{"id":28239,"date":"2025-11-26T09:56:22","date_gmt":"2025-11-26T01:56:22","guid":{"rendered":"https:\/\/aif.amtbbs.org\/?p=28239"},"modified":"2025-11-26T09:56:22","modified_gmt":"2025-11-26T01:56:22","slug":"%e4%bb%8e%e4%ba%ba%e7%b1%bb%e6%bc%94%e7%a4%ba%e5%88%b0%e6%99%ba%e8%83%bd%e4%bd%93%e8%a1%8c%e4%b8%ba%ef%bc%8c%e9%80%9a%e8%bf%87%e5%8a%a8%e4%bd%9c%e7%89%87%e6%ae%b5%e9%87%8f%e5%8c%96%e5%ae%9e%e7%8e%b0","status":"publish","type":"post","link":"https:\/\/aif.amtbbs.org\/index.php\/2025\/11\/26\/28239\/","title":{"rendered":"\u4ece\u4eba\u7c7b\u6f14\u793a\u5230\u667a\u80fd\u4f53\u884c\u4e3a\uff0c\u901a\u8fc7\u52a8\u4f5c\u7247\u6bb5\u91cf\u5316\u5b9e\u73b0\u4eba\u7c7b\u98ce\u683c\u7684\u5f3a\u5316\u5b66\u4e60\u667a\u80fd\u4f53"},"content":{"rendered":"
\u56fe\u7075\u6d4b\u8bd5\u7684\u6838\u5fc3\u5e76\u4e0d\u662f\u8003\u5bdf\u667a\u80fd\u4f53\u662f\u5426\u80fd\u89e3\u51b3\u590d\u6742\u95ee\u9898\uff0c\u800c\u662f\u80fd\u5426\u5728\u884c\u4e3a\u548c\u4ea4\u6d41\u4e2d\u201c\u4ee5\u5047\u4e71\u771f\u201d\uff0c\u8ba9\u4eba\u7c7b\u65e0\u6cd5\u8f7b\u6613\u533a\u5206\u5176\u4e0e\u771f\u5b9e\u4eba\u7c7b\u7684\u5dee\u522b\u3002\u7c7b\u4eba\u667a\u80fd\u7684\u957f\u671f\u76ee\u6807\uff0c\u662f\u8ba9AI\u4e0d\u4ec5\u806a\u660e\uff0c\u800c\u4e14\u201c\u50cf\u4eba\u201d\u3002<\/div>\n
\n
\n

\u201c\u7c7b\u4eba\u667a\u80fd\u4f53\u201d\u4e00\u76f4\u662f\u4e00\u4e2a\u4ee4\u4eba\u7740\u8ff7\u7684\u76ee\u6807\u3002<\/p>\n

\u4eba\u4eec\u4e0d\u4ec5\u5e0c\u671b\u667a\u80fd\u4f53\u80fd\u591f\u5b8c\u6210\u4efb\u52a1\uff0c\u66f4\u5e0c\u671b\u5b83\u4eec\u7684\u884c\u4e3a\u65b9\u5f0f\u63a5\u8fd1\u4eba\u7c7b\uff0c\u5c55\u73b0\u51fa\u81ea\u7136\u3001\u53ef\u4fe1\u7684\u52a8\u4f5c\u4e0e\u51b3\u7b56\u3002\u8fd9\u79cd\u8ffd\u6c42\u4e0d\u4ec5\u5173\u4e4e\u6280\u672f\u6027\u80fd\uff0c\u66f4\u5173\u4e4e\u53ef\u89e3\u91ca\u6027\u4e0e\u793e\u4f1a\u63a5\u53d7\u5ea6\u3002<\/p>\n

\u4eba\u7c7b\u98ce\u683c\u667a\u80fd\u4f53\u7684\u610f\u4e49\u5728\u4e8e\uff0c\u5b83\u4eec\u80fd\u591f\u5728\u4e0e\u4eba\u7c7b\u4ea4\u4e92\u65f6\u51cf\u5c11\u8fdd\u548c\u611f\uff0c\u8ba9\u4eba\u4eec\u66f4\u5bb9\u6613\u4fe1\u4efb\u548c\u7406\u89e3\u5176\u884c\u4e3a\u3002<\/p>\n

\u65e0\u8bba\u662f\u5728\u673a\u5668\u4eba\u64cd\u4f5c\u3001\u865a\u62df\u52a9\u624b\uff0c\u8fd8\u662f\u6e38\u620f\u5bf9\u6218\u4e2d\uff0c\u4e00\u4e2a\u52a8\u4f5c\u6d41\u7545\u3001\u7b26\u5408\u4eba\u7c7b\u4e60\u60ef\u7684\u667a\u80fd\u4f53\uff0c\u5f80\u5f80\u6bd4\u4e00\u4e2a\u4ec5\u4ec5\u8ffd\u6c42\u9ad8\u5206\u6216\u6548\u7387\u7684\u667a\u80fd\u4f53\u66f4\u5177\u4ef7\u503c\u3002<\/p>\n

\u8fd9\u4e00\u76ee\u6807\u4e0e\u56fe\u7075\u6d4b\u8bd5\u7684\u7cbe\u795e\u9ad8\u5ea6\u5951\u5408\u3002<\/p>\n

\u56fe\u7075\u6d4b\u8bd5\u7684\u6838\u5fc3\u5e76\u4e0d\u662f\u8003\u5bdf\u667a\u80fd\u4f53\u662f\u5426\u80fd\u89e3\u51b3\u590d\u6742\u95ee\u9898\uff0c\u800c\u662f\u80fd\u5426\u5728\u884c\u4e3a\u548c\u4ea4\u6d41\u4e2d\u201c\u4ee5\u5047\u4e71\u771f\u201d\uff0c\u8ba9\u4eba\u7c7b\u65e0\u6cd5\u8f7b\u6613\u533a\u5206\u5176\u4e0e\u771f\u5b9e\u4eba\u7c7b\u7684\u5dee\u522b\u3002\u7c7b\u4eba\u667a\u80fd\u7684\u957f\u671f\u76ee\u6807\uff0c\u662f\u8ba9AI\u4e0d\u4ec5\u806a\u660e\uff0c\u800c\u4e14\u201c\u50cf\u4eba\u201d\u3002<\/p>\n

\u7136\u800c\uff0c\u73b0\u5b9e\u4e2d\u7684\u5f3a\u5316\u5b66\u4e60\u667a\u80fd\u4f53\u867d\u7136\u5728\u8bb8\u591a\u4efb\u52a1\u4e0a\u5df2\u7ecf\u8d85\u8d8a\u4eba\u7c7b\uff0c\u6bd4\u5982\u5728\u56f4\u68cb\u3001\u7535\u5b50\u6e38\u620f\u6216\u673a\u5668\u4eba\u63a7\u5236\u4e2d\u53d6\u5f97\u4e86\u8d85\u4eba\u8868\u73b0\uff0c\u4f46\u5b83\u4eec\u7684\u884c\u4e3a\u5f80\u5f80\u7f3a\u4e4f\u81ea\u7136\u6027\u3002<\/p>\n

\u5e38\u89c1\u7684\u95ee\u9898\u5305\u62ec\u52a8\u4f5c\u4e0d\u5e73\u6ed1\u3001\u5bfc\u822a\u65b9\u5f0f\u602a\u5f02\u3001\u751a\u81f3\u51fa\u73b0\u6296\u52a8\u6216\u65cb\u8f6c\u7b49\u201c\u975e\u4eba\u7c7b\u201d\u884c\u4e3a\u3002\u8fd9\u4e9b\u73b0\u8c61\u8ba9\u667a\u80fd\u4f53\u5728\u5b8c\u6210\u4efb\u52a1\u65f6\u663e\u5f97\u673a\u68b0\u800c\u50f5\u786c\uff0c\u96be\u4ee5\u83b7\u5f97\u4eba\u7c7b\u7684\u76f4\u89c9\u8ba4\u540c\u3002<\/p>\n

\u73b0\u6709\u7684\u7814\u7a76\u5c1d\u8bd5\u8fc7\u51e0\u79cd\u65b9\u6cd5\u6765\u89e3\u51b3\u8fd9\u4e00\u95ee\u9898\u3002\u5176\u4e00\u662f\u57fa\u4e8e\u89c4\u5219\u7684\u7ea6\u675f\uff0c\u901a\u8fc7\u4eba\u4e3a\u8bbe\u5b9a\u60e9\u7f5a\u673a\u5236\u6765\u907f\u514d\u975e\u4eba\u7c7b\u52a8\u4f5c\uff0c\u4f8b\u5982\u5bf9\u65cb\u8f6c\u3001\u6296\u52a8\u7b49\u884c\u4e3a\u52a0\u4ee5\u60e9\u7f5a\u3002\u4f46\u8fd9\u79cd\u65b9\u6cd5\u9700\u8981\u5927\u91cf\u4eba\u5de5\u8bbe\u8ba1\uff0c\u6210\u672c\u9ad8\u6602\u4e14\u96be\u4ee5\u63a8\u5e7f\u3002<\/p>\n

