{"id":107605,"date":"2026-09-19T21:40:34","date_gmt":"2026-09-19T13:40:34","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/107605.html"},"modified":"2026-09-19T21:40:34","modified_gmt":"2026-09-19T13:40:34","slug":"iace-%e6%b7%b1%e5%ba%a6%e8%a7%a3%e6%9e%90%ef%bc%9a%e5%8f%8c%e8%87%82%e6%9c%ba%e5%99%a8%e4%ba%ba%e8%af%a5%e6%80%8e%e4%b9%88%e5%88%86%e5%b7%a5%ef%bc%9faist-%e7%94%a8-14-%e7%bb%84","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/107605.html","title":{"rendered":"IACE \u6df1\u5ea6\u89e3\u6790\uff1a\u53cc\u81c2\u673a\u5668\u4eba\u8be5\u600e\u4e48\u201c\u5206\u5de5\u201c\uff1fAIST \u7528 14 \u7ec4\u771f\u673a\u5b9e\u9a8c\u7ed9\u51fa\u67b6\u6784\u9009\u578b\u6307\u5357"},"content":{"rendered":"

\u8bba\u6587\u6807\u9898: Learning Bimanual Manipulation via Action Chunking and Inter-Arm Coordination with Transformers
\n\u4f5c\u8005: Tomohiro Motoda, Ryo Hanai, Ryoichi Nakajo, Masaki Murooka, Floris Erich, Yukiyasu Domae
\n\u673a\u6784: \u4ea7\u4e1a\u6280\u672f\u7efc\u5408\u7814\u7a76\u6240 (AIST), \u65e5\u672c
\n\u53d1\u8868: arXiv:2503.13916 (v1, 2025-03-18)
\n\u57fa\u5ea7: ACT (Action Chunking with Transformers)
\n\u5173\u952e\u8bcd: \u53cc\u81c2\u64cd\u4f5c\u3001\u6a21\u4eff\u5b66\u4e60\u3001\u52a8\u4f5c\u5206\u5757\u3001Transformer\u3001\u81c2\u95f4\u534f\u8c03\u3001\u884c\u4e3a\u514b\u9686<\/p>\n

\n

\u4e00\u53e5\u8bdd\u603b\u7ed3：AIST \u5728 ACT \u57fa\u7840\u4e0a\u52a0\u4e86\u4e00\u4e2a\u81c2\u95f4\u534f\u8c03\u7f16\u7801\u5668（IACE），\u5e76\u628a"\u6bcf\u81c2\u72ec\u7acb\u7f16\u7801\u5668 + \u5355\/\u5206\u88c2\u89e3\u7801\u5668"\u7684\u67b6\u6784\u7ec4\u5408\u5728 8 \u4e2a\u771f\u673a\u53cc\u81c2\u4efb\u52a1\u4e0a\u505a\u4e86\u7cfb\u7edf\u6d88\u878d\u2014\u2014\u7ed3\u8bba\u6e05\u6670\u5f97\u50cf\u9009\u578b\u624b\u518c：\u5f02\u6b65\u4efb\u52a1（\u4e24\u624b\u5404\u5e72\u5404\u7684）\u7528\u5206\u88c2\u89e3\u7801\u5668，\u540c\u6b65\u4efb\u52a1（\u4e24\u624b\u4e00\u8d77\u5e72）\u7528\u5355\u89e3\u7801\u5668，\u800c IACE \u662f\u4e24\u6761\u8def\u7ebf\u7684\u5171\u540c\u5730\u57fa（\u53bb\u6389\u540e\u540c\u6b65\u4efb\u52a1\u6210\u529f\u7387\u4ece 98% \u8dcc\u5230 57%）\u3002<\/p>\n

\n

\u4e00\u3001\u4ecb\u7ecd<\/h3>\n

1.1 \u53cc\u81c2\u7684"1+1"\u96be\u9898<\/h4>\n

\u53cc\u81c2\u64cd\u4f5c\u662f\u673a\u5668\u4eba\u64cd\u4f5c\u9886\u57df\u7684\u786c\u9aa8\u5934，\u4e5f\u662f\u8fd9\u4e2a\u7cfb\u5217\u535a\u5ba2\u7684\u8001\u670b\u53cb\u4e86\u2014\u2014ALOHA\u3001BRS\u3001iDP3 \u91cc\u90fd\u7ed5\u4e0d\u5f00\u5b83\u3002\u4f46 AIST \u8fd9\u7bc7\u63d0\u51fa\u7684\u95ee\u9898\u66f4"\u5143"\u4e00\u5c42：\u53cc\u81c2\u7b56\u7565\u7f51\u7edc\u7684\u67b6\u6784\u672c\u8eab\u8be5\u600e\u4e48\u8bbe\u8ba1？<\/p>\n

\u76f4\u63a5\u7684\u505a\u6cd5\u662f\u628a\u53cc\u81c2\u81ea\u7531\u5ea6\u62fc\u8d77\u6765\u7aef\u5230\u7aef\u5b66（ACT \u7684\u505a\u6cd5），\u4f46\u4f5c\u8005\u6307\u51fa\u8fd9\u9690\u542b\u4e24\u4e2a\u95ee\u9898：<\/p>\n

  • \u6570\u636e\u5f02\u6784\u6027：\u628a\u4e24\u4e2a 6-DoF \u5355\u81c2（ViperX\u3001UR5）\u6216 7-DoF \u81c2（Franka）\u62fc\u6210 12\/14-DoF \u53cc\u81c2，\u786c\u4ef6\u6784\u578b\u4e00\u53d8，\u6a21\u578b\u5c31\u5f97\u5fae\u8c03\u751a\u81f3\u91cd\u8bad\u2014\u2014\u62fc\u7ef4\u5ea6\u7684\u505a\u6cd5\u5bf9\u786c\u4ef6\u5dee\u5f02\u96f6\u9c81\u68d2\u6027；<\/li>\n
  • \u5355\u81c2\u80fd\u529b\u88ab\u7a00\u91ca：\u53cc\u81c2\u4efb\u52a1\u91cc\u5de6\u53f3\u624b\u89d2\u8272\u5f80\u5f80\u4e0d\u540c\u2014\u2014\u201c\u53f3\u624b\u6293\u8d77\u7269\u4f53\u3001\u5de6\u624b\u63a5\u4f4f\u201d（hand-over）\u672c\u8d28\u4e0a\u662f\u4e24\u4e2a\u5355\u81c2\u6280\u80fd\u7684\u63a5\u529b\u3002\u7aef\u5230\u7aef flatten \u5b66\u4e60\u8ba9\u7f51\u7edc\u5f88\u96be"\u6df1\u5b66"\u6bcf\u53ea\u624b\u7684\u72ec\u7acb\u6280\u80fd\u3002<\/li>\n

    \u4f5c\u8005\u7684\u6838\u5fc3\u5047\u8bbe（hypothesis）\u5f88\u660e\u786e：\u6df1\u5ea6\u5355\u81c2\u5b66\u4e60\u662f\u5b66\u597d\u53cc\u81c2\u64cd\u4f5c\u7684\u524d\u63d0\u3002\u6240\u4ee5\u67b6\u6784\u4e0a\u5e94\u8be5\u7ed9\u6bcf\u53ea\u624b\u4e00\u6761\u72ec\u7acb\u7684\u4fe1\u606f\u901a\u8def（\u72ec\u7acb\u7f16\u7801\u5668），\u518d\u663e\u5f0f\u5730\u5efa\u4e00\u6761"\u534f\u8c03\u901a\u8def"（IACE）\u3002<\/p>\n

    1.2 \u4e0e\u524d\u4f5c\u4eec\u7684\u4f4d\u7f6e\u5173\u7cfb<\/h4>\n

    \u8fd9\u7bc7\u7684\u5b9a\u4f4d\u662f\u67b6\u6784\u7814\u7a76\u800c\u975e\u7b97\u6cd5\u521b\u65b0\u2014\u2014\u5b83\u7ad9\u5728 ACT \u7684\u80a9\u8180\u4e0a（CVAE + \u52a8\u4f5c\u5206\u5757 + \u65f6\u5e8f\u96c6\u6210\u5168\u90e8\u4fdd\u7559），\u7814\u7a76\u7684\u95ee\u9898\u662f"\u7f16\u7801\u5668\u548c\u89e3\u7801\u5668\u600e\u4e48\u6446"\u3002\u76f8\u5173\u7684\u540c\u7c7b\u5de5\u4f5c：<\/p>\n

