{"id":856,"date":"2025-07-22T01:10:00","date_gmt":"2025-07-21T17:10:00","guid":{"rendered":"https:\/\/blog.sssn.tech\/?p=856"},"modified":"2025-07-22T01:11:05","modified_gmt":"2025-07-21T17:11:05","slug":"%e5%9f%ba%e4%ba%8eisaaclab%e7%9a%84%e4%ba%ba%e5%bd%a2%e6%9c%ba%e5%99%a8%e4%ba%ba%e4%bb%bf%e7%9c%9f%e4%b8%8e%e5%bc%ba%e5%8c%96%e5%ad%a6%e4%b9%a0%e8%ae%ad%e7%bb%83","status":"publish","type":"post","link":"https:\/\/blog.sssn.tech\/?p=856","title":{"rendered":"\u57fa\u4e8eIsaaclab\u7684\u4eba\u5f62\u673a\u5668\u4eba\u4eff\u771f\u4e0e\u5f3a\u5316\u5b66\u4e60\u8bad\u7ec3"},"content":{"rendered":"<p>\u5f00\u6e90\u4ee3\u7801\uff0c\u5b9e\u9a8c\u6570\u636e\uff0c\u89c6\u9891\u5c55\u793a: <a href=\"https:\/\/github.com\/sssn-tech\/Learn2Walk\">https:\/\/github.com\/sssn-tech\/Learn2Walk<\/a><\/p>\n<h1>Task0: Isaaclab\u73af\u5883\u914d\u7f6e<\/h1>\n<h2>\u5e38\u89c1\u5751\u70b9<\/h2>\n<p>Isaaclab\u7684\u521d\u6b65\u73af\u5883\u642d\u5efa\u5c31\u6d89\u53ca\u5230\u975e\u5e38\u591a\u7684\u5751\u70b9\uff0c\u8fd9\u91cc\u5217\u4e3e\u4e00\u4e9b\u3002<\/p>\n<p>\u8fd9\u4e9b\u5751\u70b9\u7684\u4e3b\u8981\u539f\u56e0\u662fIsaaclab\u7684\u5b98\u65b9\u6587\u6863\u5c06Isaacsim\uff08Isaaclab\u7684\u57fa\u7840\u73af\u5883\uff09\u7684\u5b89\u88c5\u4e00\u7b14\u5e26\u8fc7\u4e86\uff0c\u5b98\u65b9\u7ed9\u51fa\u4e86\u4e00\u4e2aIsaacsim\u5b89\u88c5\u547d\u4ee4\uff0c\u4f46\u662f\u6ca1\u6709\u4f5c\u51fa\u89e3\u91ca\u548c\u660e\u786e\u9700\u6c42\u3002<\/p>\n<p>\u6211\u5f3a\u70c8\u5efa\u8bae\u6240\u6709\u5c1d\u8bd5\u5b89\u88c5Isaaclab\u7684\u7528\u6237\u5bf9\u7167\u68c0\u67e5\u81ea\u8eab\u73af\u5883\u60c5\u51b5\u3002\u9010\u70b9\u5bf9\u7167\u53c2\u8003Isaacsim\u7684\u5b98\u65b9\u9700\u6c42\u6587\u6863\uff1a<\/p>\n<pre><code>https:\/\/docs.isaacsim.omniverse.nvidia.com\/latest\/installation\/requirements.html<\/code><\/pre>\n<p><img decoding=\"async\" src=\"https:\/\/pic.cirno.fun\/sssn-blog-pics\/image-20250709221440081.png\" alt=\"image-20250709221440081\" style=\"zoom:33%;\" \/><\/p>\n<h3>\u5c1d\u8bd5\u5728AutoDL\u8fd9\u6837\u7684\u5bb9\u5668\u7b97\u529b\u5e73\u53f0\u4e0a\u642d\u5efa<\/h3>\n<p>\u5c3d\u7ba1AutoDL\u5b98\u65b9\u6ca1\u6709\u544a\u8bc9\u7528\u6237\uff08\u6587\u6863\u91cc\u4e5f\u6ca1\u6709\uff09\uff0cAutoDL\u7684\u6240\u6709\u73af\u5883\u5747\u5b58\u5728\u4e8eDocker\u4e4b\u5185\uff0c\u800c\u4e0d\u662f\u5b8c\u6574\u7684\u865a\u62df\u673a\u3002<\/p>\n<p>Isaac Sim\u4f5c\u4e3aIsaaclab\u7684\u5fc5\u8981\u57fa\u7840\uff0c\u4f9d\u8d56\u4e8e CUDA\u3001OptiX\u3001RTX \u7b49\u7ec4\u4ef6\uff0c\u5fc5\u987b\u8981\u6c42\uff1a<\/p>\n<ul>\n<li>\u5bbf\u4e3b\u673a\u652f\u6301 NVIDIA GPU \u9a71\u52a8<\/li>\n<li>Docker \u5bb9\u5668\u4f7f\u7528 <code>--gpus all<\/code> \u5e76\u6302\u8f7d <code>\/dev\/nvidia*<\/code>\u3001<code>nvidia-container-runtime<\/code>\u7b49<\/li>\n<li>\u5b89\u88c5\u7279\u5b9a\u7248\u672c\u7684 NVIDIA \u9a71\u52a8\u3001CUDA \u5de5\u5177\u5305\u7b49<\/li>\n<\/ul>\n<p>AutoDL \u5e73\u53f0\u867d\u7136\u652f\u6301 GPU\uff0c\u4f46\u4e0d\u5141\u8bb8\u63a7\u5236\u5e95\u5c42 NVIDIA \u9a71\u52a8\u548c\u5bb9\u5668\u8fd0\u884c\u65f6\u914d\u7f6e\uff0c\u4e5f\u4e0d\u66b4\u9732\u76f8\u5173\u63a5\u53e3\u3002\u5982\u679c\u4f60\u5c1d\u8bd5\u5728AutoDL\u5e73\u53f0\u4e0a\u5b89\u88c5Isaacsim\uff0c\u662f\u53ef\u4ee5\u6210\u529f\u5b89\u88c5\u7684\uff08\u65e0\u8bba\u662f\u4e8c\u8fdb\u5236\uff0c\u8fd8\u662f<code>pip install<\/code>\uff09\u3002\u4f46\u5982\u679c\u4f60\u5c1d\u8bd5\u8fd0\u884c\uff0c\u4f1a\u63d0\u793a\u7f3a\u5c11\u9a71\u52a8\u63a5\u53e3\uff0c\u6216\u8005\u7f3a\u5c11\u5173\u952e\u9a71\u52a8\u6587\u4ef6\u3002\u8fd9\u901a\u5e38\u53ef\u4ee5\u5728\u5bbf\u4e3b\u73af\u5883\u5185\u4e3aDocker\u6302\u8f7d\uff0c\u4f46AutoDL\u663e\u7136\u4e0d\u5141\u8bb8\u7528\u6237\u8bbf\u95ee\u5bbf\u4e3b\u673a\u3002<\/p>\n<blockquote>\n<p>[!NOTE]<\/p>\n<p>\u5c3d\u7ba1Isaaclab\u53ef\u4ee5\u4f7f\u7528<code>--headless<\/code>\u53c2\u6570\u53d6\u6d88\u6e32\u67d3\uff0c\u53ea\u8fd0\u884c\u4eff\u771f\u529f\u80fd\uff0c\u4f46\u8fd9\u4e2a\u53c2\u6570\u6ca1\u6709\u5b8c\u5168\u53d6\u6d88\u4ed6\u5bf9\u4e8eGPU\u63a5\u53e3\u7684\u5e95\u5c42\u4f9d\u8d56\u3002\u56e0\u6b64\uff0c\u8fd9\u4e2a\u53c2\u6570\u65e0\u6cd5\u4f7f\u5b83\u53ef\u4ee5\u8fd0\u884c\u7684\u73af\u5883\u8981\u6c42\u653e\u5bbd\u3002\u8fd9\u4e2a\u53c2\u6570\u7684\u4f18\u5316\u7a7a\u95f4\u4e5f\u8bb8\u5f88\u5927\uff1a\u76ee\u524d\u4ed3\u5e93\u91cc\u8fd8\u6709\u5f31\u4ea4\u4e92\uff0c\u975e\u4ea4\u4e92\u5f0f\u811a\u672c\u7684\u4eff\u771f\u811a\u672c\u4e0d\u80fd\u901a\u8fc7<code>--headless<\/code>\u8fd0\u884c\u3002<\/p>\n<p>[!NOTE]<\/p>\n<p>Isaaclab\u662f\u4e00\u4e2a\u6bd4\u8f83\u65b0\u7684\u4ea7\u54c1\uff0c\u793e\u533a\u4e5f\u4ecd\u5728\u6784\u5efa\u4e2d\u3002\u4e0a\u8ff0\u95ee\u9898\u53cd\u6620\u4e86Isaaclab\u5728\u63cf\u8ff0\u81ea\u8eab\u5b89\u88c5\u9700\u6c42\u65f6\u5019\u5b58\u5728\u90e8\u5206\u4e0d\u6e05\u6670\u3002\u5728\u540e\u9762\u4f60\u8fd8\u4f1a\u770b\u5230\uff0c\u6211\u9047\u5230\u4e86<strong>Isaaclab\u5b98\u65b9\u811a\u672c\u4e0d\u5b8c\u5584<\/strong>\uff0c \u548c<strong>\u5b87\u6811\u5b98\u65b9\u811a\u672c\u8fd0\u884c\u9519\u8bef<\/strong>\u7684\u95ee\u9898\u3002\u9700\u8981\u63d0\u51faIssue\uff0c\u548c\u5f00\u53d1\u8005\u76f4\u63a5\u6c9f\u901a\u3002<\/p>\n<\/blockquote>\n<h3>\u5c1d\u8bd5\u5728Nvidia Tesla T4\u6216\u8005\u7c7b\u4f3c\u7684GPU\u4e0a\u642d\u5efa<\/h3>\n<p><strong>Isaaclab\u5fc5\u987bRTX\u6280\u672f\u6765\u652f\u6301\u5e95\u5c42\u529f\u80fd<\/strong>\uff0c\u8fd9\u5728Tesla T4\u4e0a\u884c\u4e0d\u901a\u3002\u540c\u6837\u7684\uff0c<code>--headless<\/code>\u53c2\u6570\u4e0d\u80fd\u89e3\u51b3\u95ee\u9898\u3002<\/p>\n<h3>\u5c1d\u8bd5\u5728\u672c\u5730\/\u8fdc\u7a0b\u7684\u7eaf\u51c0Ubuntu\u4e0a\u642d\u5efa<\/h3>\n<p>\u51c6\u5907\u9762\u5bf9<strong>\u5728Linux\u4e0a\u5b89\u88c5Nvidia\u9a71\u52a8<\/strong>\u8fd9\u4e2a\u88ab\u55b7\u4e86\u65e0\u6570\u5e74\u7684\uff0c\u7ecf\u4e45\u4e0d\u8870\u7684\u8bdd\u9898<\/p>\n<ul>\n<li>\u5047\u9152\u9a71\u52a8\uff0c<code>nvidia-smi<\/code>\u6b63\u5e38\uff0c\u4f46\u662f\u7528\u4e0d\u4e86\uff0c\u77e5\u4e4e  <code>https:\/\/zhuanlan.zhihu.com\/p\/19148652598<\/code><\/li>\n<li>\u6362\u9a71\u52a8\u5bfc\u81f4\u66b4\u6bd9\uff0c\u77e5\u4e4e  <code>https:\/\/zhuanlan.zhihu.com\/p\/6518449814<\/code><\/li>\n<li>\u5b89\u5168\u542f\u52a8\u6bd9\u6389\u9a71\u52a8\uff0cNvidia\u5b98\u65b9\u8bba\u575b <code>https:\/\/forums.developer.nvidia.com\/t\/nvidia-drivers-not-working-while-secure-boot-is-enabled-after-updating-to-ubuntu-24-04\/305351<\/code><\/li>\n<li>Mock\u6ce8\u518c\u4e86\uff0c\u4f46\u662f\u65e0\u6548<\/li>\n<li>\u88c5\u597d\u4e86Nvidia\u9a71\u52a8\uff0c\u91cd\u542f\u4e00\u4e0b\u53d1\u73b0\u7f51\u5361\u9a71\u52a8\u6302\u4e86<\/li>\n<\/ul>\n<p>\u4f60\u53ef\u80fd\u5728\u60f3\uff1a<\/p>\n<ul>\n<li>\u82f1\u4f1f\u8fbe\u5b98\u7f51\u641c\u7d22\u663e\u5361\u578b\u53f7\uff0c\u4e0b\u8f7d<code>run.sh<\/code>\u79bb\u7ebf\u5b89\u88c5<\/li>\n<li>\u4f9d\u7167\u81ea\u52a8\u5efa\u8bae\u5b89\u88c5<code>sudo ubuntu-drivers autoinstall<\/code><\/li>\n<li>\u5728\u82f1\u4f1f\u8fbe\u5b98\u7f51\u67e5\u8868\u5b89\u88c5<code>sudo apt install nvidia-driver-525<\/code><\/li>\n<li>\u66f4\u591a\u770b\u4f3c\u89c4\u8303\uff0c\u770b\u4f3c\u5b98\u65b9\u7684\u5b89\u88c5<\/li>\n<\/ul>\n<p>\u5728\u7ecf\u5386\u957f\u4e45\u7684\u6298\u78e8\u540e\uff0c\u6211\u8ba4\u4e3a\u8fd9\u4e2a\u8bdd\u9898\u5728\u4e92\u8054\u7f51\u4e0a\u88ab\u957f\u4e45\u8ba8\u8bba\u662f\u6709\u81ea\u5df1\u7684\u539f\u56e0\u7684\uff0c\u4eba\u751f\u82e6\u77ed\uff0c\u8bf7\u627e\u4e2a\u6210\u54c1\u9a71\u52a8\u955c\u50cf\u7528\u3002<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/pic.cirno.fun\/sssn-blog-pics\/image-20250709223228630.png\" alt=\"image-20250709223228630\" style=\"zoom:33%;\" \/><\/p>\n<h3>\u5c0f\u7ed3<\/h3>\n<p>\u4ee5\u4e0a\u662f\u4e00\u4e9bIsaaclab\u5b89\u88c5\u73af\u5883\u7684\u5e38\u89c1\u5751\u70b9\uff0c\u6211\u4e2a\u4eba\u8ba4\u4e3a\u5b98\u65b9\u53ef\u4ee5\u51fa\u4e00\u4e2a\u5c0f\u5de7\u7684\u811a\u672c\uff0c\u7528\u6765\u68c0\u67e5\u5f53\u524d\u6240\u5904\u7684\u73af\u5883\u662f\u5426\u6ee1\u8db3\u5b89\u88c5\u8981\u6c42\u3002\u8ba9\u7528\u6237\u4e0d\u5fc5\u82b1\u8d39\u5927\u91cf\u65f6\u95f4\u4e0b\u8f7d\uff0c\u90e8\u7f72\uff0c\u7136\u540e\u5728\u8fd0\u884c\u8fc7\u7a0b\u4e2d\u62a5\u9519\u5d29\u6e83\u3002\u6216\u8005\u6700\u5c11\u7684\uff0c\u4e5f\u5e94\u8be5\u5728\u5b98\u65b9\u6587\u6863\u4e2d\uff0cIsaacsim\u7684\u5b89\u88c5\u547d\u4ee4\u4e0b\u6ce8\u91ca\u9700\u6c42\u6587\u6863\u5730\u5740\u3002<\/p>\n<h2>\u8f6f\u4ef6\u5b89\u88c5<\/h2>\n<h3>\u793a\u4f8b\u914d\u7f6e<\/h3>\n<p>\u4ece\u8fd9\u91cc\u5f00\u59cb\uff0c\u6211\u5f00\u59cb\u4f7f\u7528\u963f\u91cc\u4e91ECS\u4f5c\u4e3a\u8bd5\u9a8c\u73af\u5883\u3002\u5b9e\u4f8b\u89c4\u683c<code>ecs.gn7i-c32g1.8xlarge<\/code>\uff1a<\/p>\n<ul>\n<li>GPU\uff1aNvidia