\u53e6\u4e00\u79cd\u65b9\u6cd5\u662f\u6a21\u4eff\u5b66\u4e60\uff0c\u901a\u8fc7\u76f4\u63a5\u5b66\u4e60\u4eba\u7c7b\u6f14\u793a\u6765\u63d0\u5347\u667a\u80fd\u4f53\u7684\u7c7b\u4eba\u6027\u3002\u7136\u800c\u6a21\u4eff\u5b66\u4e60\u7684\u6548\u679c\u9ad8\u5ea6\u4f9d\u8d56\u6f14\u793a\u6570\u636e\u7684\u8d28\u91cf\uff0c\u5f80\u5f80\u5728\u6027\u80fd\u4e0e\u4eba\u7c7b\u98ce\u683c\u4e4b\u95f4\u96be\u4ee5\u517c\u987e\u3002<\/p>\n

\u56e0\u6b64\uff0c\u6838\u5fc3\u6311\u6218\u5728\u4e8e\u5982\u4f55\u5728\u4fdd\u6301\u4efb\u52a1\u6210\u529f\u7387\u7684\u540c\u65f6\uff0c\u8ba9\u667a\u80fd\u4f53\u5c55\u73b0\u4eba\u7c7b\u822c\u7684\u884c\u4e3a\uff1f\u8fd9\u4e0d\u4ec5\u662f\u6280\u672f\u95ee\u9898\uff0c\u66f4\u662fAI\u53ef\u4fe1\u5ea6\u4e0e\u793e\u4f1a\u63a5\u53d7\u5ea6\u7684\u5173\u952e\u3002<\/p>\n

\u672c\u7814\u7a76\u7531\u4e2d\u56fd\u53f0\u6e7e\u7684\u4e24\u5927\u9876\u5c16\u5b66\u672f\u673a\u6784\u8054\u5408\u5b8c\u6210\u3002\u4e3b\u8981\u4f5c\u8005\u6765\u81ea\u56fd\u7acb\u9633\u660e\u4ea4\u901a\u5927\u5b66\uff08NYCU\uff09\u8ba1\u7b97\u673a\u79d1\u5b66\u7cfb\uff0c\u8be5\u6821\u5728\u4eba\u5de5\u667a\u80fd\u3001\u5f3a\u5316\u5b66\u4e60\u4e0e\u673a\u5668\u4eba\u63a7\u5236\u9886\u57df\u6709\u6df1\u539a\u79ef\u7d2f\u3002\u56e2\u961f\u6210\u5458\u5305\u62ec\u535a\u58eb\u751f\u4e0e\u9752\u5e74\u7814\u7a76\u4eba\u5458\uff0c\u8d1f\u8d23\u65b9\u6cd5\u8bbe\u8ba1\u3001\u5b9e\u9a8c\u5b9e\u73b0\u4e0e\u6570\u636e\u5206\u6790\u3002<\/p>\n

1.\u7406\u8bba\u57fa\u7840\u4e0e\u76f8\u5173\u5de5\u4f5c<\/h3>\n

\u8981\u7406\u89e3\u8fd9\u7bc7\u7814\u7a76\u7684\u521b\u65b0\u70b9\uff0c\u9996\u5148\u9700\u8981\u56de\u5230\u201c\u7c7b\u4eba\u5f3a\u5316\u5b66\u4e60\u201d\u7684\u57fa\u672c\u8109\u7edc\u3002<\/p>\n

\u8fc7\u53bb\u7684\u7814\u7a76\u5927\u81f4\u53ef\u4ee5\u5206\u4e3a\u4e24\u6761\u8def\u5f84\uff1a\u4e00\u662f\u901a\u8fc7\u884c\u4e3a\u7ea6\u675f\u6765\u907f\u514d\u667a\u80fd\u4f53\u51fa\u73b0\u4e0d\u81ea\u7136\u7684\u52a8\u4f5c\uff0c\u4e8c\u662f\u76f4\u63a5\u4ece\u4eba\u7c7b\u6f14\u793a\u4e2d\u5b66\u4e60\uff0c\u8bd5\u56fe\u8ba9\u667a\u80fd\u4f53\u6a21\u4eff\u4eba\u7c7b\u7684\u884c\u4e3a\u6a21\u5f0f\u3002<\/p>\n

\u5728\u884c\u4e3a\u7ea6\u675f\u7684\u8def\u5f84\u4e0a\uff0c\u7814\u7a76\u8005\u901a\u5e38\u4f1a\u5f15\u5165\u989d\u5916\u7684\u60e9\u7f5a\u673a\u5236\u3002\u4f8b\u5982\uff0c\u5f53\u667a\u80fd\u4f53\u51fa\u73b0\u65cb\u8f6c\u3001\u6296\u52a8\u7b49\u660e\u663e\u201c\u975e\u4eba\u7c7b\u201d\u7684\u52a8\u4f5c\u65f6\uff0c\u5c31\u4f1a\u88ab\u65bd\u52a0\u8d1f\u5956\u52b1\u3002<\/p>\n

\u8fd9\u79cd\u65b9\u6cd5\u7684\u4f18\u70b9\u662f\u76f4\u89c2\u6709\u6548\uff0c\u4f46\u7f3a\u70b9\u4e5f\u5f88\u660e\u663e\uff1a\u9700\u8981\u5927\u91cf\u4eba\u5de5\u8bbe\u8ba1\u89c4\u5219\uff0c\u4e14\u96be\u4ee5\u8986\u76d6\u6240\u6709\u53ef\u80fd\u7684\u975e\u4eba\u7c7b\u884c\u4e3a\u3002\u6362\u53e5\u8bdd\u8bf4\uff0c\u8fd9\u662f\u4e00\u79cd\u201c\u786c\u7f16\u7801\u201d\u7684\u65b9\u5f0f\uff0c\u6210\u672c\u9ad8\u6602\uff0c\u7075\u6d3b\u6027\u4e0d\u8db3\u3002<\/p>\n

\u53e6\u4e00\u6761\u8def\u5f84\u662f\u6a21\u4eff\u5b66\u4e60\u3002\u901a\u8fc7\u6536\u96c6\u4eba\u7c7b\u6f14\u793a\u6570\u636e\uff0c\u667a\u80fd\u4f53\u53ef\u4ee5\u76f4\u63a5\u5b66\u4e60\u4eba\u7c7b\u7684\u52a8\u4f5c\u5e8f\u5217\uff0c\u4ece\u800c\u5728\u884c\u4e3a\u4e0a\u66f4\u63a5\u8fd1\u4eba\u7c7b\u3002\u6a21\u4eff\u5b66\u4e60\u7684\u4ee3\u8868\u65b9\u6cd5\u5305\u62ec\u884c\u4e3a\u514b\u9686\uff08Behavior Cloning\uff09\u548c\u9006\u5f3a\u5316\u5b66\u4e60\uff08Inverse RL\uff09\u3002<\/p>\n

\u5b83\u4eec\u7684\u5171\u540c\u70b9\u662f\u4f9d\u8d56\u4eba\u7c7b\u6f14\u793a\u6765\u6784\u5efa\u5b66\u4e60\u76ee\u6807\u3002\u7136\u800c\uff0c\u8fd9\u79cd\u65b9\u6cd5\u7684\u5c40\u9650\u5728\u4e8e\uff1a\u6f14\u793a\u6570\u636e\u7684\u8d28\u91cf\u51b3\u5b9a\u4e86\u667a\u80fd\u4f53\u7684\u4e0a\u9650\u3002\u5982\u679c\u6f14\u793a\u6570\u636e\u6709\u9650\u6216\u4e0d\u591f\u7cbe\u7ec6\uff0c\u667a\u80fd\u4f53\u5f80\u5f80\u96be\u4ee5\u5728\u590d\u6742\u4efb\u52a1\u4e2d\u4fdd\u6301\u9ad8\u6027\u80fd\u3002<\/p>\n

\u4e3a\u4e86\u7a81\u7834\u8fd9\u4e9b\u9650\u5236\uff0c\u7814\u7a76\u56e2\u961f\u5f15\u5165\u4e86\u534a\u9a6c\u5c14\u53ef\u592b\u51b3\u7b56\u8fc7\u7a0b\uff08Semi-Markov Decision Process, SMDP\uff09\u7684\u6982\u5ff5\u3002\u4e0e\u4f20\u7edf\u7684\u9a6c\u5c14\u53ef\u592b\u51b3\u7b56\u8fc7\u7a0b\uff08MDP\uff09\u4e0d\u540c\uff0cSMDP\u5141\u8bb8\u667a\u80fd\u4f53\u5728\u4e00\u4e2a\u72b6\u6001\u4e0b\u6267\u884c\u4e00\u6bb5\u8fde\u7eed\u7684\u52a8\u4f5c\u5e8f\u5217\uff0c\u8fd9\u6bb5\u5e8f\u5217\u88ab\u79f0\u4e3a\u201c\u5b8f\u52a8\u4f5c\u201d\u3002<\/p>\n