      \n
    • INTERACT（Lee et al., CoRL 2024）：\u5c42\u7ea7\u6ce8\u610f\u529b Transformer \u540c\u6b65\u53cc\u81c2\u8f68\u8ff9\u2014\u2014\u601d\u8def\u540c\u6e90，\u4f46\u90a3\u662f\u5b8c\u6574\u7684\u65b0\u65b9\u6cd5，\u672c\u6587\u662f ACT \u4e0a\u7684\u6700\u5c0f\u4fb5\u5165\u5f0f\u6539\u9020 + \u7cfb\u7edf\u6d88\u878d；<\/li>\n
    • Shikada et al.（IROS 2024）：\u5c42\u7ea7\u6df1\u5ea6\u9884\u6d4b\u5b66\u4e60\u6a21\u578b\u505a\u53cc\u81c2\u540c\u6b65\u2014\u2014\u7528 RNN \u8def\u7ebf，\u672c\u6587\u8d70 Transformer；<\/li>\n
    • \u03c00\u3001RDT-1B：\u57fa\u7840\u6a21\u578b\u8def\u7ebf，\u5927\u6570\u636e\u5806\u6cdb\u5316\u2014\u2014\u672c\u6587\u53cd\u5176\u9053\u800c\u884c，\u7814\u7a76\u5c0f\u6570\u636e（\u6bcf\u4efb\u52a1 50 \u6761\u6f14\u793a）\u4e0b\u7684\u67b6\u6784\u6548\u7387\u3002<\/li>\n<\/ul>\n

      1.3 \u4efb\u52a1\u7684\u4e8c\u5206\u6cd5：\u5f02\u6b65 vs \u540c\u6b65<\/h4>\n

      \u5b9e\u9a8c\u8bbe\u8ba1\u6cbf\u7528\u4e86 Krebs & Asfour \u7684\u53cc\u81c2\u64cd\u4f5c\u5206\u7c7b\u6cd5（bimanual taxonomy），\u628a\u4efb\u52a1\u5207\u6210\u4e24\u7c7b：<\/p>\n

        \n
      • \u5f02\u6b65\u4efb\u52a1（Asynchronous）：\u5de6\u53f3\u81c2\u52a8\u4f5c\u76f8\u4e92\u72ec\u7acb\u3001\u5148\u540e\u6267\u884c\u3002\u4f8b：\u6298\u6bdb\u5dfe（\u5148\u62ff\u518d\u653e）\u3001\u9524\u51fb（\u5148\u6293\u9524\u518d\u6325）\u3001\u4ea4\u63a5（\u4e00\u53ea\u624b\u9012\u4e00\u53ea\u624b\u63a5）\u2014\u20144 \u9879；<\/li>\n
      • \u540c\u6b65\u4efb\u52a1（Synchronous）：\u53cc\u81c2\u540c\u65f6\u3001\u534f\u8c03\u5730\u52a8\u3002\u4f8b：\u53cc\u4eba\u62ac\u6bdb\u5dfe（\u5fc5\u987b\u540c\u65f6\u62ac\u5426\u5219\u6389）\u3001\u5f00\u5de5\u5177\u7bb1（\u4e00\u624b\u6309\u7bb1\u4f53\u4e00\u624b\u5f00\u76d6）\u3001\u62ac\u677f（\u4e24\u70b9\u540c\u65f6\u53d1\u529b\u5426\u5219\u677f\u7ffb）\u2014\u20144 \u9879\u3002<\/li>\n<\/ul>\n

        \u8fd9\u4e2a\u4e8c\u5206\u6cd5\u76f4\u63a5\u51b3\u5b9a\u4e86\u540e\u9762\u7684\u6838\u5fc3\u7ed3\u8bba：\u4e24\u7c7b\u4efb\u52a1\u7684\u6700\u4f18\u89e3\u7801\u5668\u67b6\u6784\u662f\u76f8\u53cd\u7684\u3002\u8fd9\u662f\u5168\u6587\u6700\u6709\u4ef7\u503c\u7684\u53d1\u73b0\u3002<\/p>\n

        \n

        \u4e8c\u3001\u539f\u7406<\/h3>\n

        2.1 \u57fa\u5ea7：ACT \u7684\u5feb\u901f\u56de\u987e<\/h4>\n

        ACT（Action Chunking with Transformers）\u7684\u8981\u7d20\u5168\u90e8\u4fdd\u7559：<\/p>\n

          \n
        • \u89c2\u6d4b\u7f16\u7801：4 \u8def RGB \u76f8\u673a（\u5de6\u53f3\u8155 + \u4e0a\u65b9 + \u4e0b\u65b9\u4fef\u89c6）\u2192 ResNet18 \u2192 \u89c6\u89c9\u7279\u5f81 token；\u53cc\u81c2\u5173\u8282\u4f4d\u7f6e\u5e8f\u5217 \u2192 \u62fc\u8fdb\u8f93\u5165\u5e8f\u5217；<\/li>\n
        • CVAE：\u6761\u4ef6\u53d8\u5206\u81ea\u7f16\u7801\u5668\u5efa\u6a21\u4eba\u7c7b\u793a\u6559\u7684\u98ce\u683c\u53d8\u5316（\u540c\u4e00\u4e2a\u4eba\u505a 50 \u6b21\u6298\u6bdb\u5dfe，\u6bcf\u6b21\u8f68\u8ff9\u7ec6\u8282\u90fd\u4e0d\u540c，CVAE \u7528\u9690\u53d8\u91cf z \u5438\u6536\u8fd9\u79cd\u591a\u6837\u6027）；<\/li>\n
        • \u52a8\u4f5c\u5206\u5757（action chunking）：\u4e0d\u9010\u6b65\u9884\u6d4b，\u4e00\u6b21\u9884\u6d4b\u672a\u6765\u4e00\u6bb5\u52a8\u4f5c\u5e8f\u5217；<\/li>\n
        • \u65f6\u5e8f\u96c6\u6210（temporal ensemble）：\u63a8\u7406\u65f6\u5bf9\u8fc7\u53bb\u591a\u4e2a\u65f6\u95f4\u6b65\u7684\u9884\u6d4b\u6309\u6307\u6570\u6743\u91cd\u878d\u5408，\u5e73\u6ed1\u8f93\u51fa\u3002<\/li>\n<\/ul>\n

          2.2 \u67b6\u6784：\u4e09\u6bb5\u5f0f\u4fe1\u606f\u6d41<\/h4>\n

          \u672c\u6587\u7684\u6539\u9020\u96c6\u4e2d\u5728\u7f16\u7801\u5668-\u89e3\u7801\u5668\u7684\u7ec4\u7ec7\u65b9\u5f0f\u4e0a，\u5f62\u6210\u4e09\u6bb5\u5f0f\u7ed3\u6784（\u5bf9\u5e94\u8bba\u6587 Figure 1）：<\/p>\n

          \u8f93\u5165\u5e8f\u5217\u7684\u4e09\u4e2a segment（\u5404\u5e26 CLS token + \u4f4d\u7f6e\u7f16\u7801）:
          \n \u2460 \u5de6\u81c2 segment: \u5de6\u81c2\u5173\u8282\u5e8f\u5217 + \u5de6\u8155\u76f8\u673a + \u5168\u5c40\u76f8\u673a\u7279\u5f81
          \n \u2461 \u53f3\u81c2 segment: \u53f3\u81c2\u5173\u8282\u5e8f\u5217 + \u53f3\u8155\u76f8\u673a + \u5168\u5c40\u76f8\u673a\u7279\u5f81
          \n \u2462 \u5168\u5c40 segment: \u53cc\u81c2\u5173\u8282 + \u5168\u90e8\u76f8\u673a\u7279\u5f81<\/p>\n