A10<\/li>\n<li>CPU\uff1a32vCPU Intel(R) Xeon(R) Platinum 8369B CPU @ 2.90GHz<\/li>\n<li>\u5185\u5b58\uff1a32G\uff0c\u5b9e\u9645\u4e0a\u8fc7\u5269\u4e86<\/li>\n<li>\u955c\u50cf\uff1a\u4e91\u5e02\u573a\u955c\u50cf-&gt;\u641c\u7d22\u9a71\u52a8+Ubuntu-&gt;\u9009\u62e9v3<\/li>\n<\/ul>\n<p><img decoding=\"async\" src=\"https:\/\/pic.cirno.fun\/sssn-blog-pics\/image-20250709224304681.png\" alt=\"image-20250709224304681\" style=\"zoom:33%;\" \/><\/p>\n<blockquote>\n<p>[!NOTE]<\/p>\n<p>\u963f\u91cc\u4e91ECS\u80fd\u8dd1\u7684\u672c\u8d28\u539f\u56e0\u662f\u4ed6\u662f\u865a\u62df\u673a\uff0c\u800c\u4e0d\u662fDocker\u5bb9\u5668\u3002<\/p>\n<p>[!CAUTION]<\/p>\n<p>\u4e0d\u80fd\u89c9\u5f97\u963f\u91cc\u9a71\u52a8\u5e02\u573a\u7684\u955c\u50cf\u4e00\u5b9awork\uff0c\u5f00\u673a\u6302\u6389\u7684\u6bd4\u6bd4\u7686\u662f\uff0c\u4f46\u6211\u6311\u7684\u8fd9\u4e2a\u9a71\u52a8\u548c\u5bb9\u5668\u89c4\u683c\u80af\u5b9a\u6ca1\u95ee\u9898<\/p>\n<\/blockquote>\n<h3>\u5b89\u88c5\u6d41\u7a0b<\/h3>\n<h4>\u5b89\u88c5Isaacsim\u548cIssaclab<\/h4>\n<p>\u9605\u8bfb\u5230\u8fd9\u91cc\uff0c\u4f60\u5e94\u8be5\u5df2\u7ecf\u53ef\u7528\u7684\u8fd0\u884c\u73af\u5883\uff0c\u8fd9\u4e2a\u73af\u5883<\/p>\n<ul>\n<li>GPU\u652f\u6301RTX\uff0cCUDA\u7248\u672c\u8f83\u65b0<\/li>\n<li>Nvidia\u9a71\u52a8\u6b63\u5e38\uff0c\u4e0d\u4ec5\u53ef\u4ee5\u663e\u793a\u7248\u672c\uff0c\u8fd8\u80fd\u5f80GPU\u8fc1\u79fb\u6570\u636e\uff08\u53ef\u4ee5\u53c2\u89c1\u4e0a\u9762\u7684\u201c\u5047\u9a71\u52a8\u201d\u5e16\u5b50\uff09<\/li>\n<li>\u7ea7\u522b\u4e3a\u865a\u62df\u673a\uff0c\u6216\u8005\u7269\u7406\u673a<\/li>\n<\/ul>\n<p>\u73b0\u5728\u4f60\u53ef\u4ee5\u6309\u7167Isaaclab\u5b98\u65b9\u6587\u6863\u7684\u6d41\u7a0b\u6765\u5b89\u88c5\u8f6f\u4ef6\uff0c\u5e76\u5c1d\u8bd5\u8fd0\u884c<\/p>\n<ol>\n<li>\u5b89\u88c5Miniconda<\/li>\n<li>\u521b\u5efa\u4e00\u4e2apython 3.10\u73af\u5883<\/li>\n<li>\u5b89\u88c5Isaacsim\n<ul>\n<li>\u5982\u679c\u4f60\u662fUbuntu 22.04\uff0c\u4f60\u5e94\u8be5\u7528<code>pip<\/code>\u5b89\u88c5<\/li>\n<li>\u5982\u679c\u4f60\u662fUbuntu 20.04\uff0c\u4f60\u5e94\u8be5\u7528\u4e8c\u8fdb\u5236\u5b89\u88c5<\/li>\n<\/ul>\n<\/li>\n<li>\u5b89\u88c5Isaaclab\n<ol>\n<li><code>git clone<\/code>\u5b98\u65b9\u4ed3\u5e93<\/li>\n<li><code>conda activate [\u4f60\u7684\u73af\u5883]<\/code><\/li>\n<li><code>chmod +x Issaclab.sh &amp;&amp; Issaclab.sh --install<\/code>  <\/li>\n<\/ol>\n<\/li>\n<\/ol>\n<p>\u53ef\u4ee5\u53c2\u7167\u5b98\u65b9\u6587\u6863\uff0c\u6587\u6863\u5df2\u7ecf\u6709\u7684\u6d41\u7a0b\u5c31\u4e0d\u8d58\u8ff0\u4e86<\/p>\n<pre><code>https:\/\/isaac-sim.github.io\/IsaacLab\/main\/source\/setup\/installation\/index.html<\/code><\/pre>\n<h4>\u66f4\u591a\u914d\u7f6e<\/h4>\n<p>\u4e3a\u4e86\u66f4\u65b9\u4fbf\u7684\u89c2\u6d4b\u8bd5\u9a8c\u8fd0\u884c\u60c5\u51b5\uff0c\u4f60\u6700\u597d\u5b89\u88c5\u8fdc\u7a0b\u684c\u9762\u8f6f\u4ef6\uff0c\u5e76\u914d\u7f6e\u7f51\u7edc\u4ee3\u7406\u3002\u6211\u5728\u8fd9\u91cc\u7ed9\u51fa\u4e00\u4e2a\u53ef\u884c\u7684\u89e3\u51b3\u65b9\u6848\uff1a<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/pic.cirno.fun\/sssn-blog-pics\/image-20250709230952856.png\" alt=\"image-20250709230952856\" style=\"zoom:33%;\" \/><\/p>\n<blockquote>\n<p>[!CAUTION]<\/p>\n<p>Isaaclab\u7684\u5f88\u591a\u811a\u672c\u4f1a\u8c03\u7528\u7f51\u7edc\u8d44\u6e90\uff0c\u8fd9\u4e9b\u8d44\u6e90\u4f4d\u4e8e<code>http:\/\/omniverse-content-production.s3-us-west-2.amazonaws.com\/Assets\/Isaac\/<\/code>\uff0c\u7f51\u7edc\u4ee3\u7406\u662f\u5fc5\u987b\u7684<\/p>\n<\/blockquote>\n<p>\u4f7f\u7528SSH\u96a7\u9053\u642d\u5efa\u4e24\u4e2a\u6838\u5fc3\u670d\u52a1\uff1a\u7f51\u7edc\u4ee3\u7406\u548cVNC\u684c\u9762\u7684\u901a\u9053\uff0c\u7528<code>curl<\/code>\u7b80\u8981\u6d4b\u8bd5\u4ee3\u7406\u8fd0\u884c\u60c5\u51b5<\/p>\n<pre><code class=\"language-shell\">(base) ecs-user@iZuf6g6vii5ci3ikxssy9lZ:~$ source ~\/proxy-on.sh 7899\n127.0.0.1:7899\n(base) ecs-user@iZuf6g6vii5ci3ikxssy9lZ:~$ curl google.com\n&lt;HTML&gt;&lt;HEAD&gt;&lt;meta http-equiv=&quot;content-type&quot; content=&quot;text\/html;charset=utf-8&quot;&gt;\n&lt;TITLE&gt;301 Moved&lt;\/TITLE&gt;&lt;\/HEAD&gt;&lt;BODY&gt;\n&lt;H1&gt;301 Moved&lt;\/H1&gt;\nThe document has moved\n&lt;A HREF=&quot;http:\/\/www.google.com\/&quot;&gt;here&lt;\/A&gt;.\n&lt;\/BODY&gt;&lt;\/HTML&gt;<\/code><\/pre>\n<h1>Task1: \u57fa\u7840\u73af\u5883\u642d\u5efa<\/h1>\n<p>\u5230\u8fd9\u4e00\u6b65\uff0c\u4f60\u5e94\u8be5\u5df2\u7ecf\u5b89\u88c5\u5e76\u6d4b\u8bd5\u597d\u4e86Issaclab<\/p>\n<h2>\u8981\u6c42<\/h2>\n<ol>\n<li>\u573a\u666f\u6784\u5efa\uff1a<\/li>\n<\/ol>\n<ul>\n<li>\n<p>\u4f7f\u7528Isaac Sim\u521b\u5efa10m\u00d710m\u7684\u5ba4\u5185\u5e73\u9762\u573a\u666f\uff08\u53ef\u9009\uff1a\u589e\u52a0\u969c\u788d\u7269\/\u659c\u5761\uff09<\/p>\n<\/li>\n<li>\n<p>\u6dfb\u52a01\u4e2a\u4eba\u5f62\u673a\u5668\u4eba\uff08\u4f8b\u5982\u5b87\u6811H1\/Go2\u7b49\uff09<\/p>\n<\/li>\n<li>\n<p>\u90e8\u7f72\u57fa\u7840\u5149\u6e90\u548c\u7269\u7406\u6750\u8d28<\/p>\n<\/li>\n<\/ul>\n<ol start=\"2\">\n<li>\u8fd0\u52a8\u63a7\u5236\u6d4b\u8bd5\u200b<\/li>\n<\/ol>\n<ul>\n<li>\n<p>\u901a\u8fc7Isaac Lab\u7684Python API\u63a7\u5236\u673a\u5668\u4eba\u5b8c\u6210\u57fa\u7840\u52a8\u4f5c<\/p>\n<\/li>\n<li>\n<p>\u8f93\u51fa\u673a\u5668\u4eba\u5b8c\u6210\u7ad9\u7acb\u52a8\u4f5c\u76845\u79d2\u5173\u8282\u89d2\u5ea6\u53d8\u5316\u66f2\u7ebf\u56fe<\/p>\n<\/li>\n<\/ul>\n<h2>\u5b8c\u6210\u60c5\u51b5<\/h2>\n<ul>\n<li><strong>\u5168\u90e8\u5b8c\u6210\uff0c\u6e90\u7801\u8bf7\u53c2\u8003\u4ed3\u5e93<\/strong>\n<ul>\n<li>\u57fa\u7840\u73af\u5883\u5185\u542b\u51e0\u4e2a\u57fa\u672c\u51e0\u4f55\u7269\u4f53\uff0c\u673a\u5668\u4eba\u4f7f\u7528G1\uff0c\u6709\u9ed8\u8ba4\u5149\u6e90\u548c\u6750\u8d28<\/li>\n<li>\u5f00\u59cb\u4eff\u771f\u540e\uff0c\u673a\u5668\u4eba\u5355\u811a\u72ec\u7acb\uff0c\u5168\u8eab\u6267\u884c\u5c0f\u5e45\u5ea6\u7684\u968f\u673a\u52a8\u4f5c\uff0c\u201c\u62bd\u6410\u201d\u7740\u5012\u4e0b<\/li>\n<li>\u8bb0\u5f55\u91cd\u8981\u5173\u8282\u7684\u5f27\u5ea6\u53d8\u5316\u5e76\u6253\u5370\u6210\u6298\u7ebf\u56fe<\/li>\n<\/ul>\n<\/li>\n<li>\u6269\u5c55\u5185\u5bb9\uff1a\u6f14\u793a\u89c6\u9891\u7684\u5f55\u5236\n<ul>\n<li>\u5728\u4e00\u4e2a\u5355\u72ec\u7684<code>.py<\/code>\u811a\u672c\u4e2d\u5f55\u5236\u89c6\u9891\u7adf\u7136\u5982\u6b64\u7684\u96be\uff0c<strong>\u8c03\u8bd5\u96be\u5ea6\u53ef\u4ee5\u6392\u6574\u4e2a\u9879\u76ee\u7684\u524d\u4e09<\/strong><\/li>\n<li>\u5927\u6a21\u578b\u7684\u8bb0\u5fc6\u90fd\u662f\u8001api\uff0c\u8c03\u6559\u4e86\u4e00\u6574\u5929\u90fd\u65e0\u6cd5\u4ece\u9ed8\u8ba4\u89c6\u89d2\u4e2d\u6293\u53d6\u51fa\u7167\u7247<\/li>\n<li>\u5b98\u65b9\u6587\u6863\u4e2d\u7684<code>beta: StageRecorder[1]<\/code>\u7528\u4e0d\u4e86\uff0c\u672a\u77e5\u539f\u56e0\uff0c\u5f55\u5236\u51fa\u7684\u6587\u4ef6\u662f\u7a7a\u7684<\/li>\n<li>\u5b98\u65b9\u6587\u6863\u4e2d\u7684<code>ReplicatorBasicWriter[2]<\/code>\u53ef\u4ee5\u7528\uff0c\u85cf\u7684\u6709\u70b9\u6df1\u4e86\uff0c\u627e\u4e86\u5f88\u4e45\u5f88\u4e45<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<pre><code>[1] https:\/\/isaac-sim.github.io\/IsaacLab\/main\/source\/how-to\/record_animation.html\n[2] https:\/\/isaac-sim.github.io\/IsaacLab\/main\/source\/how-to\/save_camera_output.html<\/code><\/pre>\n<p><img decoding=\"async\" src=\"https:\/\/pic.cirno.fun\/sssn-blog-pics\/image-20250710000558629.png\" alt=\"image-20250710000558629\" style=\"zoom:50%;\" \/><\/p>\n<h2>\u5173\u952e\u4ee3\u7801(\u968f\u673a\u52a8\u4f5c\u90e8\u5206)<\/h2>\n<pre><code class=\"language-python\">for i, name in enumerate(joint_names):\n    set = False\n    # \u8df3\u8fc7\u53f3\u817f\u5173\u8282\uff08pitch\u65b9\u5411\uff09\u4fdd\u6301\u8e22\u817f\u52a8\u4f5c\n    if any(key in name for key in [&quot;right_hip_pitch_joint&quot;, &quot;right_knee_joint&quot;, &quot;right_ankle_pitch_joint&quot;]):\n        continue  # \u4e0d\u4fee\u6539\u8e22\u817f\u59ff\u6001\n    # \u5339\u914d\u4e0a\u80a2\u5173\u8282\n    for key, (low, high) in upper_body_joint_ranges.items():\n        if key in name:\n            joint_pos[:, i] = torch.rand(1, device=sim.device) * (high - low) + low\n            set = True\n            break\n    # \u5339\u914d\u4e0b\u80a2\u5173\u8282\n    if not set:\n        for key, (low, high) in lower_body_joint_ranges.items():\n            