\u5b8f\u52a8\u4f5c\u7684\u5b9a\u4e49\u5f88\u7b80\u5355\uff0c\u5b83\u662f\u4e00\u7ec4\u8fde\u7eed\u7684\u52a8\u4f5c\u7247\u6bb5\uff0c\u4f8b\u5982\u5728\u673a\u5668\u4eba\u63a7\u5236\u4e2d\uff0c\u4e00\u4e2a\u201c\u63a8\u95e8\u201d\u7684\u5b8f\u52a8\u4f5c\u53ef\u80fd\u5305\u542b\u591a\u4e2a\u8fde\u7eed\u7684\u624b\u81c2\u52a8\u4f5c\u3002<\/p>\n

\u5b8f\u52a8\u4f5c\u7684\u4f18\u52bf\u5728\u4e8e\u5b83\u964d\u4f4e\u4e86\u667a\u80fd\u4f53\u7684\u51b3\u7b56\u9891\u7387\u3002\u76f8\u6bd4\u9010\u6b65\u9009\u62e9\u6bcf\u4e00\u4e2a\u539f\u5b50\u52a8\u4f5c\uff0c\u667a\u80fd\u4f53\u53ef\u4ee5\u4e00\u6b21\u6027\u9009\u62e9\u4e00\u4e2a\u5b8f\u52a8\u4f5c\uff0c\u4ece\u800c\u66f4\u597d\u5730\u6355\u6349\u957f\u65f6\u4f9d\u8d56\u5173\u7cfb\u3002<\/p>\n

\u8fd9\u4e0d\u4ec5\u63d0\u5347\u4e86\u89c4\u5212\u6548\u7387\uff0c\u4e5f\u8ba9\u667a\u80fd\u4f53\u7684\u884c\u4e3a\u66f4\u63a5\u8fd1\u4eba\u7c7b\uff0c\u56e0\u4e3a\u4eba\u7c7b\u5728\u6267\u884c\u4efb\u52a1\u65f6\u5f80\u5f80\u4e5f\u662f\u4ee5\u201c\u52a8\u4f5c\u7247\u6bb5\u201d\u4e3a\u5355\u4f4d\uff0c\u800c\u4e0d\u662f\u9010\u6b65\u8ba1\u7b97\u6bcf\u4e00\u4e2a\u7ec6\u5fae\u52a8\u4f5c\u3002<\/p>\n

\u5728\u5b9e\u73b0\u5b8f\u52a8\u4f5c\u7684\u8fc7\u7a0b\u4e2d\uff0c\u7814\u7a76\u56e2\u961f\u91c7\u7528\u4e86\u5411\u91cf\u91cf\u5316\u53d8\u5206\u81ea\u7f16\u7801\u5668\uff08Vector-Quantized Variational Autoencoder, VQVAE\uff09\u3002VQVAE\u7684\u6838\u5fc3\u601d\u60f3\u662f\u5c06\u8fde\u7eed\u7684\u6f5c\u5728\u7a7a\u95f4\u79bb\u6563\u5316\uff0c\u5b66\u4e60\u51fa\u4e00\u7ec4\u6709\u9650\u7684\u201c\u7801\u672c\u201d\u3002<\/p>\n

\u6bcf\u4e00\u4e2a\u7801\u672c\u6761\u76ee\u4ee3\u8868\u4e00\u79cd\u6f5c\u5728\u7684\u5b8f\u52a8\u4f5c\u6a21\u5f0f\u3002\u901a\u8fc7\u8fd9\u79cd\u65b9\u5f0f\uff0c\u590d\u6742\u7684\u8fde\u7eed\u52a8\u4f5c\u7a7a\u95f4\u88ab\u8f6c\u5316\u4e3a\u7ed3\u6784\u5316\u7684\u79bb\u6563\u7a7a\u95f4\u3002<\/p>\n

\u8fd9\u79cd\u79bb\u6563\u6f5c\u5728\u7a7a\u95f4\u7684\u5b66\u4e60\u6709\u4e24\u4e2a\u5173\u952e\u597d\u5904\u3002\u9996\u5148\uff0c\u5b83\u8ba9\u667a\u80fd\u4f53\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u80fd\u591f\u66f4\u9ad8\u6548\u5730\u641c\u7d22\u52a8\u4f5c\u7a7a\u95f4\uff0c\u56e0\u4e3a\u4e0d\u518d\u9700\u8981\u5728\u65e0\u9650\u7684\u8fde\u7eed\u7a7a\u95f4\u4e2d\u63a2\u7d22\uff0c\u800c\u662f\u53ea\u9700\u5728\u6709\u9650\u7684\u7801\u672c\u4e2d\u9009\u62e9\u3002\u5176\u6b21\uff0c\u5b83\u4fdd\u8bc1\u4e86\u667a\u80fd\u4f53\u7684\u52a8\u4f5c\u7ed3\u6784\u66f4\u63a5\u8fd1\u4eba\u7c7b\u6f14\u793a\uff0c\u56e0\u4e3a\u7801\u672c\u672c\u8eab\u662f\u4ece\u4eba\u7c7b\u6570\u636e\u4e2d\u63d0\u70bc\u51fa\u6765\u7684\u3002\u667a\u80fd\u4f53\u7684\u6bcf\u4e00\u4e2a\u52a8\u4f5c\u9009\u62e9\uff0c\u90fd\u88ab\u7ea6\u675f\u5728\u201c\u4eba\u7c7b\u98ce\u683c\u201d\u7684\u8f68\u8ff9\u6d41\u5f62\u4e4b\u4e2d\u3002<\/p>\n

2.\u65b9\u6cd5\u6846\u67b6\uff1aMacro Action Quantization (MAQ)<\/h3>\n

\u8981\u8ba9\u5f3a\u5316\u5b66\u4e60\u667a\u80fd\u4f53\u771f\u6b63\u5c55\u73b0\u4eba\u7c7b\u98ce\u683c\uff0c\u7814\u7a76\u56e2\u961f\u63d0\u51fa\u4e86\u4e00\u4e2a\u5168\u65b0\u7684\u6846\u67b6\u2014\u2014\u5b8f\u52a8\u4f5c\u91cf\u5316\uff08Macro Action Quantization, MAQ\uff09\u3002<\/p>\n

\"\u56fe\u7247\"<\/p>\n

\u56fe1:MAQ\u7684\u6574\u4f53\u57f9\u8bad\u8fc7\u7a0b\u3002\uff081\uff09\u4eba\u7c7b\u884c\u4e3a\u84b8\u998f\uff1a\u5bf9\u72b6\u6001\uff08st\uff09\u548c\u5b8f\u89c2\u52a8\u4f5c\uff08mt=\uff08at\uff0cat+1\uff0c\u2026\uff0cat+H\u22121\uff09\uff09\u8bad\u7ec3\u6761\u4ef6VQVAE\uff0c\u4ee5\u5b66\u4e60\u79bb\u6563\u7801\u672c\uff1b\u901a\u8fc7\u52a8\u4f5c\u8f68\u8ff9\u4e0a\u7684\u6ed1\u52a8\u7a97\u53e3\u4ece\u4eba\u7c7b\u6f14\u793a\u4e2d\u63d0\u53d6\u5b8f\u89c2\u52a8\u4f5c\uff08mt\uff09\u3002\uff082\uff09MAQ\u5f3a\u5316\u5b66\u4e60\uff1a\u5728\u7ebf\u7b56\u7565\uff08\u03c0\u03b8\uff09\u901a\u8fc7\u9009\u62e9\u7801\u672c\u7d22\u5f15\u5728\u5b66\u4e60\u7684\u79bb\u6563\u7801\u7a7a\u95f4\u4e2d\u8d77\u4f5c\u7528\uff0cVQVAE\u5c06\u7801\u672c\u7d22\u5f15\u89e3\u7801\u4e3a\u5728\u73af\u5883\u4e2d\u6267\u884c\u7684\u5b8f\u52a8\u4f5c\u3002<\/p>\n