          \u7b2c\u4e00\u5c42：\u53cc\u81c2\u5c40\u90e8\u7f16\u7801\u5668\u3002 \u6bcf\u53ea\u624b\u81c2\u4e00\u4e2a\u72ec\u7acb\u7684 Transformer encoder，\u53ea\u5403\u81ea\u5df1\u7684\u5173\u8282 + \u81ea\u5df1\u7684\u8155\u90e8\u76f8\u673a（+ \u5168\u5c40\u89c6\u89c9\u7279\u5f81）\u3002\u8fd9\u5c31\u662f"\u6df1\u5b66\u5355\u81c2\u6280\u80fd"\u7684\u67b6\u6784\u5316\u4f53\u73b0\u2014\u2014\u5de6\u81c2\u7f16\u7801\u5668\u4e0d\u9700\u8981\u7406\u89e3\u53f3\u81c2\u7684\u8155\u90e8\u89c6\u91ce，\u4e13\u6ce8\u628a\u81ea\u5df1\u7684\u8f68\u8ff9\u5b66\u624e\u5b9e\u3002<\/p>\n

          \u7b2c\u4e8c\u5c42：IACE（Inter-Arm Coordinated Transformer Encoder）\u3002 \u5168\u5c40\u7684\u534f\u8c03\u7f16\u7801\u5668，\u540c\u65f6\u5403\u53cc\u81c2\u72b6\u6001 + \u6355\u6349\u6574\u4e2a\u573a\u666f\u7684\u89c6\u89c9\u4fe1\u606f，\u8d1f\u8d23\u5efa\u6a21\u5f0f\u540c\u6b65\u548c\u65f6\u5e8f\u5bf9\u9f50\u2014\u2014\u201c\u5de6\u624b\u4ec0\u4e48\u65f6\u5019\u8be5\u52a8\u3001\u52a8\u591a\u5feb，\u53d6\u51b3\u4e8e\u53f3\u624b\u73b0\u5728\u5728\u54ea\u201d\u3002\u5b83\u662f\u72ec\u7acb\u4e8e\u4e24\u4e2a\u5c40\u90e8\u7f16\u7801\u5668\u7684\u4e00\u6761\u663e\u5f0f\u534f\u8c03\u901a\u8def，\u8f93\u51fa\u534f\u8c03 token \u4f9b\u89e3\u7801\u5668\u4f7f\u7528\u3002<\/p>\n

          \u7b2c\u4e09\u5c42：\u89e3\u7801\u5668\u7684\u4e24\u79cd\u6446\u6cd5（\u6d88\u878d\u6838\u5fc3）\u3002<\/p>\n

            \n
          • \u5355\u89e3\u7801\u5668（Single Decoder）：\u628a\u5de6\u53f3\u81c2\u7f16\u7801\u5668\u8f93\u51fa + IACE \u8f93\u51fa\u62fc\u63a5\u540e\u5582\u7ed9\u4e00\u4e2a\u5171\u4eab\u7684 Transformer decoder，\u4e00\u6b21\u6027\u8f93\u51fa\u53cc\u81c2\u52a8\u4f5c\u5757（batch output）\u3002\u4fe1\u606f\u5168\u5c40\u5171\u4eab，\u9002\u5408\u9700\u8981\u7d27\u5bc6\u8026\u5408\u7684\u4efb\u52a1；<\/li>\n
          • \u5206\u88c2\u89e3\u7801\u5668（Split Decoders）：\u4e24\u4e2a\u72ec\u7acb decoder，\u5404\u81ea\u53ea\u5403\u81ea\u5df1\u81c2\u7684\u7f16\u7801\u5668\u8f93\u51fa + IACE \u8f93\u51fa，\u5206\u522b\u8f93\u51fa\u5de6\/\u53f3\u81c2\u52a8\u4f5c\u5757\u3002\u5404\u7ba1\u5404\u7684，IACE \u4f5c\u4e3a\u552f\u4e00\u7684\u534f\u8c03\u6865\u6881，\u9002\u5408\u4e24\u81c2\u4efb\u52a1\u89e3\u8026\u7684\u573a\u666f\u3002<\/li>\n<\/ul>\n

            \u76f4\u89c9\u7c7b\u6bd4（\u7f16\u7801\u89c6\u89d2）：\u5355\u89e3\u7801\u5668\u50cf\u8054\u5408\u7f16\u7801\u2014\u2014\u628a\u5de6\u53f3\u4e24\u8def\u4fe1\u53f7\u5f53\u4e00\u4e2a\u5bf9\u8c61\u6574\u4f53\u538b\u7f29，\u5197\u4f59\u4fe1\u606f\u5171\u4eab，\u540c\u6b65\u6027\u597d\u4f46\u7075\u6d3b\u6027\u5dee；\u5206\u88c2\u89e3\u7801\u5668\u50cf\u72ec\u7acb\u7f16\u7801 + \u4e00\u6761\u516c\u5171\u53c2\u8003\u6d41\u2014\u2014\u4e24\u8def\u5404\u81ea\u6700\u4f18，\u9760 IACE \u8fd9\u6761"\u53c2\u8003\u6d41"\u5bf9\u9f50\u65f6\u5e8f\u3002\u54ea\u4e2a\u597d\u53d6\u51b3\u4e8e\u6e90\u4fe1\u53f7\u7684\u8026\u5408\u5ea6\u2014\u2014\u8fd9\u6070\u597d\u5c31\u662f\u5f02\u6b65\/\u540c\u6b65\u4efb\u52a1\u7684\u5206\u91ce\u3002<\/p>\n

            2.3 \u8bad\u7ec3\u4e0e\u63a8\u7406\u7ec6\u8282<\/h4>\n
              \n
            • \u6570\u636e：ALOHA \u9065\u64cd\u4f5c\u7cfb\u7edf\u91c7\u96c6（\u64cd\u4f5c\u5458\u6162\u901f\u64cd\u4f5c\u63d0\u7cbe\u5ea6），\u5173\u8282\u4f4d\u7f6e + RGB \u56fe\u50cf，\u539f\u59cb 50 Hz；<\/li>\n
            • 50 Hz \u2192 25 Hz \u964d\u91c7\u6837：\u4f5c\u8005\u53d1\u73b0\u539f\u59cb\u5e27\u7387\u4e0b\u5173\u8282\u503c\u7684\u5fae\u5c0f\u6296\u52a8\u5bfc\u81f4\u8bad\u7ec3\u8f93\u51fa\u4e0d\u7a33\u5b9a，\u964d\u91c7\u6837\u5230 25 Hz \u540e\u8bad\u7ec3\u7a33\u5b9a\u2014\u2014\u52a8\u4f5c\u5206\u5757\u6a21\u578b\u5bf9\u9ad8\u9891\u9065\u64cd\u4f5c\u566a\u58f0\u654f\u611f，\u8fd9\u662f\u4e2a\u5b9e\u7528\u7684\u7ecf\u9a8c\u503c；<\/li>\n
            • \u6bcf\u4efb\u52a1 50 \u6761\u6f14\u793a；AdamW，\u03bb=1e-5，5000 epochs，batch 8，\u5355\u5361 RTX 4090（24GB）；<\/li>\n
            • \u63a8\u7406\u8017\u65f6\u7ea6 0.02 s，\u52a8\u4f5c\u5206\u5757 + \u65f6\u5e8f\u96c6\u6210\u8f93\u51fa\u5e73\u6ed1\u8f68\u8ff9\u3002<\/li>\n<\/ul>\n
              \n

              \u4e09\u3001\u516c\u5f0f<\/h3>\n

              \u8fd9\u7bc7\u8bba\u6587\u6b63\u6587\u6ca1\u6709\u516c\u5f0f（6 \u9875\u77ed\u6587，\u67b6\u6784\u4ee5\u56fe\u4ee3\u6570），\u6838\u5fc3\u6570\u5b66\u90fd\u7ee7\u627f\u81ea ACT\u3002\u4e3a\u5b8c\u6574\u6027，\u628a\u88ab\u7ee7\u627f\u7684\u5173\u952e\u516c\u5f0f\u5217\u51fa，\u5e76\u6807\u6ce8 IACE \u6539\u52a8\u7684\u843d\u70b9\u3002<\/p>\n