if key in name:\n                joint_pos[:, i] = torch.rand(1, device=sim.device) * (high - low) + low\n                set = True\n                break\n    # \u5982\u679c\u6ca1\u6709\u5339\u914d\u5230\u4efb\u4f55\u8bbe\u7f6e\uff0c\u5219\u4f7f\u7528\u9ed8\u8ba4\u5c0f\u5e45\u6270\u52a8\n    if not set:\n        joint_pos[:, i] = torch.rand(1, device=sim.device) * 0.2 - 0.1  # [-0.1, 0.1]<\/code><\/pre>\n<h1>Task2: \u590d\u73b0\u5b87\u6811\u673a\u5668\u4eba\u884c\u8d70<\/h1>\n<h2>\u8981\u6c42<\/h2>\n<p>\u76ee\u6807\uff1a\u57fa\u4e8e\u5b87\u6811\u5f00\u6e90\u4ee3\u7801\u5b9e\u73b0H1\u673a\u5668\u4eba\u884c\u8d70\u7b56\u7565\u590d\u73b0<\/p>\n<ol>\n<li>\u4ee3\u7801\u90e8\u7f72\u4e0e\u8bad\u7ec3\u200b<\/li>\n<\/ol>\n<ul>\n<li>\n<p>\u4f7f\u7528\u5b87\u6811\u5b98\u65b9\u63d0\u4f9b\u7684\u5f3a\u5316\u5b66\u4e60\u793a\u4f8b<\/p>\n<\/li>\n<li>\n<p>\u89e3\u6790\u5956\u52b1\u51fd\u6570\u8bbe\u8ba1\uff1a\u901f\u5ea6\u8ddf\u8e2a\u5956\u52b1\uff1b\u6b65\u6001\u5bf9\u79f0\u6027\u60e9\u7f5a\u7b49<\/p>\n<\/li>\n<\/ul>\n<h2>\u5b8c\u6210\u60c5\u51b5<\/h2>\n<ul>\n<li><strong>\u5168\u90e8\u5b8c\u6210<\/strong>\uff0c\u5176\u4e2d\u4e3b\u8981\u590d\u73b0\u4e24\u90e8\u5206\u5b9e\u9a8c\n<ul>\n<li>\u5b87\u6811\u5b98\u65b9\u4ed3\u5e93\uff1a<code>https:\/\/github.com\/unitreerobotics\/unitree_rl_lab<\/code><\/li>\n<li>Isaaclab\u7684\u65b0\u7248\u672c\u4e5f\u5305\u542b\u5b87\u6811\u673a\u5668\u4eba\u884c\u8d70\uff1a<code>https:\/\/isaac-sim.github.io\/IsaacLab\/main\/source\/overview\/environments.html#locomotion<\/code><\/li>\n<\/ul>\n<\/li>\n<li>\u6269\u5c55\u5185\u5bb9\uff1a\u53d1\u73b0\u4e86\u4e00\u4e2a\u5b87\u6811\u5b98\u65b9\u4ee3\u7801\u7684bug\uff0c\u63d0\u4ea4issue\u5e76\u89e3\u51b3<\/li>\n<li>\u6269\u5c55\u5185\u5bb9\uff1a\u53d1\u73b0\u4e86\u4e00\u4e2aIssaclab\u5b98\u65b9\u4ee3\u7801\u7684bug\uff0c\u63d0\u4ea4issue\u5e76\u89e3\u51b3<\/li>\n<\/ul>\n<p><img decoding=\"async\" src=\"https:\/\/pic.cirno.fun\/sssn-blog-pics\/image-20250710013622535.png\" alt=\"image-20250710013622535\" style=\"zoom:33%;\" \/><\/p>\n<h3>\u62d3\u5c55\u5185\u5bb9<\/h3>\n<blockquote>\n<p>[!CAUTION]<\/p>\n<p>\u5b87\u6811\u5b98\u65b9\u4ed3\u5e93\u7684\u4ee3\u7801\u5b58\u5728\u7684\u8bad\u7ec3\u6570\u503c\u5d29\u6e83\u7684\u95ee\u9898<\/p>\n<\/blockquote>\n<pre><code class=\"language-shell\">...\n  File &quot;\/home\/ecs-user\/miniconda3\/envs\/is\/lib\/python3.10\/site-packages\/torch\/distributions\/normal.py&quot;, line 73, in sample\n    return torch.normal(self.loc.expand(shape), self.scale.expand(shape))\nRuntimeError: normal expects all elements of std &gt;= 0.0<\/code><\/pre>\n<p>\u8fd9\u662f\u56e0\u4e3a\u5728 PyTorch \u4e2d\uff0c<code>torch.distributions.Normal(loc, scale)<\/code> \u662f\u4e00\u4e2a\u6b63\u6001\u5206\u5e03\uff0c\u5176 <code>scale<\/code>\uff08\u6807\u51c6\u5dee\uff09\u5fc5\u987b\u662f \u975e\u8d1f\u6570\u3002<\/p>\n<p>\u4f46\u662f\u6a21\u578b\u4e2d <code>scale<\/code> \u51fa\u73b0\u4e86\u8d1f\u503c\u6216 NaN\uff0c\u5bfc\u81f4 <code>torch.normal(...)<\/code> \u629b\u51fa\u5f02\u5e38\u3002<\/p>\n<p>\u8fd9\u53ef\u80fd\u548c\u68af\u5ea6\u7206\u70b8\uff0c\u6570\u503c\u4e0d\u7a33\u5b9a\uff0c\u6ca1\u6709\u526a\u88c1\u68af\u5ea6\u6709\u5173\u3002<\/p>\n<p>\u6211\u63d0\u4ea4\u4e86<code>issue[1]<\/code> \uff0c\u6536\u5230\u56de\u590d\u8bf4\u95ee\u9898\u5728\u6700\u8fd1\u4e00\u6b21\u7684commit\u4fee\u590d\u4e86\u3002<\/p>\n<ul>\n<li>\u6211\u57287-7\u514b\u9686\u7684\u4ee3\u7801\uff0c\u57287-9\u9047\u5230\u7684\u95ee\u9898<\/li>\n<li>\u4fee\u590d\u7684commit\u65e5\u671f\u662f7-8\uff0c\u53ef\u4ee5\u8bf4\u662f\u975e\u5e38\u5de7\u4e86<\/li>\n<\/ul>\n<pre><code>[1] https:\/\/github.com\/unitreerobotics\/unitree_rl_lab\/issues\/7<\/code><\/pre>\n<blockquote>\n<p>[!CAUTION]<\/p>\n<p>Isaaclab\u5b98\u65b9\u4ee3\u7801\u4e2d\uff0c\u6559\u7a0b\u811a\u672c\u6709\u70b9\u5c0f\u6bdb\u75c5<\/p>\n<\/blockquote>\n<pre><code>(is) ecs-user@iZuf6g6vii5ci3ikxssy9lZ:~\/IsaacLab$ .\/isaaclab.sh -p scripts\/demos\/h1_locomotion.py --headless\n...\nTraceback (most recent call last):\n  File &quot;\/home\/ecs-user\/IsaacLab\/scripts\/demos\/h1_locomotion.py&quot;, line 50, in &lt;module&gt;\n    from omni.kit.viewport.utility import get_viewport_from_window_name\nModuleNotFoundError: No module named &#039;omni.kit.viewport&#039;<\/code><\/pre>\n<p>\u9047\u5230\u95ee\u9898\u7684\u65f6\u5019\u6ca1\u4ec0\u4e48\u60f3\u6cd5\uff0c\u63d0\u4ea4\u4e86<code>Issue[1]<\/code><\/p>\n<ul>\n<li>\u6536\u5230\u56de\u590d\u8bf4\u8fd9\u4e2a\u6559\u7a0b\u811a\u672c\u662f\u4ea4\u4e92\u5f0f\u7684\uff0c\u4e0d\u80fd\u6dfb\u52a0<code>--headless<\/code>\u53c2\u6570<\/li>\n<li>\u4f46\u662f\u6240\u6709\u540c\u4e00\u76ee\u5f55\u4e0b\u7684\u811a\u672c\u90fd\u53ef\u4ee5<code>--headless<\/code>\uff0c\u5f00\u53d1\u8005\u8bf4\u4f1a\u6dfb\u52a0\u4e00\u4e2a\u62a5\u9519\uff0c\u63d0\u9192\u8fd9\u4e2a\u811a\u672c\u4e0d\u80fd\u52a0<\/li>\n<\/ul>\n<pre><code class=\"language-shell\">[1] https:\/\/github.com\/isaac-sim\/IsaacLab\/issues\/2858<\/code><\/pre>\n<h3>\u5956\u52b1\u51fd\u6570\u8bbe\u8ba1<\/h3>\n<p>\u5956\u52b1\u51fd\u6570\u7684\u4ee3\u7801\u4f4d\u4e8e\uff1a<\/p>\n<pre><code class=\"language-shell\">source\/unitree_rl_lab\/unitree_rl_lab\/tasks\/locomotion\/robots\/h1\/velocity_env_cfg.py\n238-338\u884c<\/code><\/pre>\n<p>\u6bcf\u4e00\u9879\u5956\u52b1\u90fd\u662f\u4e00\u4e2a <code>RewTerm<\/code> \u5b9e\u4f8b\uff0c\u5b83\u5305\u542b\uff1a<\/p>\n<ul>\n<li><code>func<\/code>: \u5956\u52b1\u51fd\u6570\uff0c\u5b9e\u73b0\u5177\u4f53\u7684\u884c\u4e3a\u8bc4\u4f30\u903b\u8f91<\/li>\n<li><code>weight<\/code>: \u6743\u91cd\uff0c\u8868\u793a\u8be5\u9879\u5956\u52b1\u5bf9\u603b\u5956\u52b1\u7684\u5f71\u54cd\u7a0b\u5ea6\uff08\u6b63\u503c\u9f13\u52b1\u3001\u8d1f\u503c\u60e9\u7f5a\uff09<\/li>\n<li><code>params<\/code>: \u53ef\u9009\u53c2\u6570\uff0c\u7528\u4e8e\u914d\u7f6e\u5956\u52b1\u51fd\u6570\u7684\u884c\u4e3a<\/li>\n<\/ul>\n<p>\u603b\u5171\u6709\u8fd9\u4e9b\u5b9e\u4f8b\uff1a<\/p>\n<h4>\u4efb\u52a1\u76f8\u5173\u7684\u5956\u52b1\uff1a\u5956\u52b1\u5b8c\u6210\u4efb\u52a1\uff08\u8ddf\u968f\u76ee\u6807\u901f\u5ea6\uff09<\/h4>\n<h5>track_lin_vel_xy<\/h5>\n<ul>\n<li>\u76ee\u7684\uff1a\u9f13\u52b1\u673a\u5668\u4eba\u5728XY\u5e73\u9762\u4e0a\u8ffd\u8e2a\u76ee\u6807\u7ebf\u901f\u5ea6\u3002<\/li>\n<li>\u6743\u91cd\uff1a1.0\uff08\u8f83\u9ad8\uff0c\u8868\u660e\u8fd9\u662f\u6838\u5fc3\u4efb\u52a1\uff09<\/li>\n<\/ul>\n<h5>track_ang_vel_z<\/h5>\n<ul>\n<li>\u76ee\u7684\uff1a\u9f13\u52b1\u673a\u5668\u4eba\u8ddf\u8e2a\u76ee\u6807\u7684\u89d2\u901f\u5ea6\uff08\u7ed5Z\u8f74\u65cb\u8f6c\uff09\u3002<\/li>\n<li>\u6743\u91cd\uff1a0.5\uff08\u6b21\u91cd\u8981\uff09<\/li>\n<\/ul>\n<h5>alive<\/h5>\n<ul>\n<li>\u76ee\u7684\uff1a\u9f13\u52b1\u673a\u5668\u4eba\u4fdd\u6301\u201c\u5b58\u6d3b\u201d\uff08\u4f8b\u5982\u4e0d\u6454\u5012\u3001\u7ef4\u6301\u5e73\u8861\uff09\u3002<\/li>\n<li>\u6743\u91cd\uff1a0.15\uff08\u7ef4\u6301\u57fa\u672c\u7a33\u5b9a\u6027\uff09<\/li>\n<\/ul>\n<h4>\u57fa\u7840\u8fd0\u52a8\u7684\u5956\u52b1\uff08\u8fd0\u52a8\u72b6\u6001\u76f8\u5173\uff09<\/h4>\n<h5>base_linear_velocity<\/h5>\n<ul>\n<li>\u76ee\u7684\uff1a\u60e9\u7f5a\u5782\u76f4\u65b9\u5411\u4e0a\u7684\u901f\u5ea6\uff08Z\u8f74\uff09\uff0c\u9f13\u52b1\u7a33\u5b9a\u3002<\/li>\n<li>\u6743\u91cd\uff1a-2.0\uff08\u60e9\u7f5a\u529b\u5ea6\u8f83\u5927\uff09<\/li>\n<\/ul>\n<h5>base_angular_velocity<\/h5>\n<ul>\n<li>\u76ee\u7684\uff1a\u60e9\u7f5a\u7ed5XY\u8f74\u7684\u89d2\u901f\u5ea6\uff0c\u907f\u514d\u673a\u5668\u4eba\u5267\u70c8\u6643\u52a8\u3002<\/li>\n<li>\u6743\u91cd\uff1a-0.5<\/li>\n<\/ul>\n<h5>joint_acc<\/h5>\n<ul>\n<li>\u76ee\u7684\uff1a\u60e9\u7f5a\u5173\u8282\u52a0\u901f\u5ea6\uff0c\u9f13\u52b1\u5e73\u6ed1\u8fd0\u52a8\u3002<\/li>\n<li>\u6743\u91cd\uff1a-2.5e-7\uff08\u975e\u5e38\u5c0f\uff0c\u6b21\u8981\u60e9\u7f5a\uff09<\/li>\n<\/ul>\n<h5>action_rate<\/h5>\n<ul>\n<li>\u76ee\u7684\uff1a\u60e9\u7f5a\u52a8\u4f5c\u53d8\u5316\u7387\uff0c\u9f13\u52b1\u8fde\u7eed\u5e73\u6ed1\u7684\u52a8\u4f5c\u3002<\/li>\n<li>\u6743\u91cd\uff1a-0.05<\/li>\n<\/ul>\n<h5>dof_pos_limits<\/h5>\n<ul>\n<li>\u76ee\u7684\uff1a\u60e9\u7f5a\u5173\u8282\u4f4d\u7f6e\u8d85\u51fa\u9650\u5236\u3002<\/li>\n<li>\u6743\u91cd\uff1a-5.0\uff08\u8f83\u9ad8\u60e9\u7f5a\uff09<\/li>\n<\/ul>\n<h4>\u5173\u8282\u504f\u5dee\u5956\u52b1\uff08\u60e9\u7f5a\u4e0d\u5408\u7406\u7684\u5173\u8282\uff09<\/h4>\n<h5>joint_deviation_arms<\/h5>\n<ul>\n<li>\u76ee\u6807\uff1a\u80a9\u8180\u3001\u8098\u90e8\u7b49\u624b\u81c2\u5173\u8282<\/li>\n<li>\u6743\u91cd\uff1a-1.0<\/