/i> <img loading=\"lazy\" decoding=\"async\" width=\"16\" height=\"16\" alt=\"Loading\" src=\"https:\/\/aif.amtbbs.org\/wp-content\/plugins\/page-views-count\/ajax-loader-2x.gif\" border=0 \/><\/p>\n<div class=\"pvc_clear\"><\/div>\n","protected":false},"author":56,"featured_media":20698,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[23,80],"tags":[840],"class_list":["post-20697","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-23","category-80","tag-ai-"],"_links":{"self":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/20697","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/users\/56"}],"replies":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/comments?post=20697"}],"version-history":[{"count":1,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/20697\/revisions"}],"predecessor-version":[{"id":20702,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/20697\/revisions\/20702"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media\/20698"}],"wp:attachment":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media?parent=20697"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/categories?post=20697"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/tags?post=20697"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}