              3.1 CVAE \u7684\u53d8\u5206\u4e0b\u754c<\/h4>\n

              ACT \u7528 CVAE \u5efa\u6a21\u793a\u6559\u98ce\u683c\u591a\u6837\u6027\u3002\u53d8\u5206\u81ea\u7f16\u7801\u5668\u7684\u76ee\u6807（ELBO）：<\/p>\n

              LELBO=Ez\u223cq\u03d5(z\u2223o,y)[log\u2061p\u03b8(y\u2223z,o)]\u2212DKL(q\u03d5(z\u2223o,y)\u2009\u2225\u2009p(z))\\\\mathcal{L}_{\\\\text{ELBO}} = \\\\mathbb{E}_{z \\\\sim q_\\\\phi(z|o, y)} \\\\left[ \\\\log p_\\\\theta(y | z, o) \\\\right] – D_{\\\\text{KL}}\\\\left( q_\\\\phi(z | o, y) \\\\,\\\\|\\\\, p(z) \\\\right)<\/span><\/span>L<\/span><\/span>ELBO<\/span><\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>=<\/span><\/span><\/span><\/span>E<\/span><\/span>z<\/span>\u223c<\/span>q<\/span><\/span>\u03d5<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>(<\/span>z<\/span>\u2223<\/span>o<\/span>,<\/span>y<\/span>)<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>[<\/span>log<\/span><\/span><\/span>p<\/span><\/span>\u03b8<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>(<\/span>y<\/span>\u2223<\/span>z<\/span>,<\/span><\/span>o<\/span>)<\/span>]<\/span><\/span><\/span>\u2212<\/span><\/span><\/span><\/span>D<\/span><\/span>KL<\/span><\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>(<\/span>q<\/span><\/span>\u03d5<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>(<\/span>z<\/span>\u2223<\/span>o<\/span>,<\/span><\/span>y<\/span>)<\/span><\/span>\u2225<\/span><\/span>p<\/span>(<\/span>z<\/span>)<\/span>)<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n

                \n
              • yy<\/span><\/span>y<\/span><\/span><\/span><\/span><\/span>：\u76ee\u6807\u52a8\u4f5c\u5757（\u672a\u6765\u591a\u6b65\u53cc\u81c2\u5173\u8282\u4f4d\u7f6e\u5e8f\u5217）；<\/li>\n
              • oo<\/span><\/span>o<\/span><\/span><\/span><\/span><\/span>：\u89c2\u6d4b（\u56fe\u50cf\u7279\u5f81 + \u5173\u8282\u4f4d\u7f6e）；<\/li>\n
              • zz<\/span><\/span>z<\/span><\/span><\/span><\/span><\/span>：\u98ce\u683c\u9690\u53d8\u91cf，\u7f16\u7801\u5668 q\u03d5q_\\\\phi<\/span><\/span>q<\/span><\/span>\u03d5<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> \u4ee5 (\u89c2\u6d4b, \u76ee\u6807) \u4e3a\u6761\u4ef6\u2014\u2014\u6ce8\u610f z \u7f16\u7801\u5668\u770b\u5f97\u89c1\u771f\u503c\u52a8\u4f5c，\u6240\u4ee5\u5b66\u5230\u7684\u662f"\u8fd9\u6761\u8f68\u8ff9\u7684\u98ce\u683c"；\u91c7\u6837\u9636\u6bb5 zz<\/span><\/span>z<\/span><\/span><\/span><\/span><\/span> \u4ece\u5148\u9a8c p(z)\u223cN(0,I)p(z) \\\\sim \\\\mathcal{N}(0,I)<\/span><\/span>p<\/span>(<\/span>z<\/span>)<\/span><\/span>\u223c<\/span><\/span><\/span><\/span>N<\/span>(<\/span>0<\/span>,<\/span><\/span>I<\/span>)<\/span><\/span><\/span><\/span><\/span> \u62bd\u53d6；<\/li>\n
              • \u7b2c\u4e00\u9879\u662f\u91cd\u5efa\u9879，\u7b2c\u4e8c\u9879 KL \u7ea6\u675f\u628a\u98ce\u683c\u5206\u5e03\u62c9\u5411\u6807\u51c6\u6b63\u6001\u3002<\/li>\n<\/ul>\n

                IACE \u4e0d\u6539\u8fd9\u4e00\u5c42\u2014\u2014\u98ce\u683c\u5efa\u6a21\u4e0e\u534f\u8c03\u5efa\u6a21\u89e3\u8026\u3002<\/p>\n

                3.2 \u8f93\u5165\u5e8f\u5217\u7ec4\u88c5（IACE \u843d\u70b9）<\/h4>\n

                \u4e09\u6bb5\u5e8f\u5217\u7684\u5f62\u5f0f\u5316（\u4ee5\u5355\u89e3\u7801\u5668\u4e3a\u4f8b）：<\/p>\n

                input=[CLSL,s1L,\u2026,skL]\u23df\u5de6\u81c2\u6bb5\u2295[CLSR,s1R,\u2026,skR]\u23df\u53f3\u81c2\u6bb5\u2295[CLSG,g1,\u2026,gm]\u23df\u5168\u5c40\u6bb5\u2295PE\\\\text{input} = \\\\underbrace{[\\\\text{CLS}_L, s^L_1, \\\\dots, s^L_k]}_{\\\\text{\u5de6\u81c2\u6bb5}} \\\\oplus \\\\underbrace{[\\\\text{CLS}_R, s^R_1, \\\\dots, s^R_k]}_{\\\\text{\u53f3\u81c2\u6bb5}} \\\\oplus \\\\underbrace{[\\\\text{CLS}_G, g_1, \\\\dots, g_m]}_{\\\\text{\u5168\u5c40\u6bb5}} \\\\oplus \\\\text{PE}<\/span><\/span>input<\/span><\/span><\/span>=<\/span><\/span><\/span><\/span><\/span>\u5de6\u81c2\u6bb5<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>[<\/span>CLS<\/span><\/span><\/span>L<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>,<\/span><\/span>s<\/span><\/span>1<\/span><\/span><\/span><\/span>L<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>,<\/span><\/span>\u2026<\/span><\/span>,<\/span><\/span>s<\/span><\/span>k<\/span><\/span><\/span><\/span>L<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>]<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u2295<\/span><\/span><\/span><\/span><\/span>\u53f3\u81c2\u6bb5<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>[<\/span>CLS<\/span><\/span><\/span>R<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>,<\/span><\/span>s<\/span><\/span>1<\/span><\/span><\/span><\/span>R<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>,<\/span><\/span>\u2026<\/span><\/span>,<\/span><\/span>s<\/span><\/span>k<\/span><\/span><\/span><\/span>R<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>]<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u2295<\/span><\/span><\/span><\/span><\/span>\u5168\u5c40\u6bb5<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>[<\/span>CLS<\/span><\/span><\/span>G<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>,<\/span><\/span>g<\/span><\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>,<\/span><\/span>\u2026<\/span><\/span>,<\/span><\/span>g<\/span><\/span>m<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>]<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u2295<\/span><\/span><\/span><\/span>PE<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n

                  \n
                • siL,siRs^L_i, s^R_i<\/span><\/span>s<\/span><\/span>i<\/span><\/span><\/span><\/span>L<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>,<\/span><\/span>s<\/span><\/span>i<\/span><\/span><\/span><\/span>R<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>：\u5de6\/\u53f3\u81c2\u7684\u5173\u8282\u503c\u4e0e\u5bf9\u5e94\u8155\u90e8\u76f8\u673a ResNet18 \u7279\u5f81 token；<\/li>\n
                • gjg_j<\/span><\/span>g<\/span><\/span>j<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>：\u53cc\u81c2\u5173\u8282 + \u5168\u90e8\u76f8\u673a\u7279\u5f81；<\/li>\n
                • PE \u4e3a\u4f4d\u7f6e\u7f16\u7801；\u6bcf\u6bb5\u5f00\u5934\u7684 CLS token \u6c47\u805a\u672c\u6bb5\u4fe1\u606f（\u8bba\u6587\u5f15\u7528 BERT \u7684 CLS \u673a\u5236）\u3002<\/li>\n<\/ul>\n