li>\n<li>\u6b63\u5219\u8868\u8fbe\u5f0f\uff1a\u5339\u914d\u80a9\u3001\u8098\u5173\u8282\u7684\u540d\u5b57<\/li>\n<\/ul>\n<h5>joint_deviation_torso<\/h5>\n<ul>\n<li>\u76ee\u6807\uff1a\u8eaf\u5e72\u5173\u8282<\/li>\n<li>\u6743\u91cd\uff1a-1.0<\/li>\n<\/ul>\n<h5>joint_deviation_hips<\/h5>\n<ul>\n<li>\u76ee\u6807\uff1a\u81c0\u90e8\u8fde\u63a5\u7684\u9acb\u5173\u8282<\/li>\n<li>\u6743\u91cd\uff1a-1.0<\/li>\n<\/ul>\n<h4>\u673a\u5668\u4eba\u59ff\u6001\u5956\u52b1\uff1a\u9f13\u52b1\u76f4\u7acb\uff0c\u60e9\u7f5a\u6c34\u5e73<\/h4>\n<h5>flat_orientation_l2<\/h5>\n<ul>\n<li>\u76ee\u7684\uff1a\u60e9\u7f5a\u673a\u5668\u4eba\u59ff\u6001\u504f\u79bb\u6c34\u5e73\uff08\u4f8b\u5982\u503e\u659c\uff09<\/li>\n<li>\u6743\u91cd\uff1a-1.0<\/li>\n<\/ul>\n<h5>base_height<\/h5>\n<ul>\n<li>\u76ee\u7684\uff1a\u9f13\u52b1\u673a\u5668\u4eba\u7ef4\u6301\u4e00\u4e2a\u76ee\u6807\u9ad8\u5ea6\uff080.9\u7c73\uff09<\/li>\n<li>\u6743\u91cd\uff1a-10.0\uff08\u975e\u5e38\u91cd\u8981\uff09<\/li>\n<\/ul>\n<h4>\u8db3\u90e8\u5956\u52b1<\/h4>\n<h5>gait<\/h5>\n<ul>\n<li>\u76ee\u7684\uff1a\u9f13\u52b1\u673a\u5668\u4eba\u5f62\u6210\u81ea\u7136\u6b65\u6001\uff08\u5982\u5bf9\u79f0\u6b65\u9891\uff09<\/li>\n<li>\u6743\u91cd\uff1a0.5<\/li>\n<\/ul>\n<h5>feet_slide<\/h5>\n<ul>\n<li>\u76ee\u7684\uff1a\u60e9\u7f5a\u811a\u63a5\u89e6\u5730\u9762\u65f6\u6ed1\u52a8\uff0c\u9f13\u52b1\u7a33\u5b9a\u7740\u5730<\/li>\n<li>\u6743\u91cd\uff1a-0.2<\/li>\n<\/ul>\n<h5>feet_clearance<\/h5>\n<ul>\n<li>\u76ee\u7684\uff1a\u9f13\u52b1\u62ac\u811a\u65f6\u6709\u4e00\u5b9a\u9ad8\u5ea6\uff0c\u9632\u6b62\u7eca\u5012<\/li>\n<li>\u6743\u91cd\uff1a20.0\uff08\u975e\u5e38\u9ad8\uff0c\u91cd\u8981\u6027\u5927\uff09<\/li>\n<\/ul>\n<h5>feet_contact_forces<\/h5>\n<ul>\n<li>\u76ee\u7684\uff1a\u60e9\u7f5a\u5f02\u5e38\u7684\u5927\u811a\u90e8\u63a5\u89e6\u529b\uff08\u53ef\u80fd\u8868\u793a\u649e\u51fb\uff09<\/li>\n<li>\u6743\u91cd\uff1a-0.0002\uff08\u60e9\u7f5a\u8f7b\u5fae\uff09<\/li>\n<\/ul>\n<h4>\u5176\u4ed6<\/h4>\n<h5>undesired_contacts<\/h5>\n<ul>\n<li>\u76ee\u7684\uff1a\u60e9\u7f5a\u9664\u811a\u4ee5\u5916\u7684\u8eab\u4f53\u4e0e\u5730\u9762\u63a5\u89e6\uff08\u4f8b\u5982\u624b\u3001\u819d\u76d6\u78b0\u5730\uff09\u3002<\/li>\n<li>\u6743\u91cd\uff1a-1.0\uff08\u9632\u6b62\u6454\u5012\u7b49\uff09<\/li>\n<\/ul>\n<h1>Task3\uff1a\u8de8\u5e73\u53f0\u6b65\u6001\u8fc1\u79fb<\/h1>\n<h2>\u8981\u6c42<\/h2>\n<p>\u76ee\u6807\uff1a\u5c06\u8bad\u7ec3\u7b56\u7565\u8fc1\u79fb\u81f3\u5176\u4ed6\u4eba\u5f62\u673a\u5668\u4eba<\/p>\n<p>\u6a21\u578b\u9002\u914d<\/p>\n<ul>\n<li>\n<p>\u5bfc\u5165\u7b2c\u4e09\u65b9\u673a\u5668\u4ebaURDF\u6587\u4ef6\uff08\u9700\u5bf9\u9f50\u5173\u8282\u547d\u540d\u89c4\u8303\uff0c\u5982.*_hip_pitch\uff09<\/p>\n<\/li>\n<li>\n<p>\u8c03\u6574\u52a8\u529b\u5b66\u53c2\u6570\uff1a\u4f8b\u5982\u5173\u8282\u529b\u77e9\u9650\u5236\uff08\u7b2c\u4e09\u65b9\u673a\u5668\u4eba\u7535\u673a\u626d\u77e9\u53ef\u80fd\u4f4e\u4e8e\u5b87\u6811H1\uff09\/\u817f\u90e8\u8d28\u91cf\u5206\u5e03<\/p>\n<\/li>\n<\/ul>\n<h2>\u5b8c\u6210\u60c5\u51b5<\/h2>\n<ul>\n<li><strong>\u5168\u90e8\u5b8c\u6210<\/strong><\/li>\n<li>\u5c06\u4e0a\u6587\u4e2d\u8bad\u7ec3\u597d\u7684\u5b87\u6811H1\u884c\u8d70\u7b56\u7565\u8fc1\u79fb\u81f3\u5b87\u6811G1\u4e0a\n<ol>\n<li>\u4f7f\u7528\u4ee3\u7801\u63a2\u9488\u5206\u6790H1\u548cG1\u7684\u5173\u8282\uff0c\u5bf9\u7167\u5206\u6790.usd\u6587\u4ef6<\/li>\n<li>\u5bf9\u4e8e\u8f93\u51fa\uff08\u52a8\u4f5c\uff09\uff0c\u5efa\u7acbH1\u5230G1\u7684\u5173\u8282\u6620\u5c04\u8868<\/li>\n<li>\u5bf9\u4e8e\u8f93\u5165\uff08\u89c2\u6d4b\uff09\uff0c\u91cd\u6392G1\u7684\u5173\u8282\u987a\u5e8f\uff0c\u5bf9\u9f50H1\u7b56\u7565\u7684\u8f93\u5165\u7ef4\u5ea6\u548c\u987a\u5e8f<\/li>\n<li>\u4fee\u6539G1\u7684\u8bbe\u7f6e\uff08\u521d\u59cb\u7ad9\u59ff\uff0c\u529b\u77e9\uff09\u9002\u914dH1\u7b56\u7565<\/li>\n<li>\u6d41\u7a0b\uff1aG1\u89c2\u6d4b-&gt;\u8f6c\u6362\u4e3aH1\u89c2\u6d4b-&gt;\u8f93\u5165H1\u7b56\u7565-&gt;\u5f97\u5230H1\u52a8\u4f5c\uff08\u5173\u8282\u52a8\u4f5c\u8868\uff09-&gt;\u6620\u5c04\u4e3aG1\u52a8\u4f5c<\/li>\n<\/ol>\n<\/li>\n<\/ul>\n<h3>\u4f7f\u7528\u4ee3\u7801\u63a2\u9488\u5206\u6790\u673a\u5668\u4eba\u5efa\u6a21\uff0c\u5f97\u5230\u52a8\u4f5c\u6620\u5c04\u8868<\/h3>\n<p>\u6211\u4eec\u53ef\u4ee5\u627e\u5230<\/p>\n<pre><code class=\"language-python\">source\/isaaclab_tasks\/isaaclab_tasks\/manager_based\/locomotion\/velocity\/velocity_env_cfg.py\n109\u884c\u5f00\u59cb\n...\n\n@configclass\nclass ActionsCfg:\n    &quot;&quot;&quot;Action specifications for the MDP.&quot;&quot;&quot;\n    joint_pos = mdp.JointPositionActionCfg(asset_name=&quot;robot&quot;, joint_names=[&quot;.*&quot;], scale=0.5, use_default_offset=True)<\/code><\/pre>\n<p>\u8fd9\u91cc\u5b9a\u4e49\u4e86\u52a8\u4f5c\u8f93\u51fa\u65f6\u6240\u6709\u5173\u8282\u3002\u4e3a\u4e86\u63d0\u9ad8\u6269\u5c55\u6027\uff0c\u8fd9\u4e2a\u5b9a\u4e49\u51e0\u4e4e\u6ca1\u6709\u6211\u4eec\u53ef\u4ee5\u53c2\u8003\u7684\u5730\u65b9\uff0c\u6211\u4eec\u8981\u4f7f\u7528\u4ee3\u7801\u63a2\u9488\u5206\u6790\u8fd0\u884c\u65f6\u73af\u5883\u7684\u6784\u9020<\/p>\n<blockquote>\n<p>[!NOTE]<\/p>\n<p>\u9009\u62e9<code>rsl_rl<\/code>\u7684\u7406\u7531\u53ea\u662f\u56e0\u4e3a\u5927\u90e8\u5206\u73af\u5883\u90fd\u63d0\u4f9b\u4e86\u5b83\u7684\u8bad\u7ec3\u5165\u53e3<\/p>\n<\/blockquote>\n<p>\u4fee\u6539<code>rsl<\/code>\u5e93\u7684\u8bad\u7ec3\u4ee3\u7801, \u4ed6\u4f4d\u4e8e\uff1a<\/p>\n<pre><code>scripts\/reinforcement_learning\/rsl_rl\/train.py<\/code><\/pre>\n<p>\u52a0\u5165\u4ee3\u7801\u63a2\u9488\uff0c\u8fd9\u662f\u4e00\u4e2a\u8fd0\u884c\u65f6\u53c2\u6570<code>--extract<\/code>\uff0c\u4ee3\u8868\u4e0d\u8bad\u7ec3\uff0c\u800c\u662f\u5206\u6790\u73af\u5883\u5185\u5bb9<\/p>\n<p>\u8fd9\u4e2a\u53c2\u6570\uff1a<\/p>\n<ul>\n<li>\u5728\u5d4c\u5957\u7684\u73af\u5883\u4e2d\u9012\u5f52\u641c\u7d22\u540d\u4e3a\u540d\u4e3a<code>robot<\/code>\u7684\u5bf9\u8c61\uff0c\u5982\u679c\u627e\u5230\uff0c\u6253\u5370\u5b83\u7684\u5173\u8282\u987a\u5e8f\uff0c\u8fd9\u5c06\u7528\u6765\u5bf9\u9f50\u52a8\u4f5c\u7a7a\u95f4\uff08\u7f51\u7edc\u8f93\u51fa\uff09<\/li>\n<li>\u63d0\u53d6\u4e00\u6b21\u89c2\u6d4b\uff0c\u5373<code>observation<\/code>\uff0c\u6253\u5370\u5f62\u72b6\uff0c\u8fd9\u5c06\u7528\u6765\u5bf9\u9f50\u89c2\u6d4b\u7a7a\u95f4\uff08\u7f51\u7edc\u8f93\u5165\uff09<\/li>\n<\/ul>\n<pre><code class=\"language-python\">...\n# \u65b0\u589e\uff1a\u63d0\u53d6\u4fe1\u606f\u7684\u64cd\u4f5c\nparser.add_argument(&quot;--extract&quot;, action=&quot;store_true&quot;, default=False, help=&quot;Extract joint names &amp; observations and exit.&quot;)\n...\n\n# \u5982\u679c\u542f\u7528\u4e86 --extract \u53c2\u6570\uff0c\u5219\u53ea\u63d0\u53d6\u5173\u8282\u5e76\u9000\u51fa\n    if args_cli.extract:\n        def find_robot(env): #\u5728\u5d4c\u5957\u73af\u5883\u4e2d\u9012\u5f52\u7684\u5bfb\u627e\u540d\u4e3arobot\u7684\u5bf9\u8c61\n            max_depth = 10\n            for _ in range(max_depth):\n                if hasattr(env, &quot;scene&quot;):\n                    if hasattr(env.scene, &quot;robot&quot;):\n                        return env.scene.robot\n                    if hasattr(env.scene, &quot;articulations&quot;) and &quot;robot&quot; in env.scene.articulations:\n                        return env.scene.articulations[&quot;robot&quot;]\n                if hasattr(env, &quot;envs&quot;):\n                    env = env.envs[0]\n                elif hasattr(env, &quot;env&quot;):\n                    env = env.env\n                elif hasattr(env, &quot;unwrapped&quot;):\n                    env = env.unwrapped\n                else:\n                    break\n            raise AttributeError(&quot;Could not find robot in environment.