\u5b83\u7684\u6838\u5fc3\u601d\u60f3\u662f\u5c06\u4eba\u7c7b\u98ce\u683c\u5b9a\u4e49\u4e3a\u4e00\u4e2a\u8f68\u8ff9\u4f18\u5316\u95ee\u9898\uff0c\u5e76\u901a\u8fc7\u6eda\u52a8\u65f6\u57df\u63a7\u5236\uff08HRC\uff09\u5728\u77ed\u7247\u6bb5\u5185\u8fdb\u884c\u52a8\u4f5c\u4f18\u5316\uff0c\u8ba9\u667a\u80fd\u4f53\u7684\u884c\u4e3a\u65e2\u80fd\u5b8c\u6210\u4efb\u52a1\uff0c\u53c8\u80fd\u4fdd\u6301\u81ea\u7136\u6d41\u7545\u3002<\/p>\n

\u5728\u4f20\u7edf\u7684\u5f3a\u5316\u5b66\u4e60\u4e2d\uff0c\u8f68\u8ff9\u4f18\u5316\u901a\u5e38\u88ab\u8868\u8ff0\u4e3a\uff1a<\/p>\n

\"\u56fe\u7247\"\u56fe\u7247<\/span><\/p>\n

\u8fd9\u91cc\uff0c\u667a\u80fd\u4f53\u9700\u8981\u5728\u6574\u4e2a\u65f6\u95f4\u8303\u56f4\u00a0TT\u00a0\u5185\u5bfb\u627e\u6700\u4f18\u52a8\u4f5c\u5e8f\u5217\u3002\u7136\u800c\uff0c\u8fd9\u79cd\u5168\u5c40\u4f18\u5316\u5728\u590d\u6742\u4efb\u52a1\u4e2d\u5f80\u5f80\u4e0d\u53ef\u884c\u3002\u4e8e\u662f\uff0c\u7814\u7a76\u56e2\u961f\u5f15\u5165\u4e86\u6eda\u52a8\u65f6\u57df\u63a7\u5236\uff08HRC\uff09\uff0c\u5728\u77ed\u7247\u6bb5\u5185\u8fdb\u884c\u4f18\u5316\uff1a<\/p>\n

\"\u56fe\u7247\"\u56fe\u7247<\/span><\/p>\n

\u8fd9\u79cd\u65b9\u6cd5\u8ba9\u667a\u80fd\u4f53\u5728\u6bcf\u4e2a\u51b3\u7b56\u70b9\u53ea\u9700\u8003\u8651\u957f\u5ea6\u4e3a\u00a0HH\u00a0\u7684\u52a8\u4f5c\u5e8f\u5217\uff0c\u5e76\u4e0d\u65ad\u6eda\u52a8\u66f4\u65b0\uff0c\u4ece\u800c\u5728\u4fdd\u6301\u53cd\u5e94\u7075\u6d3b\u6027\u7684\u540c\u65f6\uff0c\u751f\u6210\u66f4\u81ea\u7136\u7684\u52a8\u4f5c\u8f68\u8ff9\u3002<\/p>\n

\u4e3a\u4e86\u8ba9\u8fd9\u4e9b\u77ed\u7247\u6bb5\u771f\u6b63\u5177\u5907\u4eba\u7c7b\u98ce\u683c\uff0cMAQ\u5f15\u5165\u4e86\u5b8f\u52a8\u4f5c\u91cf\u5316\u6d41\u7a0b\u3002\u9996\u5148\u662f\u4eba\u7c7b\u884c\u4e3a\u84b8\u998f\uff1a\u7814\u7a76\u56e2\u961f\u5229\u7528\u6761\u4ef6VQVAE\uff08Vector-Quantized Variational Autoencoder\uff09\u4ece\u4eba\u7c7b\u6f14\u793a\u4e2d\u63d0\u53d6\u5b8f\u52a8\u4f5c\uff0c\u5e76\u5efa\u7acb\u4e00\u4e2a\u79bb\u6563\u7684\u52a8\u4f5c\u7801\u672c\u3002<\/p>\n

\u6bcf\u4e2a\u7801\u672c\u6761\u76ee\u4ee3\u8868\u4e00\u79cd\u53ef\u80fd\u7684\u5b8f\u52a8\u4f5c\u6a21\u5f0f\uff0c\u76f8\u5f53\u4e8e\u628a\u4eba\u7c7b\u7684\u884c\u4e3a\u7247\u6bb5\u201c\u7f16\u7801\u201d\u6210\u53ef\u4f9b\u667a\u80fd\u4f53\u9009\u62e9\u7684\u79bb\u6563\u5355\u5143\u3002<\/p>\n

\u63a5\u4e0b\u6765\u662f\u52a8\u4f5c\u7a7a\u95f4\u7684\u8f6c\u5316\u3002\u901a\u8fc7\u5b8f\u52a8\u4f5c\u7684\u5f15\u5165\uff0c\u539f\u672c\u7684\u9a6c\u5c14\u53ef\u592b\u51b3\u7b56\u8fc7\u7a0b\uff08MDP\uff09\u88ab\u8f6c\u5316\u4e3a\u534a\u9a6c\u5c14\u53ef\u592b\u51b3\u7b56\u8fc7\u7a0b\uff08SMDP\uff09\u3002\u8fd9\u610f\u5473\u7740\u667a\u80fd\u4f53\u4e0d\u518d\u5728\u539f\u59cb\u52a8\u4f5c\u7a7a\u95f4\u4e2d\u9010\u6b65\u9009\u62e9\uff0c\u800c\u662f\u5728\u5b8f\u52a8\u4f5c\u7a7a\u95f4\u4e2d\u8fdb\u884c\u51b3\u7b56\uff1a<\/p>\n

\"\u56fe\u7247\"\u56fe\u7247<\/span><\/p>\n

\u5176\u4e2d\uff0c\u5b8f\u52a8\u4f5c\u4ee3\u8868\u4e00\u6bb5\u8fde\u7eed\u7684\u52a8\u4f5c\u5e8f\u5217\u3002<\/p>\n

\u6700\u540e\u662f\u667a\u80fd\u4f53\u8bad\u7ec3\u3002\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0c\u667a\u80fd\u4f53\u7684\u7b56\u7565\u4e0d\u518d\u76f4\u63a5\u8f93\u51fa\u4e00\u4e2a\u52a8\u4f5c\uff0c\u800c\u662f\u9009\u62e9\u7801\u672c\u4e2d\u7684\u7d22\u5f15\u3002\u9009\u5b9a\u7d22\u5f15\u540e\uff0c\u89e3\u7801\u5668\u4f1a\u5c06\u5176\u8fd8\u539f\u4e3a\u5bf9\u5e94\u7684\u5b8f\u52a8\u4f5c\uff0c\u5e76\u5728\u73af\u5883\u4e2d\u6267\u884c\u3002<\/p>\n

\u8fd9\u4e00\u8fc7\u7a0b\u8ba9\u667a\u80fd\u4f53\u7684\u884c\u4e3a\u59cb\u7ec8\u4fdd\u6301\u5728\u4eba\u7c7b\u6f14\u793a\u7684\u201c\u8f68\u8ff9\u6d41\u5f62\u201d\u4e2d\uff0c\u907f\u514d\u4e86\u504f\u79bb\u4eba\u7c7b\u98ce\u683c\u7684\u52a8\u4f5c\u3002<\/p>\n

MAQ\u7684\u4f18\u52bf\u975e\u5e38\u7a81\u51fa\u3002\u9996\u5148\uff0c\u5b83\u663e\u8457\u51cf\u5c11\u4e86\u52a8\u4f5c\u7a7a\u95f4\u7684\u590d\u6742\u5ea6\u3002\u539f\u59cb\u52a8\u4f5c\u7a7a\u95f4\u5f80\u5f80\u662f\u8fde\u7eed\u4e14\u5e9e\u5927\u7684\uff0c\u800c\u5b8f\u52a8\u4f5c\u7801\u672c\u5219\u662f\u6709\u9650\u7684\u3001\u7ed3\u6784\u5316\u7684\uff0c\u6781\u5927\u5730\u63d0\u5347\u4e86\u8bad\u7ec3\u6548\u7387\u3002<\/p>\n

\u5176\u6b21\uff0c\u5b83\u4fdd\u8bc1\u4e86\u667a\u80fd\u4f53\u7684\u884c\u4e3a\u59cb\u7ec8\u843d\u5728\u4eba\u7c7b\u6f14\u793a\u7684\u8f68\u8ff9\u8303\u56f4\u5185\u3002\u8fd9\u610f\u5473\u7740\u667a\u80fd\u4f53\u7684\u52a8\u4f5c\u4e0d\u4ec5\u80fd\u5b8c\u6210\u4efb\u52a1\uff0c\u8fd8\u80fd\u5c55\u73b0\u51fa\u81ea\u7136\u3001\u53ef\u4fe1\u7684\u7c7b\u4eba\u98ce\u683c\u3002<\/p>\n