                  IACE \u7684\u6539\u52a8：\u5728\u5c40\u90e8\u7f16\u7801\u5668\u4e4b\u540e\u63d2\u5165\u4e00\u4e2a\u5168\u5c40\u534f\u8c03\u7f16\u7801\u5668，\u5176\u81ea\u6ce8\u610f\u529b\u4f5c\u7528\u4e8e\u62fc\u63a5\u540e\u7684\u53cc\u81c2 + \u5168\u5c40 token：<\/p>\n

                  Hcoord=TransformerEncIACE([EL;ER;EG])H_{\\\\text{coord}} = \\\\text{TransformerEnc}_{\\\\text{IACE}}\\\\left( [E_L; E_R; E_G] \\\\right)<\/span><\/span>H<\/span><\/span>coord<\/span><\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>=<\/span><\/span><\/span><\/span>TransformerEnc<\/span><\/span><\/span>IACE<\/span><\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>(<\/span>[<\/span>E<\/span><\/span>L<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>;<\/span><\/span>E<\/span><\/span>R<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>;<\/span><\/span>E<\/span><\/span>G<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>]<\/span>)<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n

                  \u5176\u4e2d EL,ERE_L, E_R<\/span><\/span>E<\/span><\/span>L<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>,<\/span><\/span>E<\/span><\/span>R<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> \u4e3a\u5de6\u53f3\u5c40\u90e8\u7f16\u7801\u5668\u8f93\u51fa，EGE_G<\/span><\/span>E<\/span><\/span>G<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> \u4e3a\u5168\u5c40\u6bb5\u7f16\u7801\u3002\u6b63\u662f\u8fd9\u4e00\u6b65\u8ba9"\u53f3\u624b\u72b6\u6001"\u6210\u4e3a"\u5de6\u624b\u51b3\u7b56"\u7684\u663e\u5f0f\u4e0a\u4e0b\u6587\u3002<\/p>\n

                  3.3 \u52a8\u4f5c\u5206\u5757\u4e0e\u65f6\u5e8f\u96c6\u6210<\/h4>\n

                  \u89e3\u7801\u5668（query \u4f4d\u7f6e\u5d4c\u5165\u4e3a\u8f93\u5165）\u8f93\u51fa\u672a\u6765 kk<\/span><\/span>k<\/span><\/span><\/span><\/span><\/span> \u6b65\u52a8\u4f5c\u5757：<\/p>\n

                  at:t+k=TransformerDec(query=Paction,\u00a0key\/value=[Hcoord;EL;ER])a_{t:t+k} = \\\\text{TransformerDec}(\\\\text{query}=P_{\\\\text{action}},\\\\ \\\\text{key\/value}=[H_{\\\\text{coord}}; E_L; E_R])<\/span><\/span>a<\/span><\/span>t<\/span>:<\/span>t<\/span>+<\/span>k<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>=<\/span><\/span><\/span><\/span>TransformerDec<\/span><\/span>(<\/span>query<\/span><\/span><\/span>=<\/span><\/span><\/span><\/span>P<\/span><\/span>action<\/span><\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>,<\/span>\u00a0<\/span><\/span>key\/value<\/span><\/span><\/span>=<\/span><\/span><\/span><\/span>[<\/span>H<\/span><\/span>coord<\/span><\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>;<\/span><\/span>E<\/span><\/span>L<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>;<\/span><\/span>E<\/span><\/span>R<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>])<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n

                  \u5355\u89e3\u7801\u5668\u4e00\u6b21\u51fa\u53cc\u81c2\u8054\u5408\u5757（2\u00d7ndof2 \\\\times n_{\\\\text{dof}}<\/span><\/span>2<\/span><\/span>\u00d7<\/span><\/span><\/span><\/span>n<\/span><\/span>dof<\/span><\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> \u7ef4）；\u5206\u88c2\u89e3\u7801\u5668\u51fa\u4e24\u4e2a\u72ec\u7acb\u5757\u3002\u63a8\u7406\u65f6\u5bf9\u65f6\u95f4\u4e0a\u91cd\u53e0\u7684\u9884\u6d4b\u505a\u6307\u6570\u52a0\u6743\u878d\u5408：<\/p>\n

                  a^t=\u2211iwi\u2009at(t\u2212i)\u2211iwi,wi=exp\u2061(\u2212\u03bbi)\\\\hat{a}_t = \\\\frac{\\\\sum_{i} w_i \\\\, a_t^{(t-i)}}{\\\\sum_i w_i}, \\\\quad w_i = \\\\exp(-\\\\lambda i)<\/span><\/span><\/span>a<\/span><\/span><\/span>^<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>t<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>=<\/span><\/span><\/span><\/span><\/span><\/span>\u2211<\/span><\/span>i<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>w<\/span><\/span>i<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u2211<\/span><\/span>i<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>w<\/span><\/span>i<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>a<\/span><\/span>t<\/span><\/span><\/span><\/span>(<\/span>t<\/span>\u2212<\/span>i<\/span>)<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>,<\/span><\/span><\/span>w<\/span><\/span>i<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>=<\/span><\/span><\/span><\/span>exp<\/span>(<\/span>\u2212<\/span>\u03bbi<\/span>)<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n

                  \u8f83\u65b0\u7684\u9884\u6d4b\u6743\u91cd\u66f4\u9ad8，\u65e7\u9884\u6d4b\u9010\u6e10\u5931\u6548\u2014\u2014\u8fd9\u5c42\u96c6\u6210\u62b9\u5e73\u4e86\u5206\u5757\u5207\u6362\u65f6\u7684\u8fb9\u754c\u6296\u52a8\u3002<\/p>\n

                  \n

                  \u56db\u3001\u56fe\u793a<\/h3>\n

                  4.1 \u4e94\u79cd\u88ab\u6d4b\u67b6\u6784（\u5bf9\u5e94\u8bba\u6587 Figure 1 & Figure 5）<\/h4>\n

                  (a) Ours \u5355\u89e3\u7801\u5668 (b) Ours \u5206\u88c2\u89e3\u7801\u5668
                  \n \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510
                  \n \u2502 \u5de6\u81c2Enc\u2510 \u2502 \u2502 \u5de6\u81c2Enc \u2500\u2510 \u2502
                  \n \u2502 \u53f3\u81c2Enc\u253c\u2500\u25b6 IACE \u2500\u2500\u253c\u2500\u25b6 \u5355Dec \u2502 \u53f3\u81c2Enc \u2500\u253c\u2500\u25b6IACE \u2502
                  \n \u2502 \u5168\u5c40Enc\u2518 \u2502 (\u62fc\u8f93\u5165) \u2502 \u5168\u5c40Enc \u2500\u2518 \u2502 \u2502
                  \n \u2502 \u2502 \u2502 \u250c\u2500\u2500\u2500\u2500\u2500\u2534\u2500\u2500\u2500\u2500\u2510 \u2502
                  \n \u2502 \u8f93\u51fa: \u53cc\u81c2\u8054\u5408\u52a8\u4f5c\u5757 \u2502 \u2502 \u5de6Dec\u25c0\u2500\u2524 \u2502 \u2502
                  \n \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2502 \u53f3Dec\u25c0\u2500\u2518 (\u5404\u81ea+IACE) \u2502
                  \n \u2502 \u8f93\u51fa: \u5de6\u53f3\u72ec\u7acb\u52a8\u4f5c\u5757 \u2502
                  \n \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518<\/p>\n

                  (c)(d) \u53bb\u6389 IACE \u7684\u5bf9\u5e94\u6d88\u878d\u7248 (e) ACT \u57fa\u7ebf (\u65e0\u5206\u81c2Enc, \u65e0IACE)<\/p>\n

                  4.2 \u4efb\u52a1\u4e8c\u5206\u6cd5\u4e0e\u67b6\u6784\u9009\u578b\u7ed3\u8bba<\/h4>\n