&quot;)\n\n        robot = find_robot(env)\n        joint_names = robot.joint_names\n\n        print(&quot;Extracted joint names:&quot;)\n        for name in joint_names: # \u627e\u5230\u7684robot\uff0c\u6253\u5370\u5173\u8282\u987a\u5e8f\n            print(name)<\/code><\/pre>\n<p>\u4f7f\u7528\u63a2\u9488\u53ef\u4ee5\u5206\u6790\u5f97\u5230H1\u548cG1\u7684\u5173\u8282\u5168\u79f0\uff0c\u5176\u4e2dH1\u670919\u4e2a\u5173\u8282\uff0cG1\u670937\u4e2a\uff0c\u5206\u522b\u662f<\/p>\n<pre><code class=\"language-python\">h1_joint_names = [&quot;left_hip_yaw&quot;, &quot;right_hip_yaw&quot;, &quot;torso&quot;, &quot;left_hip_roll&quot;, &quot;right_hip_roll&quot;, &quot;left_shoulder_pitch&quot;, &quot;right_shoulder_pitch&quot;, &quot;left_hip_pitch&quot;, &quot;right_hip_pitch&quot;, &quot;left_shoulder_roll&quot;, &quot;right_shoulder_roll&quot;, &quot;left_knee&quot;, &quot;right_knee&quot;, &quot;left_shoulder_yaw&quot;, &quot;right_shoulder_yaw&quot;, &quot;left_ankle&quot;, &quot;right_ankle&quot;, &quot;left_elbow&quot;, &quot;right_elbow&quot;]\n\ng1_joint_names = [&quot;left_hip_pitch_joint&quot;, &quot;right_hip_pitch_joint&quot;, &quot;torso_joint&quot;, &quot;left_hip_roll_joint&quot;, &quot;right_hip_roll_joint&quot;, &quot;left_shoulder_pitch_joint&quot;, &quot;right_shoulder_pitch_joint&quot;, &quot;left_hip_yaw_joint&quot;, &quot;right_hip_yaw_joint&quot;, &quot;left_shoulder_roll_joint&quot;, &quot;right_shoulder_roll_joint&quot;, &quot;left_knee_joint&quot;, &quot;right_knee_joint&quot;, &quot;left_shoulder_yaw_joint&quot;, &quot;right_shoulder_yaw_joint&quot;, &quot;left_ankle_pitch_joint&quot;, &quot;right_ankle_pitch_joint&quot;, &quot;left_elbow_pitch_joint&quot;, &quot;right_elbow_pitch_joint&quot;, &quot;left_ankle_roll_joint&quot;, &quot;right_ankle_roll_joint&quot;, &quot;left_elbow_roll_joint&quot;, &quot;right_elbow_roll_joint&quot;, &quot;left_five_joint&quot;, &quot;left_three_joint&quot;, &quot;left_zero_joint&quot;, &quot;right_five_joint&quot;, &quot;right_three_joint&quot;, &quot;right_zero_joint&quot;, &quot;left_six_joint&quot;, &quot;left_four_joint&quot;, &quot;left_one_joint&quot;, &quot;right_six_joint&quot;, &quot;right_four_joint&quot;, &quot;right_one_joint&quot;, &quot;left_two_joint&quot;, &quot;right_two_joint&quot;]<\/code><\/pre>\n<blockquote>\n<p>[!IMPORTANT]<\/p>\n<p>\u4e0a\u9762\u7684\u5173\u8282\u987a\u5e8f\u8868\u662f\u6709\u5e8f\u7684\uff01\u4ed6\u7684\u987a\u5e8f\u548c\u7b56\u7565\u8f93\u51fa\u7ef4\u5ea6\u4e00\u4e00\u5bf9\u5e94<\/p>\n<\/blockquote>\n<p>\u6211\u4eec\u518d\u5c06\u6a21\u578b\u7684\u7ed3\u6784\u6253\u5370\u51fa\u6765\uff0c\u9a8c\u8bc1\u4e0a\u6587\u5f97\u5230\u7684\u5173\u8282\u6570\u76ee<\/p>\n<p>H1\u7684Actor\u7f51\u7edc<\/p>\n<pre><code>Actor MLP: Sequential(\n  (0): Linear(in_features=69, out_features=128, bias=True)\n  (1): ELU(alpha=1.0)\n  (2): Linear(in_features=128, out_features=128, bias=True)\n  (3): ELU(alpha=1.0)\n  (4): Linear(in_features=128, out_features=128, bias=True)\n  (5): ELU(alpha=1.0)\n  (6): Linear(in_features=128, out_features=19, bias=True)\n)<\/code><\/pre>\n<p>G1\u7684Actor\u7f51\u7edc<\/p>\n<pre><code>Actor MLP: Sequential(\n  (0): Linear(in_features=123, out_features=256, bias=True)\n  (1): ELU(alpha=1.0)\n  (2): Linear(in_features=256, out_features=128, bias=True)\n  (3): ELU(alpha=1.0)\n  (4): Linear(in_features=128, out_features=128, bias=True)\n  (5): ELU(alpha=1.0)\n  (6): Linear(in_features=128, out_features=37, bias=True)\n)<\/code><\/pre>\n<p>\u53ef\u4ee5\u770b\u5230\u8f93\u51fa\u7ef4\u5ea6\u662f19\u548c37\uff0c\u8fd9\u4e0e\u6211\u4eec\u5f97\u5230\u7684\u5173\u8282\u6570\u76ee\u5339\u914d\uff0c\u8bf4\u660e\u6211\u4eec\u7684\u5173\u8282\u6570\u76ee\u6ca1\u95ee\u9898\u3002<\/p>\n<blockquote>\n<p>[!NOTE]<\/p>\n<p>\u6253\u5370\u7f51\u7edc\u4e0d\u4ec5\u53ef\u4ee5\u5f97\u5230\u8f93\u51fa\u7ef4\u5ea6\uff08\u5173\u8282\u6570\u76ee\uff09\uff0c\u8fd8\u5f97\u5230\u4e86\u8f93\u5165\u7ef4\u5ea6\uff08\u89c2\u6d4b\u7ef4\u5ea6\uff09\uff0c\u672c\u8282\u4e3b\u8981\u662f\u5206\u6790\u5173\u8282\u7684\u5bf9\u9f50\uff0c\u89c2\u6d4b\u7684\u5bf9\u9f50\u5728\u4e0b\u4e00\u8282\u9610\u8ff0<\/p>\n<\/blockquote>\n<p>\u73b0\u5728\u6211\u4eec\u53ef\u4ee5\u518d\u786e\u8ba4\u4e00\u4e0b\u5173\u8282\u7ed3\u6784\uff0c\u5e76\u987a\u5e26\u83b7\u53d6\u673a\u5668\u4eba\u7684<code>.usd<\/code>\u5efa\u6a21\u6587\u4ef6\u3002<\/p>\n<p><strong>\u6211\u7528\u5230\u4e00\u4e2a\u6709\u8da3\u7684\u5de5\u7a0b\u6280\u5de7<\/strong>\uff1aIsaaclab\u7684\u5b98\u65b9\u4ed3\u5e93\u91cc\u6ca1\u6709\u673a\u5668\u4eba\u7684<code>.usd<\/code>\u6587\u4ef6\uff0c\u4f46\u4ed6\u80fd\u6b63\u5e38\u5b8c\u6210\u8bad\u7ec3\uff0c\u8fd9\u8bf4\u660e\u4ed6\u5927\u6982\u7387\u8c03\u7528\u4e86\u7f51\u7edc\u4e0a\u7684\u8d44\u6e90\u3002<strong>\u6211\u4eec\u53ef\u4ee5\u624b\u52a8\u65ad\u7f51\uff0c\u6e05\u7406\u7f13\u5b58\uff0c\u7136\u540e\u8fd0\u884c\u8bad\u7ec3\uff0c\u7b49\u5f85\u8fde\u63a5\u8d85\u65f6\u62a5\u9519<\/strong>\uff1a<\/p>\n<pre><code class=\"language-shell\">(is) ecs-user@iZuf6g6vii5ci3ikxssy9lZ:~\/IsaacLab$ source ~\/proxy-on.sh 9999 # \u8bbe\u7f6e\u4e00\u4e2a\u4e0d\u5b58\u5728\u7684\u4ee3\u7406\n127.0.0.1:9999\n(is) ecs-user@iZuf6g6vii5ci3ikxssy9lZ:~\/IsaacLab$ curl baidu.com # \u786e\u8ba4\u4e00\u4e0b\u7f51\u7edc\u662f\u5426\u771f\u7684\u5931\u6548\u4e86\ncurl: (7) Failed to connect to 127.0.0.1 port 9999 after 0 ms: Connection refused\n(is) ecs-user@iZuf6g6vii5ci3ikxssy9lZ:~\/IsaacLab$ python scripts\/reinforcement_learning\/rsl_rl\/train.py --task=Isaac-Velocity-Flat-G1-v0 --headless\n...\nFile &quot;\/home\/ecs-user\/IsaacLab\/source\/isaaclab\/isaaclab\/sim\/spawners\/from_files\/from_files.py&quot;, line 66, in spawn_from_usd\n    return _spawn_from_usd_file(prim_path, cfg.usd_path, cfg, translation, orientation)\n  File &quot;\/home\/ecs-user\/IsaacLab\/source\/isaaclab\/isaaclab\/sim\/spawners\/from_files\/from_files.py&quot;, line 237, in _spawn_from_usd_file\n    raise FileNotFoundError(f&quot;USD file not found at path: &#039;{usd_path}&#039;.&quot;)\nFileNotFoundError: USD file not found at path: &#039;http:\/\/omniverse-content-production.s3-us-west-2.amazonaws.com\/Assets\/Isaac\/4.5\/Isaac\/IsaacLab\/Robots\/Unitree\/G1\/g1_minimal.usd&#039;.<\/code><\/pre>\n<p>\u5728\u7b49\u5f85\u6570\u5206\u949f\u540e\uff0cpython\u629b\u51fa\u4e86\u4e00\u4e2a\u627e\u4e0d\u5230\u6587\u4ef6\u7684\u5f02\u5e38\uff0c\u6211\u4eec\u770b\u5230\u8fd9\u4e2a\u6a21\u578b\u6587\u4ef6\u662f<\/p>\n<pre><code class=\"language-shell\">http:\/\/omniverse-content-production.s3-us-west-2.amazonaws.com\/Assets\/Isaac\/4.5\/Isaac\/IsaacLab\/Robots\/Unitree\/G1\/g1_minimal.usd<\/code><\/pre>\n<p>\u8fd9\u6b63\u662f\u6211\u4eec\u8981\u627e\u7684<code>.usd<\/code>\u6587\u4ef6\uff01\u6211\u4eec\u53ef\u4ee5\u5199\u4e00\u4e2a\u811a\u672c\u5206\u6790\u4ed6\u7684\u7ed3\u6784\uff1a<\/p>\n<pre><code class=\"language-python\">def explore_usd_file(path):\n    stage = Usd.Stage.Open(path)\n    if not stage:\n        print(f&quot;Failed to open {path}&quot;)\n        return\n    print(f&quot;\\n--- Exploring: {path} ---&quot;)\n    # \u904d\u5386\u6240\u6709 Prim\uff08\u7269\u4f53\u8282\u70b9\uff09\n    for prim in stage.Traverse():\n        print(f&quot;Prim: {prim.GetPath()}  Type: {prim.GetTypeName()}&quot;)\n        # \u793a\u4f8b\uff1a\u5982\u679c\u662f Mesh\uff0c\u6253\u5370\u70b9\u6570\n        if prim.GetTypeName() == &quot;Mesh&quot;:\n            mesh = UsdGeom.Mesh(prim)\n            points_attr = mesh.GetPointsAttr()\n            points = points_attr.Get()\n            print(f&quot;  Mesh with {len(points)} points&quot;)<\/code><\/pre>\n<p>\u7136\u540e\u53ef\u4ee5\u5f97\u5230<\/p>\n<pre><code>--- Exploring: \/home\/ecs-user\/working-labs\/H1transferG1\/assets\/g1_minimal.usd ---\nPrim: \/physicsScene  Type: PhysicsScene\nPrim: \/g1  Type: Xform\nPrim: \/g1\/pelvis  Type: Xform\nPrim: \/g1\/pelvis\/visuals  Type: \nPrim: \/g1\/pelvis\/left_hip_pitch_joint  Type: PhysicsRevoluteJoint\nPrim: \/g1\/pelvis\/pelvis_contour_joint  Type: PhysicsFixedJoint\nPrim: \/g1\/pelvis\/right_hip_pitch_joint  Type: PhysicsRevoluteJoint\n...