\u4ece\u6574\u4f53\u6765\u770b\uff0cMAQ\u6846\u67b6\u7684\u8bbe\u8ba1\u65e2\u6709\u7406\u8bba\u4e0a\u7684\u4f18\u96c5\u6027\uff0c\u53c8\u6709\u5b9e\u8df5\u4e0a\u7684\u9ad8\u6548\u6027\u3002\u5b83\u901a\u8fc7\u8f68\u8ff9\u4f18\u5316\u4e0e\u52a8\u4f5c\u91cf\u5316\u7684\u7ed3\u5408\uff0c\u8ba9\u5f3a\u5316\u5b66\u4e60\u667a\u80fd\u4f53\u5728\u6027\u80fd\u4e0e\u4eba\u7c7b\u98ce\u683c\u4e4b\u95f4\u627e\u5230\u4e86\u65b0\u7684\u5e73\u8861\u70b9\u3002\u8fd9\u4e00\u65b9\u6cd5\u7684\u63d0\u51fa\uff0c\u6807\u5fd7\u7740\u7c7b\u4eba\u5f3a\u5316\u5b66\u4e60\u7814\u7a76\u8fc8\u51fa\u4e86\u5173\u952e\u4e00\u6b65\u3002<\/p>\n

3.\u5b9e\u9a8c\u8bbe\u8ba1\u4e0e\u8bc4\u4f30<\/h3>\n

\u4e3a\u4e86\u68c0\u9a8cMAQ\u6846\u67b6\u7684\u6709\u6548\u6027\uff0c\u7814\u7a76\u56e2\u961f\u9009\u62e9\u4e86D4RL Adroit\u57fa\u51c6\u4efb\u52a1\u4f5c\u4e3a\u5b9e\u9a8c\u573a\u666f\u3002<\/p>\n

\u8fd9\u4e00\u5957\u4efb\u52a1\u5728\u5f3a\u5316\u5b66\u4e60\u9886\u57df\u5e7f\u4e3a\u4eba\u77e5\uff0c\u6db5\u76d6\u4e86\u56db\u4e2a\u5177\u6709\u4ee3\u8868\u6027\u7684\u673a\u5668\u4eba\u64cd\u4f5c\uff1a\u5f00\u95e8\uff08Door\uff09\u3001\u9524\u51fb\uff08Hammer\uff09\u3001\u7b14\u8f6c\u52a8\uff08Pen\uff09\u3001\u4ee5\u53ca\u7269\u4f53\u642c\u79fb\uff08Relocate\uff09\u3002\u8fd9\u4e9b\u4efb\u52a1\u4e0d\u4ec5\u52a8\u4f5c\u590d\u6742\uff0c\u800c\u4e14\u90fd\u63d0\u4f9b\u4e86\u4eba\u7c7b\u6f14\u793a\u6570\u636e\uff0c\u56e0\u6b64\u975e\u5e38\u9002\u5408\u7528\u6765\u8bc4\u4f30\u667a\u80fd\u4f53\u7684\u201c\u7c7b\u4eba\u6027\u201d\u3002<\/p>\n

\u5728\u5bf9\u6bd4\u5b9e\u9a8c\u4e2d\uff0c\u56e2\u961f\u5f15\u5165\u4e86\u51e0\u79cd\u4e3b\u6d41\u7684\u5f3a\u5316\u5b66\u4e60\u7b97\u6cd5\u4f5c\u4e3a\u53c2\u7167\u3002\u5305\u62ecIQL\uff08Implicit Q-Learning\uff09\uff0c\u4e00\u79cd\u7ed3\u5408\u79bb\u7ebf\u4e0e\u5728\u7ebf\u8bad\u7ec3\u7684\u7b97\u6cd5\uff1bSAC\uff08Soft Actor-Critic\uff09\uff0c\u4ee5\u7a33\u5b9a\u6027\u548c\u63a2\u7d22\u80fd\u529b\u8457\u79f0\u7684\u5f3a\u5316\u5b66\u4e60\u65b9\u6cd5\uff1b\u4ee5\u53caRLPD\uff08Reinforcement Learning with Policy Distillation\uff09\uff0c\u4ee3\u8868\u4e86\u5f53\u524d\u79bb\u7ebf\u5f3a\u5316\u5b66\u4e60\u7684\u5148\u8fdb\u6c34\u5e73\u3002\u9664\u6b64\u4e4b\u5916\uff0c\u7814\u7a76\u56e2\u961f\u8fd8\u52a0\u5165\u4e86\u884c\u4e3a\u514b\u9686\uff08Behavior Cloning, BC\uff09\u4f5c\u4e3a\u57fa\u7ebf\u3002<\/p>\n

BC\u76f4\u63a5\u6a21\u4eff\u4eba\u7c7b\u6f14\u793a\uff0c\u867d\u7136\u5728\u52a8\u4f5c\u76f8\u4f3c\u5ea6\u4e0a\u6709\u4e00\u5b9a\u4f18\u52bf\uff0c\u4f46\u5728\u4efb\u52a1\u5b8c\u6210\u7387\u4e0a\u5f80\u5f80\u8868\u73b0\u4e0d\u4f73\uff0c\u56e0\u6b64\u5b83\u662f\u4e00\u4e2a\u5f88\u597d\u7684\u5bf9\u7167\u70b9\u3002<\/p>\n

\u4e3a\u4e86\u5168\u9762\u8861\u91cf\u667a\u80fd\u4f53\u7684\u8868\u73b0\uff0c\u7814\u7a76\u56e2\u961f\u8bbe\u8ba1\u4e86\u591a\u7ef4\u5ea6\u7684\u8bc4\u4ef7\u6307\u6807\u3002\u9996\u5148\u662f\u52a8\u6001\u65f6\u95f4\u89c4\u6574\uff08DTW\uff09\uff0c\u7528\u4e8e\u8861\u91cf\u667a\u80fd\u4f53\u8f68\u8ff9\u4e0e\u4eba\u7c7b\u6f14\u793a\u8f68\u8ff9\u7684\u76f8\u4f3c\u5ea6\u3002DTW\u80fd\u591f\u5904\u7406\u8f68\u8ff9\u957f\u5ea6\u548c\u65f6\u95f4\u4e0a\u7684\u5dee\u5f02\uff0c\u6355\u6349\u4e24\u8005\u5728\u52a8\u4f5c\u987a\u5e8f\u4e0a\u7684\u5339\u914d\u7a0b\u5ea6\u3002<\/p>\n

\u5176\u6b21\u662fWasserstein\u8ddd\u79bb\uff08WD\uff09\uff0c\u5b83\u4ece\u5206\u5e03\u89d2\u5ea6\u8861\u91cf\u667a\u80fd\u4f53\u4e0e\u4eba\u7c7b\u6f14\u793a\u7684\u5dee\u5f02\uff0c\u53cd\u6620\u6574\u4f53\u884c\u4e3a\u6a21\u5f0f\u7684\u63a5\u8fd1\u7a0b\u5ea6\u3002\u9664\u4e86\u76f8\u4f3c\u5ea6\u6307\u6807\uff0c\u56e2\u961f\u8fd8\u8003\u5bdf\u4e86\u4efb\u52a1\u6210\u529f\u7387\uff0c\u786e\u4fdd\u667a\u80fd\u4f53\u5728\u8ffd\u6c42\u7c7b\u4eba\u6027\u7684\u540c\u65f6\u4e0d\u4e27\u5931\u5b8c\u6210\u4efb\u52a1\u7684\u80fd\u529b\u3002<\/p>\n

\u6700\u540e\uff0c\u7814\u7a76\u56e2\u961f\u8fdb\u884c\u4e86\u4eba\u7c7b\u8bc4\u4f30\u5b9e\u9a8c\uff0c\u7c7b\u4f3c\u4e8e\u56fe\u7075\u6d4b\u8bd5\uff0c\u8ba9\u4eba\u7c7b\u53c2\u4e0e\u8005\u76f4\u63a5\u5224\u65ad\u667a\u80fd\u4f53\u7684\u884c\u4e3a\u662f\u5426\u50cf\u4eba\u7c7b\uff0c\u4ece\u800c\u83b7\u5f97\u66f4\u76f4\u89c2\u7684\u9a8c\u8bc1\u3002<\/p>\n

\u8fd9\u4e00\u5957\u5b9e\u9a8c\u8bbe\u8ba1\u65e2\u6709\u7406\u8bba\u4e0a\u7684\u4e25\u8c28\u6027\uff0c\u53c8\u6709\u5b9e\u8df5\u4e0a\u7684\u9488\u5bf9\u6027\u3002\u901a\u8fc7\u591a\u7ef4\u5ea6\u7684\u91cf\u5316\u6307\u6807\u4e0e\u4eba\u7c7b\u76f4\u89c9\u7684\u7ed3\u5408\uff0c\u7814\u7a76\u56e2\u961f\u80fd\u591f\u5168\u9762\u8bc4\u4f30MAQ\u6846\u67b6\u5728\u201c\u7c7b\u4eba\u6027\u201d\u4e0e\u201c\u4efb\u52a1\u6027\u80fd\u201d\u4e4b\u95f4\u7684\u5e73\u8861\u3002\u8fd9\u4e5f\u4e3a\u540e\u7eed\u7684\u5b9e\u9a8c\u7ed3\u679c\u5206\u6790\u5960\u5b9a\u4e86\u575a\u5b9e\u57fa\u7840\u3002<\/p>\n