                  \u5f02\u6b65\u4efb\u52a1 (\u4e24\u624b\u5404\u5e72\u5404\u7684) \u540c\u6b65\u4efb\u52a1 (\u4e24\u624b\u5fc5\u987b\u914d\u5408)
                  \n \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510
                  \n \u2502 Fold Towel \u6298\u6bdb\u5dfe \u2502 \u2502 Lift & Place Towel \u62ac\u653e\u6bdb\u5dfe\u2502
                  \n \u2502 Hand-over CAN \u9012\u7f50\u5b50 \u2502 \u2502 Lift Board \u62ac\u677f \u2502
                  \n \u2502 Hit with Hammer \u6325\u9524 \u2502 \u2502 Open Toolbox \u5f00\u5de5\u5177\u7bb1 \u2502
                  \n \u2502 Hand-over Silver Bag \u9012\u5305 \u2502 \u2502 Open Lid \u5f00\u76d6 \u2502
                  \n \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518
                  \n \u5e73\u5747 89.3% \u2705 \u5206\u88c2\u89e3\u7801\u5668 \u5e73\u5747 98.4% \u2705 \u5355\u89e3\u7801\u5668
                  \n (\u5355\u89e3\u7801\u5668 85.8%) (\u5206\u88c2\u89e3\u7801\u5668 90.4%)
                  \n \u2514\u2500\u2500\u2500\u2500\u2500\u2500 \u5171\u540c\u5730\u57fa: IACE \u2500\u2500\u2500\u2500\u2500\u2500\u2518
                  \n \u53bb\u6389 IACE: \u5f02\u6b65 89\u219275\/60, \u540c\u6b65 98\u219257 (\u707e\u96be\u6027\u584c\u65b9)<\/p>\n

                  4.3 \u5b9e\u9a8c\u914d\u7f6e\u901f\u67e5\u8868<\/h4>\n\n\u9879\u76ee\u914d\u7f6e<\/tr>\n\n\n\n\n\n\n\n\n\n
                  \u673a\u5668\u4eba<\/td>\n2 \u00d7 ViperX 300S（ALOHA \u6539\u88c5：\u8c03\u76f8\u673a\u9ad8\u5ea6\u3001\u52a0\u9632\u62a4\u5e03）<\/td>\n<\/tr>\n
                  \u76f8\u673a<\/td>\n4 \u8def RGB：\u5de6\/\u53f3\u8155 + \u4e0a\u65b9\u4fef\u89c6 + \u4e0b\u65b9\u6c34\u5e73\u89c6<\/td>\n<\/tr>\n
                  \u9065\u64cd\u4f5c<\/td>\nALOHA \u4e3b\u4ece\u81c2，50 Hz \u91c7\u96c6 \u2192 25 Hz \u8bad\u7ec3<\/td>\n<\/tr>\n
                  \u6570\u636e\u91cf<\/td>\n\u6bcf\u4efb\u52a1 50 \u6761\u6f14\u793a \u00d7 8 \u4efb\u52a1<\/td>\n<\/tr>\n
                  \u4f18\u5316<\/td>\nAdamW，\u03bb=1e-5，5000 epochs，batch 8<\/td>\n<\/tr>\n
                  \u786c\u4ef6<\/td>\nCore i9 \/ 128GB RAM \/ RTX 4090 24GB<\/td>\n<\/tr>\n
                  \u63a8\u7406\u5ef6\u8fdf<\/td>\n~0.02 s<\/td>\n<\/tr>\n
                  \u8bc4\u6d4b<\/td>\n\u6bcf\u4efb\u52a1 25 episodes<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
                  \n

                  \u4e94\u3001\u8e29\u5751\u70b9<\/h3>\n

                  \u8303\u5f0f\u7ea7<\/h4>\n

                  \u5751 1：flatten \u53cc\u81c2\u81ea\u7531\u5ea6 \u2260 \u5b66\u4f1a\u53cc\u81c2\u3002
                  \n\u8fd9\u662f\u5168\u6587\u7684\u4e2d\u5fc3\u8bba\u70b9：\u628a 21 \u7ef4\u52a8\u4f5c\u5411\u91cf（\u6216\u4efb\u610f\u53cc\u81c2\u62fc\u63a5）\u76f4\u63a5\u7aef\u5230\u7aef\u5b66，\u7f51\u7edc\u6ca1\u6709\u7ed3\u6784\u6027\u52a8\u529b\u53bb\u5206\u522b\u638c\u63e1\u6bcf\u53ea\u624b\u7684\u6280\u80fd，\u4e5f\u6ca1\u6709\u663e\u5f0f\u901a\u9053\u53bb\u5efa\u6a21\u5de6\u53f3\u4e92\u4f9d\u8d56\u3002\u6d88\u878d\u6570\u636e\u652f\u6491\u5f97\u5f88\u786c：\u53bb\u6389 IACE \u540e，\u540c\u6b65\u4efb\u52a1\u5e73\u5747\u6210\u529f\u7387\u4ece 98.4%\/90.4% \u6389\u5230 56.8%\/57.6%\u2014\u2014\u6389\u6389 40 \u4e2a\u767e\u5206\u70b9，\u63a5\u8fd1\u968f\u673a\u5931\u8d25\u3002\u534f\u8c03\u4e0d\u662f"\u5e0c\u671b\u80fd\u5b66\u5230"，\u800c\u662f\u8981\u7ed9\u4e00\u6761\u4e13\u95e8\u7684\u901a\u8def\u3002<\/p>\n

                  \u5751 2：\u89e3\u7801\u5668\u67b6\u6784\u6ca1\u6709\u4e07\u80fd\u7b54\u6848\u2014\u2014\u4efb\u52a1\u8026\u5408\u5ea6\u51b3\u5b9a\u4e00\u5207\u3002
                  \n\u5206\u88c2\u89e3\u7801\u5668\u5728\u5f02\u6b65\u4efb\u52a1\u8d62（89.3% vs 85.8%），\u5355\u89e3\u7801\u5668\u5728\u540c\u6b65\u4efb\u52a1\u8d62（98.4% vs 90.4%）\u3002\u539f\u56e0\u76f4\u767d：\u5f02\u6b65\u4efb\u52a1\u91cc\u4e24\u81c2\u76ee\u6807\u76f8\u4e92\u72ec\u7acb，\u5206\u88c2\u89e3\u7801\u907f\u514d\u4e86"\u53e6\u4e00\u53ea\u624b\u7684\u566a\u58f0\u68af\u5ea6"\u6c61\u67d3\u672c\u81c2\u52a8\u4f5c；\u540c\u6b65\u4efb\u52a1\u91cc\u4e24\u81c2\u52a8\u4f5c\u672c\u8d28\u662f\u4e00\u4e2a\u6574\u4f53（\u62ac\u6bdb\u5dfe\u65f6\u5de6\u53f3\u5fc5\u987b\u955c\u50cf\u5bf9\u79f0），\u5206\u88c2\u89e3\u7801\u5207\u65ad\u4e86\u4f4e\u5c42\u4fe1\u606f\u5171\u4eab，\u5168\u9760 IACE \u5355\u70b9\u534f\u8c03，\u53cd\u800c\u5403\u4e8f\u3002\u505a\u67b6\u6784\u9009\u578b\u524d\u5148\u95ee\u81ea\u5df1\u7684\u4efb\u52a1\u5c5e\u4e8e\u54ea\u4e00\u7c7b\u3002<\/p>\n

                  \u5751 3：\u9065\u64cd\u4f5c\u6570\u636e\u7684\u9ad8\u9891\u566a\u58f0\u4f1a\u6253\u7206\u52a8\u4f5c\u5206\u5757\u8bad\u7ec3\u3002
                  \n50 Hz \u9065\u64cd\u4f5c\u6570\u636e\u91cc，\u4eba\u624b\u7684\u5fae\u5c0f\u6296\u52a8\u8ba9\u76f8\u90bb\u5e27\u5173\u8282\u503c\u9ad8\u9891\u632f\u8361，ACT \u5f0f\u5206\u5757\u9884\u6d4b\u5728\u8fd9\u79cd\u6570\u636e\u4e0a\u8f93\u51fa\u4e0d\u7a33\u5b9a\u3002\u964d\u91c7\u6837\u5230 25 Hz \u8bad\u7ec3\u540e\u7a33\u5b9a\u3002\u8fd9\u4e2a\u5751\u548c\u91c7\u6837\u7387-\u63a7\u5236\u5e26\u5bbd\u7684\u5173\u7cfb\u503c\u5f97\u7ec6\u54c1：\u5206\u5757\u6a21\u578b\u9884\u6d4b\u7684\u662f"\u5e73\u6ed1\u610f\u56fe"，\u91c7\u6837\u7387\u9ad8\u4e8e\u610f\u56fe\u5e26\u5bbd\u65f6，\u6570\u636e\u91cc\u5168\u662f\u566a\u58f0\u800c\u6ca1\u6709\u610f\u56fe\u3002<\/p>\n