<\/code><\/pre>\n<p>\u8fd9\u5c31\u662f\u8be6\u7ec6\u7684\u7ed3\u6784\u4e86\uff0c\u6211\u4eec\u53ef\u4ee5\u5bf9\u7167\u4e0a\u9762\u7684\u5217\u8868\u5206\u6790\u5173\u8282\uff0c\u6b64\u5916\uff0c\u6211\u4eec\u5c06<code>.usd<\/code>\u6587\u4ef6\u4e0b\u8f7d\u5230\u4e86\u672c\u5730\uff0c\u8fd9\u5728\u4e0b\u9762\u7684\u5de5\u4f5c\u4e2d\u975e\u5e38\u6709\u7528\u3002<\/p>\n<p>\u5728\u5f97\u5230\u5173\u8282\u547d\u540d\u540e\uff0c\u6211\u4eec\u53ef\u4ee5\u5efa\u7acbH1\u5230G1\u7684\u5173\u8282\u6620\u5c04\u8868\u3002<strong>\u56e0\u4e3aG1\u7684\u5173\u8282\u6bd4H1\u66f4\u591a\uff0c\u6211\u4eec\u5c06\u5197\u4f59\u7684\u81ea\u7531\u5ea6\u5168\u90e8\u8bbe\u4e3a0\u3002<\/strong><\/p>\n<blockquote>\n<p>[!NOTE]<\/p>\n<p>\u56de\u987e\u4e00\u4e0b\u6d41\u7a0b\uff1a<\/p>\n<p><code>G1\u89c2\u6d4b-&gt;\u8f6c\u6362\u4e3aH1\u89c2\u6d4b-&gt;\u8f93\u5165H1\u7b56\u7565-&gt;\u5f97\u5230H1\u52a8\u4f5c\uff08\u5173\u8282\u52a8\u4f5c\u8868\uff09-&gt;\u6620\u5c04\u4e3aG1\u52a8\u4f5c<\/code><\/p>\n<p>\u8fd9\u8bf4\u660e\u6211\u4eec\u53ea\u9700\u8981H1\u5230G1\u7684\u5355\u5411\u5173\u8282\u6620\u5c04\uff0c\u800c\u4e0d\u9700\u8981G1\u5230H1\u7684<\/p>\n<\/blockquote>\n<pre><code class=\"language-python\">H1_TO_G1_JOINT_MAP = {\n    # \u4e0b\u80a2\uff08\u5de6\u817f\uff09\n    &quot;left_hip_yaw&quot;: &quot;left_hip_yaw_joint&quot;,\n    &quot;left_hip_roll&quot;: &quot;left_hip_roll_joint&quot;,\n    &quot;left_hip_pitch&quot;: &quot;left_hip_pitch_joint&quot;,\n    &quot;left_knee&quot;: &quot;left_knee_joint&quot;,\n    &quot;left_ankle&quot;: &quot;left_ankle_pitch_joint&quot;,\n    # G1\u591a\u51fa\u7684\u8e1d\u5173\u8282roll\n    &quot;left_ankle_roll_joint&quot;: 0.0,\n    # \u4e0b\u80a2\uff08\u53f3\u817f\uff09\n    &quot;right_hip_yaw&quot;: &quot;right_hip_yaw_joint&quot;,\n    &quot;right_hip_roll&quot;: &quot;right_hip_roll_joint&quot;,\n    &quot;right_hip_pitch&quot;: &quot;right_hip_pitch_joint&quot;,\n    &quot;right_knee&quot;: &quot;right_knee_joint&quot;,\n    &quot;right_ankle&quot;: &quot;right_ankle_pitch_joint&quot;,\n    # G1\u591a\u51fa\u7684\u8e1d\u5173\u8282roll\n    &quot;right_ankle_roll_joint&quot;: 0.0,\n    # \u8eaf\u5e72\n    &quot;torso&quot;: &quot;torso_joint&quot;,\n    # \u4e0a\u80a2\uff08\u5de6\u81c2\uff09\n    &quot;left_shoulder_pitch&quot;: &quot;left_shoulder_pitch_joint&quot;,\n    &quot;left_shoulder_roll&quot;: &quot;left_shoulder_roll_joint&quot;,\n    &quot;left_shoulder_yaw&quot;: &quot;left_shoulder_yaw_joint&quot;,\n    &quot;left_elbow&quot;: &quot;left_elbow_pitch_joint&quot;,\n    # G1\u591a\u51fa\u7684\u8098\u5173\u8282roll\n    &quot;left_elbow_roll_joint&quot;: 0.0,\n    # \u4e0a\u80a2\uff08\u53f3\u81c2\uff09\n    &quot;right_shoulder_pitch&quot;: &quot;right_shoulder_pitch_joint&quot;,\n    &quot;right_shoulder_roll&quot;: &quot;right_shoulder_roll_joint&quot;,\n    &quot;right_shoulder_yaw&quot;: &quot;right_shoulder_yaw_joint&quot;,\n    &quot;right_elbow&quot;: &quot;right_elbow_pitch_joint&quot;,\n    # G1\u591a\u51fa\u7684\u8098\u5173\u8282roll\n    &quot;right_elbow_roll_joint&quot;: 0.0,\n}<\/code><\/pre>\n<p>\u6709\u4e86\u8fd9\u5f20\u5b8c\u6574\u7684\u5173\u8282\u6620\u5c04\u8868\uff0c\u6211\u4eec\u53ef\u4ee5\u5199\u51faH1\u5230G1\u7684\u52a8\u4f5c\u6620\u5c04\u51fd\u6570\u3002<\/p>\n<blockquote>\n<p>[!IMPORTANT]<\/p>\n<p>\u4e0b\u9762\u7684\u52a8\u4f5c\u6620\u5c04\u51fd\u6570\u6b63\u786e\u524d\u63d0\u662f\u5173\u8282\u540d\u5217\u8868\u6709\u5e8f\uff0c\u8fd9\u9700\u8981\u548c\u795e\u7ecf\u7f51\u7edc\u8f93\u51fa\u505a\u4e00\u4e00\u5bf9\u5e94\u3002\u8fd9\u4e5f\u662f\u4e3a\u4ec0\u4e48\u4e0a\u6587\u82b1\u4e86\u90a3\u4e48\u591a\u529f\u592b\u786e\u8ba4\u5173\u8282\u5217\u8868\u987a\u5e8f\u3002<\/p>\n<\/blockquote>\n<pre><code class=\"language-python\">def map_h1_action_to_g1(h1_action, h1_joint_names=h1_joint_names, g1_joint_names=g1_joint_names):\n    # H1\u5173\u8282\u540d\u5230action\u503c\u7684\u6620\u5c04\n    h1_action_dict = dict(zip(h1_joint_names, h1_action))\n    g1_action = []\n    for g1_joint in g1_joint_names:\n        # \u53cd\u67e5\u6620\u5c04\u8868\uff0c\u627e\u5230H1\u5173\u8282\u540d\n        h1_key = None\n        for k, v in H1_TO_G1_JOINT_MAP.items():\n            if v == g1_joint:\n                h1_key = k\n                break\n        if h1_key is not None and h1_key in h1_action_dict:\n            g1_action.append(h1_action_dict[h1_key])\n        elif g1_joint in H1_TO_G1_JOINT_MAP and H1_TO_G1_JOINT_MAP[g1_joint] == 0.0:\n            g1_action.append(0.0)\n        else:\n            # \u5176\u5b83G1\u5173\u8282\uff08\u5982\u624b\u6307\u7b49\uff09\uff0c\u4e5f\u8865\u96f6\n            g1_action.append(0.0)\n    return g1_action<\/code><\/pre>\n<p>\u81f3\u6b64\uff0c\u6211\u4eec\u5f97\u5230\u4e86\u5173\u952e\u51fd\u6570<code>map_h1_action_to_g1<\/code>\uff0c\u8fd9\u89e3\u51b3\u4e86\u8fc1\u79fb\u6d41\u7a0b\u4e2d\u91cd\u8981\u7684\u90e8\u5206\uff1a<strong>\u5c06H1\u7b56\u7565\u6a21\u578b\u7684\u8f93\u51fa\u5e94\u7528\u5728G1\u4e0a<\/strong>\u3002<\/p>\n<h3>\u4f7f\u7528\u4ee3\u7801\u63a2\u9488\u5206\u6790\u89c2\u6d4b\u6570\u636e\u6784\u9020\uff0c\u5f97\u5230\u89c2\u6d4b\u6620\u5c04\u8868<\/h3>\n<p>\u6ce8\u610f\u5230<\/p>\n<pre><code>source\/isaaclab_tasks\/isaaclab_tasks\/manager_based\/locomotion\/velocity\/velocity_env_cfg.py\n116\u884c\u5f00\u59cb<\/code><\/pre>\n<pre><code class=\"language-python\">@configclass\nclass ObservationsCfg:\n    &quot;&quot;&quot;Observation specifications for the MDP.&quot;&quot;&quot;\n\n    @configclass\n    class PolicyCfg(ObsGroup):\n        &quot;&quot;&quot;Observations for policy group.&quot;&quot;&quot;\n\n        # observation terms (order preserved)\n        base_lin_vel = ObsTerm(func=mdp.base_lin_vel, noise=Unoise(n_min=-0.1, n_max=0.1))\n        base_ang_vel = ObsTerm(func=mdp.base_ang_vel, noise=Unoise(n_min=-0.2, n_max=0.2))\n        projected_gravity = ObsTerm(\n            func=mdp.projected_gravity,\n            noise=Unoise(n_min=-0.05, n_max=0.05),\n        )\n        velocity_commands = ObsTerm(func=mdp.generated_commands, params={&quot;command_name&quot;: &quot;base_velocity&quot;})\n        joint_pos = ObsTerm(func=mdp.joint_pos_rel, noise=Unoise(n_min=-0.01, n_max=0.01))\n        joint_vel = ObsTerm(func=mdp.joint_vel_rel, noise=Unoise(n_min=-1.5, n_max=1.5))\n        actions = ObsTerm(func=mdp.last_action)\n        height_scan = ObsTerm(\n            func=mdp.height_scan,\n            params={&quot;sensor_cfg&quot;: SceneEntityCfg(&quot;height_scanner&quot;)},\n            noise=Unoise(n_min=-0.1, n_max=0.1),\n            clip=(-1.0, 1.0),\n        )\n\n        def __post_init__(self):\n            self.enable_corruption = True\n            self.concatenate_terms = True\n\n    # observation groups\n    policy: PolicyCfg = PolicyCfg()<\/code><\/pre>\n<p>\u8fd9\u4e2a\u914d\u7f6e\u6587\u4ef6\u5b9a\u4e49\u4e86\u89c2\u6d4b\u7684\u4e00\u822c\u5f62\u5f0f\uff1a<\/p>\n<table>\n<thead>\n<tr>\n<th>\u89c2\u6d4b\u9879<\/th>\n<th>\u8bf4\u660e<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><code>base_lin_vel<\/code>, <code>base_ang_vel<\/code><\/td>\n<td>\u5f53\u524d\u7684\u8fd0\u52a8\u72b6\u6001<\/td>\n<\/tr>\n<tr>\n<td><code>projected_gravity<\/code><\/td>\n<td>\u59ff\u6001\u4f30\u8ba1<\/td>\n<\/tr>\n<tr>\n<td><code>velocity_commands<\/code><\/td>\n<td>\u60f3\u8981\u8fbe\u5230\u7684\u901f\u5ea6\uff08\u4efb\u52a1\u76ee\u6807\uff09<\/td>\n<\/tr>\n<tr>\n<td><code>joint_pos<\/code>, <code>joint_vel<\/code><\/td>\n<td>\u8fd0\u52a8\u7cfb\u7edf\u7684\u72b6\u6001<\/td>\n<\/tr>\n<tr>\n<td><code>actions<\/code><\/td>\n<td>\u5386\u53f2\u52a8\u4f5c\uff08\u7528\u4e8e\u6784\u5efa\u8bb0\u5fc6\uff09<\/td>\n<\/tr>\n<tr>\n<td><code>height_scan<\/code><\/td>\n<td>\u5730\u5f62\u4fe1\u606f<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u7136\u800c\uff0c\u4ec5\u4ec5\u77e5\u9053\u8fd9\u4e9b\u662f\u4e0d\u8db3\u4ee5\u6784\u5efa\u89c2\u6d4b\u6620\u5c04\u8868\u7684\u3002\u6211\u4eec\u4ecd\u9700\u8981\u5728\u8bad\u7ec3\u4ee3\u7801\u4e2d\u52a0\u5165\u63a2\u9488\uff0c\u6253\u5370\u51fa\u6765\u770b\u4e00\u4e0b<\/p>\n<pre><code class=\"language-python\"># \u6cbf\u7528\u4e0a\u4e00\u8282\u63d0\u5230\u7684--extrat\u53c2\u6570\n# ====== \u65b0\u589e\uff1a\u63d0\u53d6 observation \u793a\u4f8b\u548c shape ======\nobs = env.reset()\nimport numpy as np\nprint(&quot;\\nExtracted observation structure:&quot;)\nif isinstance(obs, (tuple, list)) and len(obs) &gt; 0 and isinstance(obs[0], dict) and &#039;policy&#039; in obs[0]:\n    policy_obs = obs[0][&#039;policy&#039;]\n    print(f&quot;policy: shape={policy_obs.shape} dtype={getattr(policy_obs, &#039;dtype&#039;, type(policy_obs))}&quot;)\n    # \u6253\u5370\u7b2c\u4e00\u4e2a\u73af\u5883\u7684 observation \u5411\u91cf\n    arr = policy_obs[0].detach().cpu().numpy() if hasattr(policy_obs[0], &#039;detach&#039;) else np.array(policy_obs[0])\n    print(f&quot;  First row (len={len(arr)}): {arr}&quot;)\n    print(f&quot;  First 20 values: {arr[:20]}&quot;)\n    if hasattr(policy_obs, &#039;detach&#039;):\n        policy_obs = policy_obs.detach().cpu().numpy()\n\n    print(&quot;Index\\t[9]\\t\\t[10]\\t\\t[11]\\t\\t[12]&quot;)\n    for i in range(10):\n        print(f&quot;{i:&gt;5}\\t{policy_obs[i][9]:.6f}\\t{policy_obs[i][10]:.6f}\\t{policy_obs[i][11]:.6f}\\t{policy_obs[i][12]:.6f}&quot;)\nelse:\n    print(&quot;Observation type\/structure not recognized. Raw output:&quot;, obs)\n# ====== \u65b0\u589e\u7ed3\u675f ======<\/code><\/pre>\n<p>\u73b0\u5728\u6211\u4eec\u6253\u5370\u4e00\u4e0bH1\u7684\u89c2\u6d4b\u7a7a\u95f4<\/p>\n<pre><code class=\"language-python\">Extracted observation structure:\npolicy: shape=torch.Size([4096, 69]) dtype=torch.float32\n  First row (len=69): [-5.6066342e-02 -8.7517925e-02  4.9868785e-02 -9.6398674e-02\n -1.9630268e-01 -1.1851498e-01  2.4278294e-02  3.7880894e-02\n -1.0474691e+00  3.5836050e-01  0.0000000e+00 -4.6731085e-01\n  8.4095336e-03 -8.9571215e-03 -4.2297798e-03  4.5316387e-03\n -8.4742559e-03  3.4465501e-03  8.2933642e-03  3.7253592e-03\n  7.2544329e-03 -9.4385361e-03  4.5920312e-03 -6.5772794e-05\n -2.6932871e-03 -5.2364785e-03  8.4291194e-03  9.6111558e-03\n  8.7782890e-03  7.7467002e-03 -6.2260972e-03 -1.2100315e+00\n  5.8564234e-01 -4.7681594e-01 -7.0546228e-01  1.3577251e+00\n  3.5666263e-01  2.7153444e-01 -1.3317013e-01 -3.0642664e-01\n -1.2162994e+00 -3.2588577e-01 -4.4944549e-01 -6.0138190e-01\n -5.8937967e-01 -2.3791862e-01  9.6854997e-01  1.4231379e+00\n  5.8026981e-01  7.4255371e-01  0.0000000e+00  0.0000000e+00\n  0.0000000e+00  0.0000000e+00  0.0000000e+00  0.0000000e+00\n  0.0000000e+00  0.0000000e+00  0.0000000e+00  0.0000000e+00\n  0.0000000e+00  0.0000000e+00  0.0000000e+00  0.0000000e+00\n  0.0000000e+00  0.0000000e+00  0.0000000e+00  0.0000000e+00\n  0.0000000e+00]\n  First 20 values: [-0.05606634 -0.08751792  0.04986878 -0.09639867 -0.19630268 -0.11851498\n  0.02427829  0.03788089 -1.0474691   0.3583605   0.         -0.46731085\n  0.00840953 -0.00895712 -0.00422978  0.00453164 -0.00847426  0.00344655\n  0.00829336  0.00372536]\nIndex   [9]     [10]        [11]        [12]\n    0   0.358360    0.000000    -0.467311   0.008410\n    1   0.298718    0.000000    -0.848940   -0.000898\n    2   0.329300    0.000000    -0.413140   -0.008372\n    3   0.396957    0.000000    0.005005    -0.004949\n    4   0.635064    0.000000    0.069569    -0.008066\n    5   0.414920    0.000000    0.000215    0.006416\n    6   0.253440    0.000000    -0.763752   -0.002671\n    7   0.801332    0.000000    -0.484505   0.003594\n    8   0.278513    0.000000    0.841860    0.001726\n    9   0.966565    0.000000    -0.863862   0.000334<\/code><\/pre>\n<p><code>shape=torch.Size([4096, 69])<\/code>\u4ee3\u8868\u4e86H1\u7684\u89c2\u6d4b\u7a7a\u95f4\u662f69\u7ef4\u7684\u3002\u8fd8\u8bb0\u5f97\u4e0a\u8282\u4e2d\u63d0\u5230\u7684Actor\u7f51\u7edc\u5417\uff1fH1\u7684Actor\u7f51\u7edc\u662f<\/p>\n<pre><code class=\"language-shell\">Actor MLP: Sequential(\n  (0): Linear(in_features=69, out_features=128, bias=True)\n  (1): ELU(alpha=1.0)\n  (2): Linear(in_features=128, out_features=128, bias=True)\n  (3): ELU(alpha=1.0)\n  (4): Linear(in_features=128, out_features=128, bias=True)\n  (5): ELU(alpha=1.0)\n  (6): Linear(in_features=128, out_features=19, bias=True)\n)<\/code><\/pre>\n<p>\u592a\u597d\u4e86\uff0c\u8fd9\u8bc1\u660e\u6211\u4eec\u7684\u7ed3\u8bba\u6ca1\u6709\u95ee\u9898\u3002<\/p>\n<p>\u6211\u4eec\u9700\u8981\u5177\u4f53\u7684\u89c2\u6d4b\u6784\u9020\u65b9\u6cd5\uff0c\u4f46\u662f\u4ee3\u7801\u9ad8\u5ea6\u62bd\u8c61\uff0c\u6253\u5370\u51fa\u6765\u4e5f\u65e0\u6d4e\u4e8e\u4e8b\u3002\u6211\u4eec\u53ef\u4ee5\u603b\u7ed3\u4e00\u4e0b\u4e0a\u6587\uff0c\u6c47\u603b\u4e00\u4e0b\u5df2\u77e5\u4fe1\u606f<\/p>\n<ul>\n<li>H1\u670919\u4e2a\u5173\u8282\uff0c\u89c2\u6d4b\u7a7a\u95f469\u7ef4<\/li>\n<li>G1\u670937\u4e2a\u5173\u8282\uff0c\u89c2\u6d4b\u7a7a\u95f4123\u7ef4<\/li>\n<li>\u89c2\u6d4b\u7a7a\u95f4\u6309\u7167\u987a\u5e8f\uff0c\u5206\u522b\u662fbase_lin_vel, base_ang_vel, projected_gravity, velocity_commands, joint_pos, joint_vel, actions<\/li>\n<\/ul>\n<p>\u5f53\u5173\u8282\u6570\u76ee\u5df2\u77e5\u65f6\uff0cjoint_pos\uff0cjoint_vel, actions\u7684\u7ef4\u5ea6\u5c31\u5df2\u77e5\u4e86\uff0c\u52a0\u4e4b\u5e38\u89c1\u7684\u901f\u5ea6\u4e00\u822c\u662f2\u7ef4\u6216\u80053\u7ef4\uff0c\u6211\u4eec\u53ef\u4ee5\u89e3\u65b9\u7a0b+\u5408\u7406\u731c\u6d4b<\/p>\n<p>H1\u7684\u89c2\u6d4b\u7a7a\u95f4<\/p>\n<pre><code>[0:3]    base_lin_vel       (3)\n[3:6]    base_ang_vel       (3)\n[6:9]    projected_gravity  (3)\n[9:12]   velocity_commands  (3)\n[12:31]  joint_pos          (19)\n[31:50]  joint_vel          (19)\n[50:69]  actions            (19)<\/code><\/pre>\n<p>G1\u7684\u89c2\u6d4b\u7a7a\u95f4<\/p>\n<pre><code>[0:3]    base_lin_vel       (3)\n[3:6]    base_ang_vel       (3)\n[6:9]    projected_gravity  (3)\n[9:12]   velocity_commands  (3)\n[12:49]  joint_pos          (37)\n[49:86]  joint_vel          (37)\n[86:123] actions            (37)<\/code><\/pre>\n<p>\u8fd9\u662f\u4e00\u4e2a\u975e\u5e38\u5408\u7406\u7684\u89c2\u6d4b\u7ed3\u6784\uff01\u73b0\u5728\u6211\u4eec\u53ef\u4ee5\u5f00\u59cb\u5199\u89c2\u6d4b\u6620\u5c04\u51fd\u6570\uff1a<\/p>\n<pre><code class=\"language-python\">def map_g1_obs_to_h1_obs(policy_obs_row, g1_joint_names=g1_joint_names, h1_joint_names=h1_joint_names):\n    &quot;&quot;&quot;\n    \u5c06G1\u73af\u5883\u7684policy observation\u5411\u91cf\uff0c\u91cd\u6392\u4e3aH1\u7b56\u7565\u53ef\u7528\u7684\u8f93\u5165\u683c\u5f0f\u3002\n\n    Args:\n        policy_obs_row: G1\u73af\u5883\u8f93\u51fa\u7684\u5355\u4e2a\u73af\u5883\u7684policy observation\n        g1_joint_names: G1\u673a\u5668\u4eba\u7684\u5173\u8282\u540d\u79f0\u5217\u8868\n        h1_joint_names: H1\u673a\u5668\u4eba\u7684\u5173\u8282\u540d\u79f0\u5217\u8868\n\n    Returns:\n        h1_obs: \u91cd\u6392\u540e\u7b26\u5408H1\u7b56\u7565\u8f93\u5165\u683c\u5f0f\u7684observation\u5411\u91cf\n    &quot;&quot;&quot;\n    # \u521b\u5efa\u8f93\u51fa\u5411\u91cf\uff0c\u521d\u59cb\u4e3a\u7a7a\n    h1_obs = []\n\n    # 1. \u524d12\u7ef4\u76f4\u63a5\u590d\u5236 [0:12] (base_lin_vel, base_ang_vel, projected_gravity, velocity_commands)\n    h1_obs.extend(policy_obs_row[0:12])\n\n    # 2. \u91cd\u6392joint_pos\u90e8\u5206 [12:49] -&gt; [12:31]\n    g1_joint_pos = policy_obs_row[12:49]  # G1\u768437\u4e2a\u5173\u8282\u4f4d\u7f6e\n    g1_joint_pos_dict = dict(zip(g1_joint_names, g1_joint_pos))  # \u6784\u5efaG1\u5173\u8282\u540d\u5230\u4f4d\u7f6e\u7684\u6620\u5c04\n\n    # \u4e3aH1\u521b\u5efa\u5173\u8282\u4f4d\u7f6e\u90e8\u5206\n    h1_joint_pos = []\n    for h1_joint in h1_joint_names:\n        # \u627e\u5230\u5bf9\u5e94\u7684G1\u5173\u8282\u540d\n        g1_joint = H1_TO_G1_JOINT_MAP.get(h1_joint)\n        if g1_joint in g1_joint_pos_dict:\n            h1_joint_pos.append(g1_joint_pos_dict[g1_joint])\n        else:\n            # \u5982\u679c\u6ca1\u627e\u5230\u5bf9\u5e94\u7684G1\u5173\u8282\uff08\u7406\u8bba\u4e0a\u4e0d\u5e94\u8be5\u53d1\u751f\uff09\uff0c\u4f7f\u75280\u4f5c\u4e3a\u9ed8\u8ba4\u503c\n            h1_joint_pos.append(0.0)\n\n    h1_obs.extend(h1_joint_pos)\n\n    # 3. \u91cd\u6392joint_vel\u90e8\u5206 [49:86] -&gt; [31:50]\n    g1_joint_vel = policy_obs_row[49:86]  # G1\u768437\u4e2a\u5173\u8282\u901f\u5ea6\n    g1_joint_vel_dict = dict(zip(g1_joint_names, g1_joint_vel))  # \u6784\u5efaG1\u5173\u8282\u540d\u5230\u901f\u5ea6\u7684\u6620\u5c04\n\n    # \u4e3aH1\u521b\u5efa\u5173\u8282\u901f\u5ea6\u90e8\u5206\n    h1_joint_vel = []\n    for h1_joint in h1_joint_names:\n        # \u627e\u5230\u5bf9\u5e94\u7684G1\u5173\u8282\u540d\n        g1_joint = H1_TO_G1_JOINT_MAP.get(h1_joint)\n        if g1_joint in g1_joint_vel_dict:\n            h1_joint_vel.append(g1_joint_vel_dict[g1_joint])\n        else:\n            # \u5982\u679c\u6ca1\u627e\u5230\u5bf9\u5e94\u7684G1\u5173\u8282\uff0c\u4f7f\u75280\u4f5c\u4e3a\u9ed8\u8ba4\u503c\n            h1_joint_vel.append(0.0)\n\n    h1_obs.extend(h1_joint_vel)\n\n    # 4. \u91cd\u6392actions\u90e8\u5206 [86:123] -&gt; [50:69]\n    g1_actions = policy_obs_row[86:123]  # G1\u768437\u4e2a\u5173\u8282\u52a8\u4f5c\n    g1_actions_dict = dict(zip(g1_joint_names, g1_actions))  # \u6784\u5efaG1\u5173\u8282\u540d\u5230\u52a8\u4f5c\u7684\u6620\u5c04\n\n    # \u4e3aH1\u521b\u5efa\u52a8\u4f5c\u90e8\u5206\n    h1_actions = []\n    for h1_joint in h1_joint_names:\n        # \u627e\u5230\u5bf9\u5e94\u7684G1\u5173\u8282\u540d\n        g1_joint = H1_TO_G1_JOINT_MAP.get(h1_joint)\n        if g1_joint in g1_actions_dict:\n            h1_actions.append(g1_actions_dict[g1_joint])\n        else:\n            # \u5982\u679c\u6ca1\u627e\u5230\u5bf9\u5e94\u7684G1\u5173\u8282\uff0c\u4f7f\u75280\u4f5c\u4e3a\u9ed8\u8ba4\u503c\n            h1_actions.append(0.0)\n\n    h1_obs.extend(h1_actions)\n\n    # \u786e\u4fdd\u8f93\u51fa\u5411\u91cf\u7ef4\u5ea6\u4e3a69\n    assert len(h1_obs) == 69, f&quot;\u8f93\u51fa\u5411\u91cf\u7ef4\u5ea6\u5e94\u4e3a69\uff0c\u5f53\u524d\u4e3a{len(h1_obs)}&quot;\n\n    return h1_obs<\/code><\/pre>\n<p>\u81f3\u6b64\uff0c\u6211\u4eec\u5b8c\u6210\u4e86G1\u5230H1\u7684\u89c2\u6d4b\u6620\u5c04\uff01\u6211\u4eec\u7684\u6d41\u7a0b\uff1a<\/p>\n<pre><code>G1\u89c2\u6d4b-&gt;\u8f6c\u6362\u4e3aH1\u89c2\u6d4b-&gt;\u8f93\u5165H1\u7b56\u7565-&gt;\u5f97\u5230H1\u52a8\u4f5c\uff08\u5173\u8282\u52a8\u4f5c\u8868\uff09-&gt;\u6620\u5c04\u4e3aG1\u52a8\u4f5c<\/code><\/pre>\n<p>\u5df2\u7ecf\u6784\u5efa\u5b8c\u6210\uff01<\/p>\n<h3>\u4fee\u6539\u6742\u9879\u53c2\u6570\uff0c\u5bf9\u9f50G1\u5230H1<\/h3>\n<p>\u5728\u6211\u4eec\u8dd1\u901a\u8fd9\u4e2a\u8fc1\u79fb\u4e4b\u524d\uff0c\u6211\u4eec\u8fd8\u5e94\u8c03\u6574G1\u7684\u521d\u59cb\u7ad9\u59ff\uff0c\u8fd9\u975e\u5e38\u91cd\u8981\uff0c\u56e0\u4e3aH1\u7684\u7b56\u7565\u90fd\u662f\u4ece\u56fa\u5b9a\u7684\u7ad9\u59ff\u5f00\u59cb\u7684\uff0c\u5982\u679c\u6a21\u578b\u6536\u5230\u4e86\u672a\u77e5\u7684\u521d\u59cb\u503c\uff0c\u5f88\u53ef\u80fd\u8fd4\u56de\u968f\u673a\u7684\u7ed3\u679c\uff0c\u8fd9\u4f1a\u8ba9\u673a\u5668\u4eba\u5728\u4e00\u5f00\u59cb\u5c31\u65e0\u6cd5\u7ad9\u4f4f<\/p>\n<pre><code class=\"language-python\">G1_CFG = ArticulationCfg(\n...