4.\u5b9e\u9a8c\u7ed3\u679c\u4e0e\u5206\u6790<\/h3>\n

\u5b9e\u9a8c\u7ed3\u679c\u975e\u5e38\u4ee4\u4eba\u632f\u594b\u3002\u9996\u5148\uff0c\u5728\u8f68\u8ff9\u76f8\u4f3c\u5ea6\u65b9\u9762\uff0cMAQ\u5728\u6240\u6709\u4efb\u52a1\u4e2d\u90fd\u663e\u8457\u63d0\u9ad8\u4e86DTW\u548cWD\u5206\u6570\u3002\u65e0\u8bba\u662f\u72b6\u6001\u5e8f\u5217\u8fd8\u662f\u52a8\u4f5c\u5e8f\u5217\uff0cMAQ\u667a\u80fd\u4f53\u7684\u8868\u73b0\u90fd\u66f4\u63a5\u8fd1\u4eba\u7c7b\u6f14\u793a\u3002\u5c24\u5176\u662f\u5728SAC\u7b97\u6cd5\u4e0a\uff0c\u5dee\u8ddd\u6700\u4e3a\u660e\u663e\u3002<\/p>\n

\u4f20\u7edfSAC\u5b8c\u5168\u4f9d\u8d56\u73af\u5883\u4ea4\u4e92\uff0c\u7f3a\u4e4f\u4eba\u7c7b\u6f14\u793a\u7684\u7ea6\u675f\uff0c\u56e0\u6b64\u5176\u884c\u4e3a\u4e0e\u4eba\u7c7b\u5dee\u8ddd\u8f83\u5927\u3002\u800c\u5728\u5f15\u5165MAQ\u540e\uff0cSAC\u7684\u8f68\u8ff9\u76f8\u4f3c\u5ea6\u5927\u5e45\u63d0\u5347\uff0c\u663e\u793a\u51faMAQ\u5bf9\u7eaf\u4ea4\u4e92\u578b\u7b97\u6cd5\u7684\u6539\u9020\u6548\u679c\u5c24\u4e3a\u663e\u8457\u3002<\/p>\n

\"\u56fe\u7247\"\u56fe\u7247<\/span><\/p>\n

\u88681:D4RL\u7684Adroit\u57fa\u51c6\u6d4b\u8bd5\u7ed3\u679c\u3002\u6bcf\u79cd\u7b97\u6cd5\u7684\u57fa\u7ebf\u4e0eMAQ\u4e0d\u4e00\u81f4\u901a\u5e38\u4f1a\u63d0\u9ad8\u6bcf\u4e2a\u63a7\u5236\u4efb\u52a1\u4e2d\u6240\u6709\u539f\u59cb\u7b97\u6cd5\u7684\u76f8\u4f3c\u6027\u5f97\u5206\u3002<\/p>\n

\u5728\u4efb\u52a1\u6210\u529f\u7387\u65b9\u9762\uff0cMAQ\u667a\u80fd\u4f53\u540c\u6837\u8868\u73b0\u51fa\u8272\u3002\u5728Door\u548cHammer\u4efb\u52a1\u4e2d\uff0cMAQ\u667a\u80fd\u4f53\u7684\u6210\u529f\u7387\u63a5\u8fd1\u751a\u81f3\u8d85\u8fc7\u4e86\u57fa\u7ebf\u7b97\u6cd5\u3002\u8fd9\u8bf4\u660eMAQ\u4e0d\u4ec5\u8ba9\u667a\u80fd\u4f53\u66f4\u50cf\u4eba\u7c7b\uff0c\u800c\u4e14\u6ca1\u6709\u727a\u7272\u4efb\u52a1\u5b8c\u6210\u7684\u80fd\u529b\u3002<\/p>\n

\"\u56fe\u7247\"\u56fe\u7247<\/span><\/p>\n

\u56fe2:MAQ+RLPD\u4e2d\u4e0d\u540c\u957f\u5ea6\u7684\u8f68\u8ff9\u76f8\u4f3c\u6027\u5f97\u5206\u548c\u6210\u529f\u7387\u3002<\/p>\n

\u76f8\u6bd4\u4e4b\u4e0b\uff0cBC\u867d\u7136\u5728\u76f8\u4f3c\u5ea6\u4e0a\u6709\u4e00\u5b9a\u4f18\u52bf\uff0c\u4f46\u7531\u4e8e\u5b83\u53ea\u4f1a\u673a\u68b0\u5730\u6a21\u4eff\u6f14\u793a\u6570\u636e\uff0c\u7f3a\u4e4f\u4f18\u5316\u80fd\u529b\uff0c\u56e0\u6b64\u5728\u6210\u529f\u7387\u4e0a\u8868\u73b0\u6700\u5dee\u3002<\/p>\n

\"\u56fe3\uff1a\u56fe\u7075\u6d4b\u8bd5\u3002\u8bef\u5dee\u6761\u8868\u793a96%\u7684\u7f6e\u4fe1\u533a\u95f4\u3002\"\u56fe3\uff1a\u56fe\u7075\u6d4b\u8bd5\u3002\u8bef\u5dee\u6761\u8868\u793a96%\u7684\u7f6e\u4fe1\u533a\u95f4\u3002<\/span><\/p>\n

\u56fe3\u7684\u6761\u5f62\u56fe\u7edf\u8ba1\u4e86\u5404\u7c7b\u667a\u80fd\u4f53\u7684\u83b7\u80dc\u7387\uff0c\u5373\u5728\u8bc4\u4f30\u8005\u773c\u4e2d\uff0c\u667a\u80fd\u4f53\u7684\u884c\u4e3a\u88ab\u8bef\u8ba4\u4e3a\u6bd4\u4eba\u7c7b\u6f14\u793a\u66f4\u50cf\u4eba\u7c7b\u7684\u6bd4\u4f8b\u3002<\/p>\n

\u7ed3\u679c\u4e0e\u524d\u6587\u7684\u8f68\u8ff9\u76f8\u4f3c\u5ea6\u5206\u6790\u9ad8\u5ea6\u4e00\u81f4\uff1a\u5f15\u5165MAQ\u7684\u667a\u80fd\u4f53\u5728\u83b7\u80dc\u7387\u4e0a\u663e\u8457\u8d85\u8fc7\u4e86\u6ca1\u6709MAQ\u7684\u4f20\u7edfRL\u7b97\u6cd5\u3002\u8fd9\u8bf4\u660e\uff0c\u901a\u8fc7\u5b8f\u52a8\u4f5c\u91cf\u5316\u8fdb\u884c\u8f68\u8ff9\u4f18\u5316\uff0c\u5c24\u5176\u662f\u91c7\u7528\u8f83\u957f\u7684\u5b8f\u52a8\u4f5c\u7247\u6bb5\uff0c\u80fd\u591f\u6709\u6548\u63d0\u5347\u667a\u80fd\u4f53\u7684\u7c7b\u4eba\u6027\u3002<\/p>\n

\u6700\u5177\u8bf4\u670d\u529b\u7684\u662f\u4eba\u7c7b\u8bc4\u4f30\u7ed3\u679c\u3002\u5728\u5b9e\u9a8c\u4e2d\uff0c\u53c2\u4e0e\u8005\u80fd\u591f\u8f7b\u6613\u8bc6\u522b\u51fa\u4f20\u7edfRL\u667a\u80fd\u4f53\u7684\u201c\u975e\u4eba\u7c7b\u201d\u884c\u4e3a\uff0c\u6bd4\u5982\u4e0d\u81ea\u7136\u7684\u6296\u52a8\u6216\u5947\u602a\u7684\u52a8\u4f5c\u6a21\u5f0f\u3002<\/p>\n

\"\u56fe4\uff1a\u4eba\u7c7b\u76f8\u4f3c\u5ea6\u6392\u540d\u6d4b\u8bd5\u3002\"\u56fe4\uff1a\u4eba\u7c7b\u76f8\u4f3c\u5ea6\u6392\u540d\u6d4b\u8bd5\u3002<\/span><\/p>\n