                  \u5751 4：\u5c0f\u6570\u636e\u4e0b\u7684 Diffusion Policy \u4e0d\u516c\u5e73\u3002
                  \n\u4f5c\u8005\u660e\u786e\u8bf4\u4e0d\u8ddf Diffusion Policy \u6bd4：\u6309 INTERACT \u7684\u7ed3\u8bba，DP \u5728 50 \u6761\u6f14\u793a\u91cf\u7ea7\u4e0b\u6839\u672c\u5b66\u4e0d\u52a8\u3002\u8fd9\u4e0d\u662f DP \u4e0d\u884c，\u662f\u6570\u636e\u9884\u7b97\u51b3\u5b9a\u7b97\u6cd5\u9009\u578b\u2014\u201450 \u6761\u6f14\u793a\u30015000 epochs\u3001\u5355\u5361 24GB \u662f\u8fd9\u7bc7\u7684\u9884\u7b97\u7ea6\u675f，ACT \u7cfb\u67b6\u6784\u6b63\u662f\u4e3a\u8fd9\u4e2a\u9884\u7b97\u751f\u7684\u3002<\/p>\n

                  \u5751 5：\u57fa\u7ebf\u9009\u62e9\u7684\u6b63\u5f53\u6027\u8981\u63d0\u524d\u5e03\u9632\u3002
                  \n\u4f5c\u8005\u5f15\u7528 Lee et al. \u7684\u7ed3\u8bba\u4e3a"\u53ea\u548c ACT \u6bd4"\u8fa9\u62a4（DP \u9700\u8981\u66f4\u591a\u6570\u636e\u3001ACT \u5df2\u5728\u53cc\u81c2\u4efb\u52a1\u4e0a\u80dc\u8fc7\u591a\u4e2a\u7b56\u7565）\u3002\u8fd9\u5728\u5ba1\u7a3f\u610f\u4e49\u4e0a\u662f\u5408\u7406\u7684，\u4f46\u8bfb\u8005\u5e94\u610f\u8bc6\u5230：\u672c\u6587\u7684\u6bd4\u8f83\u5bf9\u8c61\u662f"\u67b6\u6784\u53d8\u4f53"，\u4e0d\u662f"SOTA \u65b9\u6cd5"\u2014\u2014\u5b83\u56de\u7b54"\u600e\u4e48\u6446\u7f16\u7801\u5668\u89e3\u7801\u5668"，\u4e0d\u56de\u7b54"\u54ea\u79cd\u7b97\u6cd5\u6700\u5f3a"\u3002<\/p>\n

                  \u5de5\u7a0b\u7ea7<\/h4>\n

                  \u5751 6：\u6bcf\u6bb5\u5f00\u5934\u7684 CLS token \u662f\u5ec9\u4ef7\u4f46\u5173\u952e\u7684\u8bbe\u8ba1\u3002
                  \n\u591a\u6bb5\u8f93\u5165\u5e8f\u5217\u91cc，\u6ca1\u6709 CLS \u6c47\u805a\u7684\u8bdd，\u6bb5\u7ea7\u6458\u8981\u4fe1\u606f\u53ea\u80fd\u9760\u6ce8\u610f\u529b\u81ea\u53d1\u5f62\u6210，\u8bad\u7ec3\u4e0d\u7a33\u5b9a\u3002\u8fd9\u662f\u4ece BERT \u501f\u6765\u7684\u8001\u6280\u672f，\u5728\u673a\u5668\u4eba\u7b56\u7565\u91cc\u540c\u6837\u597d\u4f7f\u3002<\/p>\n

                  \u5751 7：\u771f\u5b9e\u573a\u666f\u7684\u7269\u7406\u9632\u62a4\u3002
                  \n\u5b9e\u9a8c\u53f0\u4e0a\u52a0\u4e86\u9632\u62a4\u5e03（protective tarp），\u76f8\u673a\u9ad8\u5ea6\u91cd\u65b0\u8c03\u6574\u2014\u2014\u53cc\u81c2\u6325\u9524\u3001\u62ac\u677f\u8fd9\u7c7b\u4efb\u52a1，\u7b56\u7565\u5931\u8d25\u4e00\u6b21\u5c31\u662f\u7269\u7406\u78b0\u649e\u3002\u771f\u673a\u5b9e\u9a8c\u7684\u5de5\u7a0b\u9884\u7b97\u91cc，\u9632\u62a4\u6c38\u8fdc\u8be5\u6392\u5728\u7cbe\u5ea6\u524d\u9762\u3002<\/p>\n

                  \u5751 8：\u4f5c\u8005\u81ea\u5df1\u627f\u8ba4\u63d0\u5347\u5e45\u5ea6\u4efb\u52a1\u76f8\u5173\u3002
                  \n\u7ed3\u8bba\u91cc\u660e\u786e\u5199：\u201cdepending on the type of tasks, the improvement in success rate by the proposed method may not be significant\u201d\u3002\u770b\u6570\u636e\u786e\u5b9e\u5982\u6b64：Open Toolbox\u3001Open Lid \u4e0a\u5404\u67b6\u6784\u51e0\u4e4e\u90fd 100%，IACE \u7684\u4ef7\u503c\u96c6\u4e2d\u5728\u9700\u8981\u7cbe\u7ec6\u534f\u8c03\u7684\u4efb\u52a1（\u62ac\u653e\u6bdb\u5dfe\u3001\u6298\u6bdb\u5dfe\u3001\u6325\u9524）\u4e0a\u3002\u7b80\u5355\u540c\u6b65\u4efb\u52a1\u4e0a\u5806\u67b6\u6784\u662f\u6d6a\u8d39\u3002<\/p>\n

                  \n

                  \u516d\u3001\u6e90\u7801\u62c6\u89e3<\/h3>\n

                  \u8bba\u6587\u672a\u9644\u5b98\u65b9\u4ee3\u7801（\u4ee5\u8bba\u6587\u6b63\u6587 + \u67b6\u6784\u56fe\u4e3a\u51c6\u91cd\u5efa\u4f2a\u4ee3\u7801；\u57fa\u5ea7\u5b9e\u73b0\u53ef\u76f4\u63a5\u590d\u7528 ACT \u5f00\u6e90\u4ed3\u5e93）\u3002<\/p>\n

                  6.1 \u6574\u4f53\u524d\u5411（\u5bf9\u5e94\u8bba\u6587 Figure 1a\/1b）<\/h4>\n

                  # \u4f2a\u4ee3\u7801：\u57fa\u4e8e ACT \u7684 IACE \u67b6\u6784\u3002\u5bf9\u5e94\u8bba\u6587 Fig.1<\/span>
                  \nclass<\/span> IACE_ACT<\/span>(<\/span>nn.<\/span>Module)<\/span>:<\/span>
                  \n def<\/span> __init__<\/span>(<\/span>self,<\/span> single_decoder=<\/span>True<\/span>)<\/span>:<\/span>
                  \n self.<\/span>img_backbone =<\/span> ResNet18(<\/span>)<\/span> # 4 \u8def\u76f8\u673a\u5171\u4eab\u4e3b\u5e72<\/span>
                  \n self.<\/span>enc_left =<\/span> TransformerEncoder(<\/span>)<\/span> # \u5de6\u81c2\u5c40\u90e8\u7f16\u7801\u5668<\/span>
                  \n self.<\/span>enc_right =<\/span> TransformerEncoder(<\/span>)<\/span> # \u53f3\u81c2\u5c40\u90e8\u7f16\u7801\u5668<\/span>
                  \n self.<\/span>enc_global=<\/span> TransformerEncoder(<\/span>)<\/span> # \u5168\u5c40\u6bb5\u7f16\u7801\u5668<\/span>
                  \n self.<\/span>iace =<\/span> TransformerEncoder(<\/span>)<\/span> # \u2605 \u81c2\u95f4\u534f\u8c03\u7f16\u7801\u5668 (\u6838\u5fc3)<\/span>
                  \n if<\/span> single_decoder:<\/span>
                  \n self.<\/span>decoder =<\/span> TransformerDecoder(<\/span>)<\/span> # \u5355\u89e3\u7801\u5668: \u62fc\u63a5\u8f93\u5165<\/span>
                  \n else<\/span>:<\/span>
                  \n self.<\/span>dec_left =<\/span> TransformerDecoder(<\/span>)<\/span> # \u5206\u88c2\u89e3\u7801\u5668<\/span>
                  \n self.<\/span>dec_right =<\/span> TransformerDecoder(<\/span>)<\/span>
                  \n self.<\/span>cvae_enc =<\/span> CVAEEncoder(<\/span>)<\/span> # \u98ce\u683c\u9690\u53d8\u91cf z<\/span>
                  \n self.<\/span>chunk_size =<\/span> k # \u4e00\u6b21\u9884\u6d4b k \u6b65<\/span><\/p>\n