\n    # \u4fee\u6539\u521d\u59cb\u72b6\u6001\u4ee5\u5339\u914dH1\u7684\u914d\u7f6e\n    init_state=ArticulationCfg.InitialStateCfg(\n        pos=(0.0, 0.0, 0,74),  # \u9ad8\u5ea6\u4e0d\u7528\u6539\n        joint_pos={\n            # \u4e0b\u80a2\u5173\u8282 - \u6620\u5c04\u81eaH1\n            &quot;.*_hip_yaw_joint&quot;: 0.0,  # \u5bf9\u5e94H1\u7684 *_hip_yaw\n            &quot;.*_hip_roll_joint&quot;: 0.0,  # \u5bf9\u5e94H1\u7684 *_hip_roll\n            &quot;.*_hip_pitch_joint&quot;: -0.28,  # \u5bf9\u5e94H1\u7684 *_hip_pitch (-16\u5ea6)\n            &quot;.*_knee_joint&quot;: 0.79,  # \u5bf9\u5e94H1\u7684 *_knee (45\u5ea6)\n            &quot;.*_ankle_pitch_joint&quot;: -0.52,  # \u5bf9\u5e94H1\u7684 *_ankle (-30\u5ea6)\n            &quot;.*_ankle_roll_joint&quot;: 0.0,  # G1\u7279\u6709\uff0cH1\u6ca1\u6709\uff0c\u8bbe\u4e3a0\n\n            # \u8eaf\u5e72\u5173\u8282\n            &quot;torso_joint&quot;: 0.0,  # \u5bf9\u5e94H1\u7684 torso\n\n            # \u4e0a\u80a2\u5173\u8282 - \u6620\u5c04\u81eaH1\n            &quot;.*_shoulder_pitch_joint&quot;: 0.28,  # \u5bf9\u5e94H1\u7684 *_shoulder_pitch\n            &quot;.*_shoulder_roll_joint&quot;: 0.0,  # \u5bf9\u5e94H1\u7684 *_shoulder_roll\n            &quot;.*_shoulder_yaw_joint&quot;: 0.0,  # \u5bf9\u5e94H1\u7684 *_shoulder_yaw\n            &quot;.*_elbow_pitch_joint&quot;: 0.52,  # \u5bf9\u5e94H1\u7684 *_elbow\n            &quot;.*_elbow_roll_joint&quot;: 0.0,  # G1\u7279\u6709\uff0cH1\u6ca1\u6709\uff0c\u8bbe\u4e3a0\n\n            # \u624b\u90e8\u5173\u8282 - G1\u7279\u6709\uff0c\u4fdd\u6301\u539f\u59cb\u503c\u6216\u8bbe\u4e3a\u4e2d\u7acb\u4f4d\u7f6e\n            &quot;.*_five_joint&quot;: 0.0,  # G1\u7279\u6709\uff0c\u8bbe\u4e3a\u4e2d\u7acb\u4f4d\u7f6e\n            &quot;.*_three_joint&quot;: 0.0,  # G1\u7279\u6709\uff0c\u8bbe\u4e3a\u4e2d\u7acb\u4f4d\u7f6e\n            &quot;.*_six_joint&quot;: 0.0,  # G1\u7279\u6709\uff0c\u8bbe\u4e3a\u4e2d\u7acb\u4f4d\u7f6e\n            &quot;.*_four_joint&quot;: 0.0,  # G1\u7279\u6709\uff0c\u8bbe\u4e3a\u4e2d\u7acb\u4f4d\u7f6e\n            &quot;.*_zero_joint&quot;: 0.0,  # G1\u7279\u6709\uff0c\u8bbe\u4e3a\u4e2d\u7acb\u4f4d\u7f6e\n            &quot;left_one_joint&quot;: 1.0,  # \u4fdd\u6301\u539f\u914d\u7f6e\u4e2d\u7684\u503c\n            &quot;right_one_joint&quot;: -1.0,  # \u4fdd\u6301\u539f\u914d\u7f6e\u4e2d\u7684\u503c\n            &quot;left_two_joint&quot;: 0.52,  # \u4fdd\u6301\u539f\u914d\u7f6e\u4e2d\u7684\u503c\n            &quot;right_two_joint&quot;: -0.52,  # \u4fdd\u6301\u539f\u914d\u7f6e\u4e2d\u7684\u503c\n        },\n...<\/code><\/pre>\n<h3>\u6784\u5efa\u6d4b\u8bd5\u73af\u5883\uff0c\u8dd1\u901a\u5b9e\u9a8c<\/h3>\n<p>\u5728\u6709\u4e86\u5b8c\u6574pipline\u4e4b\u540e\uff0c\u6211\u4eec\u53ef\u4ee5\u5728Task1\u5b9e\u9a8c\u7684\u57fa\u7840\u4e0a\u642d\u5efa\u6d4b\u8bd5\u73af\u5883<\/p>\n<ol>\n<li>\u521d\u59cb\u5316\u73af\u5883\u4e3a\u5e73\u5730<\/li>\n<li>G1\u673a\u5668\u4eba\u89c2\u6d4b<\/li>\n<li>\u8c03\u7528\u51fd\u6570\uff0c\u5c06G1\u89c2\u6d4b\u8f6c\u6362\u4e3aH1\u89c2\u6d4b<\/li>\n<li>\u8c03\u7528\u6a21\u578b\uff0c\u5c06H1\u89c2\u6d4b\u63a8\u7406\u4e3aH1\u52a8\u4f5c<\/li>\n<li>\u8c03\u7528\u51fd\u6570\uff0c\u5c06H1\u52a8\u4f5c\u8f6c\u6362\u4e3aG1\u52a8\u4f5c<\/li>\n<li>\u8fed\u4ee3\u4eff\u771f<\/li>\n<\/ol>\n<p><img decoding=\"async\" src=\"https:\/\/pic.cirno.fun\/sssn-blog-pics\/image-20250710060559151.png\" alt=\"image-20250710060559151\" style=\"zoom:50%;\" \/><\/p>\n<p>\u5b9e\u9a8c\u7684\u7ed3\u679c\u662f\uff0c\u642d\u8f7dH1\u7b56\u7565\u7684G1\u53ef\u4ee5\u4f5c\u51fa\u884c\u8d70\u7684\u52a8\u4f5c\uff1a\u4f46\u4ed6\u4f1a\u5728\u5c1d\u8bd5\u8d704-6\u6b65\u5185\u56e0\u4e3a\u8eab\u4f53\u524d\u503e\u8dcc\u5012\u3002<\/p>\n<h2>\u66f4\u591a\u601d\u8003<\/h2>\n<p>\u642d\u8f7dH1\u7b56\u7565\u7684G1\u673a\u5668\u4eba\u4f1a\u5728<\/p>\n<ul>\n<li>\u5f00\u59cb\u65f6\u5c1d\u8bd5\u8d70\u8def<\/li>\n<li>\u7136\u540e\u539f\u5730\u8e0f\u6b65<\/li>\n<li>\u63a5\u7740\u8dcc\u5012<\/li>\n<\/ul>\n<p>\u8bf4\u660e\u7b56\u7565\u5728\u4e00\u5f00\u59cb\u8fd8\u201c\u6709\u70b9\u7528\u201d\uff0c\u4f46\u63a5\u7740\u65e0\u6cd5\u9002\u5e94\u65b0\u673a\u5668\u4eba\u7684\u673a\u68b0\u7ed3\u6784\u3002<\/p>\n<h3>\u4e3b\u8981\u5931\u8d25\u539f\u56e0<\/h3>\n<h4>\u5173\u8282\u8865\u96f6\u7684\u65b9\u5f0f\u5bfc\u81f4\u7b56\u7565\u5931\u6548<\/h4>\n<p>\u6211\u4eec\u628a H1 \u4e0a\u4e0d\u5b58\u5728\u7684\u5173\u8282\u5728\u52a8\u4f5c\u7a7a\u95f4\u91cc\u7f6e 0\uff0c\u8fd9\u4f1a\u8ba9 G1 \u7684\u8fd9\u4e9b\u5173\u8282\u201c\u50f5\u786c\u201d\u6216\u201c\u60ac\u7a7a\u201d\uff0c\u4ece\u800c\uff1a<\/p>\n<ul>\n<li>\u5bfc\u81f4 G1 \u7684\u8d28\u91cf\u5206\u5e03\u548c\u52a8\u6001\u884c\u4e3a\u4e0e H1 \u5dee\u5f02\u8f83\u5927<\/li>\n<\/ul>\n<h4>\u52a8\u529b\u5b66\u5dee\u5f02<\/h4>\n<p>H1 \u548c G1 \u7684\u8d28\u91cf\u5206\u5e03\u3001\u817f\u957f\u3001\u60ef\u6027\u5f20\u91cf\u3001\u6469\u64e6\u7cfb\u6570\u7b49\u90fd\u4e0d\u540c\uff0c\u5373\u4f7f\u89c2\u6d4b\u7a7a\u95f4\u548c\u52a8\u4f5c\u7a7a\u95f4\u5bf9\u9f50\uff0c\u4e5f\u4e0d\u80fd\u4fdd\u8bc1\u7b56\u7565\u80fd\u7a33\u5b9a\u8fc1\u79fb\u3002<\/p>\n<h3>RL Fine-Tuning<\/h3>\n<ul>\n<li>\n<p>\u4fdd\u7559\u539f\u59cb\u7b56\u7565\u7f51\u7edc\u7ed3\u6784\u548c\u53c2\u6570<\/p>\n<\/li>\n<li>\n<p>\u628a\u7b56\u7565\u90e8\u7f72\u5230 G1 \u7684\u4eff\u771f\u73af\u5883\u4e2d<\/p>\n<\/li>\n<li>\n<p>\u4f7f\u7528\u4e0e\u539f\u6765\u76f8\u540c\u7684\u5f3a\u5316\u5b66\u4e60\u7b97\u6cd5\u7ee7\u7eed\u8bad\u7ec3\u7b56\u7565\uff0c\u4f18\u5316\u5728 G1 \u4e0a\u7684\u8868\u73b0<\/p>\n<\/li>\n<li>\n<p>\u5b66\u4e60\u7387\u8981\u4f4e\u4e00\u70b9\uff0c\u9632\u6b62\u201c\u5fd8\u8bb0\u201d\u539f\u6765\u7684\u6280\u80fd\u3002<\/p>\n<\/li>\n<li>\n<p>\u73af\u5883\u5956\u52b1\u51fd\u6570\u8981\u4fdd\u6301\u4e00\u81f4\u6216\u7a0d\u4f5c\u8c03\u6574\u4ee5\u9002\u914d G1\u3002<\/p>\n<\/li>\n<li>\n<p>\u53ef\u4ee5\u51bb\u7ed3\u90e8\u5206\u7f51\u7edc\u5c42\uff08\u5982\u524d\u51e0\u5c42\uff09\uff0c\u53ea\u8bad\u7ec3\u8f93\u51fa\u5c42\u3002<\/p>\n<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>\u5f00\u6e90\u4ee3\u7801\uff0c\u5b9e\u9a8c\u6570\u636e\uff0c\u89c6\u9891\u5c55\u793a: https:\/\/github.com\/sssn- [&#8230;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[31],"tags":[],"class_list":["post-856","post","type-post","status-publish","format-standard","hentry","category-31"],"_links":{"self":[{"href":"https:\/\/blog.sssn.tech\/index.php?rest_route=\/wp\/v2\/posts\/856","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/blog.sssn.tech\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/blog.sssn.tech\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/blog.sssn.tech\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/blog.sssn.tech\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=856"}],"version-history":[{"count":2,"href":"https:\/\/blog.sssn.tech\/index.php?rest_route=\/wp\/v2\/posts\/856\/revisions"}],"predecessor-version":[{"id":858,"href":"https:\/\/blog.sssn.tech\/index.php?rest_route=\/wp\/v2\/posts\/856\/revisions\/858"}],"wp:attachment":[{"href":"https:\/\/blog.sssn.tech\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=856"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/blog.sssn.tech\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=856"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/blog.sssn.tech\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=856"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}