\u56fe4\u901a\u8fc7\u70ed\u56fe\u7684\u5f62\u5f0f\u603b\u7ed3\u4e86\u56db\u4e2a\u4efb\u52a1\u7684\u5bf9\u6bd4\u7ed3\u679c\uff0c\u6bcf\u4e2a\u5355\u5143\u683c\u4ee3\u8868\u8bc4\u4f30\u8005\u5728\u76f4\u63a5\u6bd4\u8f83\u4e24\u4e2a\u667a\u80fd\u4f53\u7684\u89c6\u9891\u65f6\uff0c\u8ba4\u4e3a\u7eb5\u8f74\u4e0a\u7684\u667a\u80fd\u4f53\u6bd4\u6a2a\u8f74\u4e0a\u7684\u667a\u80fd\u4f53\u66f4\u50cf\u4eba\u7c7b\u7684\u9891\u7387\u3002<\/p>\n

\u6574\u4f53\u6392\u540d\u975e\u5e38\u6e05\u6670\uff1a\u4eba\u7c7b\uff0874%\uff09\u00a0> MAQ+RLPD\uff0871%\uff09\u00a0> MAQ+SAC\uff0861%\uff09\u00a0= MAQ+IQL\uff0861%\uff09\u00a0> BC\uff0843%\uff09\u00a0> RLPD\uff0838%\uff09\u00a0= IQL\uff0838%\uff09\u00a0> SAC\uff0814%\uff09\u3002\u8fd9\u4e00\u987a\u5e8f\u4e0e\u524d\u9762\u56fe\u7075\u6d4b\u8bd5\u7684\u7ed3\u679c\u9ad8\u5ea6\u4e00\u81f4\uff0c\u4ec5\u5728IQL\u4e0eSAC\u7684\u4f4d\u7f6e\u4e0a\u51fa\u73b0\u4e86\u5c0f\u5e45\u98a0\u5012\u3002\u8fd9\u79cd\u4e00\u81f4\u6027\u8bf4\u660e\u4eba\u7c7b\u8bc4\u4f30\u8005\u7684\u5224\u65ad\u662f\u7a33\u5b9a\u7684\uff0c\u4ed6\u4eec\u80fd\u591f\u660e\u786e\u533a\u5206\u4eba\u7c7b\u6f14\u793a\u4e0e\u4f20\u7edfRL\u667a\u80fd\u4f53\u7684\u884c\u4e3a\u5dee\u5f02\u3002<\/p>\n

\u66f4\u4ee4\u4eba\u77a9\u76ee\u7684\u662f\uff0cMAQ+RLPD\u7684\u83b7\u80dc\u7387\u8fbe\u523071%\uff0c\u51e0\u4e4e\u4e0e\u4eba\u7c7b\u6f14\u793a\u768474%\u6301\u5e73\u3002\u8fd9\u610f\u5473\u7740\u5728\u76f4\u63a5\u5bf9\u6bd4\u4e2d\uff0cMAQ+RLPD\u5df2\u7ecf\u80fd\u591f\u5728\u5927\u591a\u6570\u60c5\u51b5\u4e0b\u201c\u6b3a\u9a97\u201d\u4eba\u7c7b\u8bc4\u4f30\u8005\uff0c\u8ba9\u4ed6\u4eec\u8bef\u4ee5\u4e3a\u5176\u884c\u4e3a\u4e0e\u4eba\u7c7b\u4e00\u6837\u81ea\u7136\u3002\u8fd9\u4e00\u7ed3\u679c\u4e0d\u4ec5\u662f\u6570\u503c\u4e0a\u7684\u7a81\u7834\uff0c\u66f4\u662f\u611f\u77e5\u5c42\u9762\u7684\u91cd\u5927\u8fdb\u5c55\uff1a\u5b83\u8868\u660e\u901a\u8fc7\u5b8f\u52a8\u4f5c\u91cf\u5316\u4e0e\u8f68\u8ff9\u4f18\u5316\uff0c\u5f3a\u5316\u5b66\u4e60\u667a\u80fd\u4f53\u53ef\u4ee5\u5728\u884c\u4e3a\u8868\u73b0\u4e0a\u903c\u8fd1\u4eba\u7c7b\u7684\u6c34\u5e73\u3002<\/p>\n

\u5f53\u9762\u5bf9MAQ\u667a\u80fd\u4f53\u65f6\uff0c\u53c2\u4e0e\u8005\u5f80\u5f80\u96be\u4ee5\u533a\u5206\u5176\u4e0e\u4eba\u7c7b\u6f14\u793a\u7684\u5dee\u522b\u3002\u8fd9\u4e00\u7ed3\u679c\u51e0\u4e4e\u5c31\u662f\u7c7b\u56fe\u7075\u6d4b\u8bd5\u7684\u80dc\u5229\uff0c\u8bc1\u660eMAQ\u5728\u884c\u4e3a\u5c42\u9762\u771f\u6b63\u5b9e\u73b0\u4e86\u201c\u4ee5\u5047\u4e71\u771f\u201d\u3002<\/p>\n

5.\u7814\u7a76\u8d21\u732e\u4e0e\u610f\u4e49<\/h3>\n

\u8fd9\u9879\u7814\u7a76\u7684\u6700\u5927\u4eae\u70b9\u5728\u4e8e\u65b9\u6cd5\u4e0a\u7684\u521b\u65b0\u3002\u56e2\u961f\u63d0\u51fa\u4e86\u5b8f\u52a8\u4f5c\u91cf\u5316\uff08Macro Action Quantization, MAQ\uff09\u6846\u67b6\uff0c\u901a\u8fc7\u6761\u4ef6VQVAE\u5c06\u4eba\u7c7b\u6f14\u793a\u8f6c\u5316\u4e3a\u53ef\u5b66\u4e60\u7684\u79bb\u6563\u52a8\u4f5c\u7a7a\u95f4\u3002<\/p>\n

\u76f8\u6bd4\u4f20\u7edf\u7684\u9010\u6b65\u52a8\u4f5c\u9009\u62e9\uff0cMAQ\u8ba9\u667a\u80fd\u4f53\u5728\u66f4\u9ad8\u5c42\u6b21\u7684\u201c\u5b8f\u52a8\u4f5c\u201d\u7a7a\u95f4\u4e2d\u8fdb\u884c\u51b3\u7b56\uff0c\u4ece\u800c\u907f\u514d\u4e86\u673a\u68b0\u3001\u50f5\u786c\u7684\u884c\u4e3a\u6a21\u5f0f\u3002\u8fd9\u79cd\u65b9\u6cd5\u4e0d\u4ec5\u4f18\u96c5\u5730\u89e3\u51b3\u4e86\u52a8\u4f5c\u7a7a\u95f4\u8fc7\u4e8e\u5e9e\u5927\u7684\u95ee\u9898\uff0c\u8fd8\u786e\u4fdd\u4e86\u667a\u80fd\u4f53\u7684\u884c\u4e3a\u59cb\u7ec8\u843d\u5728\u4eba\u7c7b\u6f14\u793a\u7684\u8f68\u8ff9\u6d41\u5f62\u4e4b\u4e2d\u3002<\/p>\n

\u66f4\u91cd\u8981\u7684\u662f\uff0cMAQ\u5728\u6027\u80fd\u4e0e\u4eba\u7c7b\u98ce\u683c\u4e4b\u95f4\u627e\u5230\u4e86\u5e73\u8861\u70b9\u3002\u8fc7\u53bb\u7684\u6a21\u4eff\u5b66\u4e60\u5f80\u5f80\u5728\u7c7b\u4eba\u6027\u4e0a\u6709\u6240\u63d0\u5347\uff0c\u4f46\u727a\u7272\u4e86\u4efb\u52a1\u6210\u529f\u7387\uff1b\u800c\u7eaf\u7cb9\u7684\u5f3a\u5316\u5b66\u4e60\u7b97\u6cd5\u867d\u7136\u80fd\u5b8c\u6210\u4efb\u52a1\uff0c\u5374\u7f3a\u4e4f\u81ea\u7136\u6027\u3002<\/p>\n

MAQ\u7684\u8d21\u732e\u5728\u4e8e\uff0c\u5b83\u65e2\u4fdd\u6301\u4e86\u4efb\u52a1\u5b8c\u6210\u7684\u9ad8\u6210\u529f\u7387\uff0c\u53c8\u663e\u8457\u63d0\u5347\u4e86\u4eba\u7c7b\u76f8\u4f3c\u5ea6\u3002\u8fd9\u610f\u5473\u7740\u667a\u80fd\u4f53\u4e0d\u518d\u9700\u8981\u5728\u201c\u6548\u7387\u201d\u4e0e\u201c\u7c7b\u4eba\u6027\u201d\u4e4b\u95f4\u505a\u51fa\u53d6\u820d\uff0c\u800c\u662f\u80fd\u591f\u517c\u987e\u4e24\u8005\u3002<\/p>\n

\u5728\u5e94\u7528\u6f5c\u529b\u65b9\u9762\uff0cMAQ\u5c55\u73b0\u51fa\u6781\u5927\u7684\u6269\u5c55\u6027\u3002\u5b83\u4e0d\u4ec5\u9002\u7528\u4e8e\u5f53\u524d\u7684\u57fa\u51c6\u4efb\u52a1\uff0c\u8fd8\u53ef\u4ee5\u63a8\u5e7f\u5230\u66f4\u590d\u6742\u7684\u573a\u666f\uff0c\u5c24\u5176\u662f\u8fde\u7eed\u52a8\u4f5c\u7a7a\u95f4\u7684\u4efb\u52a1\u3002<\/p>\n

\u968f\u7740\u673a\u5668\u4eba\u63a7\u5236\u3001\u81ea\u52a8\u9a7e\u9a76\u548c\u865a\u62df\u4ea4\u4e92\u7684\u590d\u6742\u5ea6\u4e0d\u65ad\u63d0\u5347\uff0cMAQ\u7684\u5b8f\u52a8\u4f5c\u91cf\u5316\u7b56\u7565\u80fd\u591f\u5e2e\u52a9\u667a\u80fd\u4f53\u5728\u8fd9\u4e9b\u573a\u666f\u4e2d\u5c55\u73b0\u66f4\u81ea\u7136\u3001\u66f4\u53ef\u4fe1\u7684\u884c\u4e3a\u3002<\/p>\n

\u672a\u6765\u7684\u7814\u7a76\u65b9\u5411\u540c\u6837\u4ee4\u4eba\u671f\u5f85\u3002\u9996\u5148\u662f\u591a\u6a21\u6001\u4eba\u7c7b\u6f14\u793a\u7684\u5f15\u5165\uff0c\u4e0d\u4ec5\u4ec5\u662f\u52a8\u4f5c\u6570\u636e\uff0c\u8fd8\u5305\u62ec\u89c6\u89c9\u4fe1\u606f\u751a\u81f3\u8bed\u8a00\u6307\u4ee4\uff0c\u8fd9\u5c06\u8ba9\u667a\u80fd\u4f53\u7684\u5b66\u4e60\u66f4\u5168\u9762\u3001\u66f4\u8d34\u8fd1\u771f\u5b9e\u4eba\u7c7b\u7684\u51b3\u7b56\u8fc7\u7a0b\u3002\u5176\u6b21\u662f\u66f4\u5927\u89c4\u6a21\u7684\u8f68\u8ff9\u4f18\u5316\uff0c\u901a\u8fc7\u66f4\u4e30\u5bcc\u7684\u6570\u636e\u548c\u66f4\u5f3a\u7684\u8ba1\u7b97\u80fd\u529b\uff0c\u8fdb\u4e00\u6b65\u63d0\u5347\u667a\u80fd\u4f53\u7684\u7c7b\u4eba\u6027\u3002<\/p>\n

\u6700\u540e\uff0c\u5728\u4eba\u673a\u4ea4\u4e92\u4e0e\u673a\u5668\u4eba\u63a7\u5236\u9886\u57df\uff0cMAQ\u7684\u5e94\u7528\u524d\u666f\u5e7f\u9614\u3002\u65e0\u8bba\u662f\u670d\u52a1\u673a\u5668\u4eba\u3001\u533b\u7597\u8f85\u52a9\uff0c\u8fd8\u662f\u865a\u62df\u52a9\u624b\uff0c\u7c7b\u4eba\u98ce\u683c\u7684\u667a\u80fd\u4f53\u90fd\u80fd\u663e\u8457\u63d0\u5347\u7528\u6237\u4f53\u9a8c\u4e0e\u4fe1\u4efb\u5ea6\u3002<\/p>\n

\u603b\u4f53\u800c\u8a00\uff0cMAQ\u6846\u67b6\u4e3a\u201c\u7c7b\u4eba\u5f3a\u5316\u5b66\u4e60\u201d\u5f00\u8f9f\u4e86\u4e00\u6761\u65b0\u7684\u7814\u7a76\u8def\u5f84\u3002\u5b83\u901a\u8fc7\u8f68\u8ff9\u4f18\u5316\u4e0e\u52a8\u4f5c\u91cf\u5316\u7684\u7ed3\u5408\uff0c\u8ba9\u667a\u80fd\u4f53\u4e0d\u4ec5\u80fd\u591f\u5b8c\u6210\u4efb\u52a1\uff0c\u8fd8\u80fd\u5c55\u73b0\u81ea\u7136\u3001\u53ef\u4fe1\u7684\u4eba\u7c7b\u98ce\u683c\u3002\u8fd9\u79cd\u65b9\u6cd5\u7684\u63d0\u51fa\uff0c\u6807\u5fd7\u7740\u5f3a\u5316\u5b66\u4e60\u7814\u7a76\u4ece\u5355\u7eaf\u8ffd\u6c42\u6027\u80fd\uff0c\u8fc8\u5411\u517c\u987e\u4eba\u7c7b\u98ce\u683c\u4e0e\u793e\u4f1a\u63a5\u53d7\u5ea6\u7684\u65b0\u9636\u6bb5\u3002<\/p>\n

\u5728\u672a\u6765\u7684AI\u8bbe\u8ba1\u4e2d\uff0c\u53ef\u4fe1\u5ea6\u4e0e\u53ef\u89e3\u91ca\u6027\u5c06\u6210\u4e3a\u6838\u5fc3\u8981\u6c42\u3002MAQ\u7684\u7814\u7a76\u6210\u679c\u4e3a\u8fd9\u4e00\u8d8b\u52bf\u63d0\u4f9b\u4e86\u91cd\u8981\u542f\u793a\uff1a\u667a\u80fd\u4f53\u7684\u6210\u529f\u4e0d\u5e94\u4ec5\u4ec5\u4ee5\u4efb\u52a1\u5b8c\u6210\u7387\u8861\u91cf\uff0c\u66f4\u5e94\u8003\u8651\u5176\u884c\u4e3a\u662f\u5426\u7b26\u5408\u4eba\u7c7b\u7684\u76f4\u89c9\u4e0e\u4e60\u60ef\u3002\u901a\u8fc7\u8fd9\u79cd\u7c7b\u4eba\u5316\u7684\u8bbe\u8ba1\uff0cAI\u4e0d\u4ec5\u80fd\u6210\u4e3a\u9ad8\u6548\u7684\u5de5\u5177\uff0c\u66f4\u80fd\u6210\u4e3a\u503c\u5f97\u4fe1\u8d56\u7684\u4f19\u4f34\u3002<\/p>\n

\u8fd9\u7bc7\u7814\u7a76\u7684\u610f\u4e49\u5728\u4e8e\uff0c\u5b83\u8ba9\u6211\u4eec\u770b\u5230\u4e86\u4e00\u4e2a\u66f4\u81ea\u7136\u3001\u66f4\u4eba\u6027\u5316\u7684AI\u672a\u6765\u3002MAQ\u6846\u67b6\u7684\u63d0\u51fa\uff0c\u4e0d\u53ea\u662f\u6280\u672f\u4e0a\u7684\u7a81\u7834\uff0c\u66f4\u662f\u5bf9\u201c\u7c7b\u4eba\u667a\u80fd\u201d\u8fd9\u4e00\u957f\u671f\u76ee\u6807\u7684\u6709\u529b\u56de\u5e94\u3002\uff08END\uff09<\/p>\n

\u53c2\u8003\u8d44\u6599\uff1ahttps:\/\/arxiv.org\/pdf\/2511.15055<\/p>\n

\u6587\u7ae0\u6765\u81ea\uff1a51CTO<\/p>\n<\/div>\n<\/div>\n

<\/div>\n

<\/g><\/svg><\/i> \"Loading\"<\/p>\n

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

\u56fe\u7075\u6d4b\u8bd5\u7684\u6838\u5fc3\u5e76\u4e0d\u662f\u8003\u5bdf\u667a\u80fd\u4f53\u662f\u5426\u80fd\u89e3\u51b3\u590d\u6742\u95ee\u9898\uff0c\u800c\u662f\u80fd\u5426\u5728\u884c\u4e3a\u548c\u4ea4\u6d41\u4e2d\u201c\u4ee5\u5047\u4e71\u771f\u201d\uff0c\u8ba9\u4eba\u7c7b\u65e0\u6cd5\u8f7b\u6613\u533a\u5206\u5176\u4e0e\u771f […]<\/p>\n

<\/div>\n

<\/g><\/svg><\/i> \"Loading\"<\/p>\n

<\/div>\n","protected":false},"author":56,"featured_media":28240,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3,23,20,80],"tags":[1996],"class_list":["post-28239","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai","category-23","category-20","category-80","tag-1996"],"_links":{"self":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/28239","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=28239"}],"version-history":[{"count":1,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/28239\/revisions"}],"predecessor-version":[{"id":28248,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/28239\/revisions\/28248"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media\/28240"}],"wp:attachment":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media?parent=28239"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/categories?post=28239"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/tags?post=28239"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}