                  def<\/span> build_sequence<\/span>(<\/span>self,<\/span> joints_L,<\/span> joints_R,<\/span> cams)<\/span>:<\/span>
                  \n # \u6bcf\u4e2a segment: CLS token + \u9010\u5e27 token + \u4f4d\u7f6e\u7f16\u7801<\/span>
                  \n seg_L =<\/span> with_cls(<\/span>self.<\/span>enc_left(<\/span> tokens(<\/span>joints_L,<\/span> feats(<\/span>cams.<\/span>wrist_L)<\/span>)<\/span>)<\/span>)<\/span>
                  \n seg_R =<\/span> with_cls(<\/span>self.<\/span>enc_right(<\/span> tokens(<\/span>joints_R,<\/span> feats(<\/span>cams.<\/span>wrist_R)<\/span>)<\/span>)<\/span>)<\/span>
                  \n seg_G =<\/span> with_cls(<\/span>self.<\/span>enc_global(<\/span>tokens(<\/span>cat(<\/span>joints_L,<\/span> joints_R)<\/span>,<\/span>
                  \n feats(<\/span>cams.<\/span>upper,<\/span> cams.<\/span>lower)<\/span>)<\/span>)<\/span>)<\/span>
                  \n return<\/span> seg_L,<\/span> seg_R,<\/span> seg_G<\/p>\n

                  def<\/span> forward<\/span>(<\/span>self,<\/span> obs,<\/span> z=<\/span>None<\/span>)<\/span>:<\/span>
                  \n seg_L,<\/span> seg_R,<\/span> seg_G =<\/span> self.<\/span>build_sequence(<\/span>**<\/span>obs)<\/span>
                  \n # \u2605 IACE: \u53cc\u81c2+\u5168\u5c40 token \u62fc\u63a5\u540e\u8fc7\u534f\u8c03\u7f16\u7801\u5668 \u2192 \u540c\u6b65\u4e0e\u65f6\u5e8f\u5bf9\u9f50<\/span>
                  \n H_coord =<\/span> self.<\/span>iace(<\/span>cat(<\/span>seg_L,<\/span> seg_R,<\/span> seg_G)<\/span>)<\/span>
                  \n if<\/span> self.<\/span>single_decoder:<\/span>
                  \n memory =<\/span> cat(<\/span>seg_L,<\/span> seg_R,<\/span> H_coord)<\/span> # \u5168\u90e8\u62fc\u7ed9\u5355\u89e3\u7801\u5668<\/span>
                  \n return<\/span> self.<\/span>decoder(<\/span>action_queries,<\/span> memory)<\/span> # (k, 2*ndof) \u8054\u5408\u5757<\/span>
                  \n else<\/span>:<\/span>
                  \n a_L =<\/span> self.<\/span>dec_left (<\/span>action_queries_L,<\/span> cat(<\/span>seg_L,<\/span> H_coord)<\/span>)<\/span>
                  \n a_R =<\/span> self.<\/span>dec_right(<\/span>action_queries_R,<\/span> cat(<\/span>seg_R,<\/span> H_coord)<\/span>)<\/span>
                  \n return<\/span> a_L,<\/span> a_R # \u5404\u81ea\u72ec\u7acb\u5757<\/span><\/p>\n

                  6.2 \u8bad\u7ec3\u5faa\u73af（\u5bf9\u5e94\u8bba\u6587 Section III-B）<\/h4>\n

                  # \u4f2a\u4ee3\u7801：\u8bad\u7ec3\u3002\u5bf9\u5e94\u8bba\u6587 III-B \u5b9e\u73b0\u7ec6\u8282<\/span>
                  \n# \u6570\u636e: ALOHA \u9065\u64cd\u4f5c 50Hz \u91c7\u96c6 \u2192 \u964d\u91c7\u6837 25Hz (\u57513: \u9ad8\u9891\u6296\u52a8\u5bfc\u81f4\u8bad\u7ec3\u4e0d\u7a33)<\/span>
                  \nfor<\/span> epoch in<\/span> range<\/span>(<\/span>5000<\/span>)<\/span>:<\/span>
                  \n joints_L,<\/span> joints_R,<\/span> images,<\/span> targets =<\/span> sample_batch(<\/span>batch=<\/span>8<\/span>)<\/span> # 25Hz<\/span>
                  \n z =<\/span> self.<\/span>cvae_enc(<\/span>images,<\/span> targets)<\/span> # z \u7f16\u7801\u5668\u770b\u5f97\u89c1\u76ee\u6807 \u2192 \u5438\u6536\u98ce\u683c\u53d8\u5316<\/span>
                  \n z =<\/span> reparameterize(<\/span>z)<\/span> # z ~ N(mu, sigma), KL \u7ea6\u675f<\/span>
                  \n pred =<\/span> model(<\/span>obs,<\/span> z)<\/span>
                  \n loss =<\/span> l1_loss(<\/span>pred,<\/span> targets)<\/span> +<\/span> 1e-5<\/span> *<\/span> kl_loss # ACT \u635f\u5931: L1 + KL<\/span>
                  \n adamw_step(<\/span>lr=<\/span>.<\/span>.<\/span>.<\/span>,<\/span> weight_decay=<\/span>1e-5<\/span>)<\/span><\/p>\n

                  6.3 \u63a8\u7406：\u5206\u5757 + \u65f6\u5e8f\u96c6\u6210（\u5bf9\u5e94\u8bba\u6587 Section III-B \u672b）<\/h4>\n

                  # \u4f2a\u4ee3\u7801：\u63a8\u7406\u3002\u5bf9\u5e94\u8bba\u6587 "temporal ensemble" + "action chunking"<\/span>
                  \naction_queue =<\/span> TemporalEnsembleBuffer(<\/span>exp_weight=<\/span>exp(<\/span>–<\/span>lambda<\/span>*<\/span>i)<\/span>)<\/span><\/p>\n

                  @torch.<\/span>no_grad<\/span>(<\/span>)<\/span>
                  \ndef<\/span> control_loop<\/span>(<\/span>t)<\/span>:<\/span> # ~0.02s \u6bcf\u6b21\u63a8\u7406<\/span>
                  \n z =<\/span> prior_sample(<\/span>)<\/span> # \u91c7\u6837\u9636\u6bb5: z \u4ece N(0,I) \u62bd<\/span>
                  \n chunk =<\/span> model(<\/span>obs,<\/span> z)<\/span> # \u4e00\u6b21\u9884\u6d4b\u672a\u6765 k \u6b65<\/span>
                  \n action_queue.<\/span>push(<\/span>chunk,<\/span> t)<\/span> # \u4e0e\u5386\u53f2\u9884\u6d4b\u6309\u6307\u6570\u6743\u91cd\u878d\u5408<\/span>
                  \n return<\/span> action_queue.<\/span>weighted_action(<\/span>t)<\/span> # \u5e73\u6ed1\u8f93\u51fa\u5f53\u524d\u6b65\u52a8\u4f5c<\/span><\/p>\n

                  6.4 \u5173\u952e\u8d85\u53c2\u6570\u4e0e\u8bbe\u8ba1\u51b3\u7b56\u8868<\/h4>\n\n\u9879\u76ee\u503c\/\u51b3\u7b56\u6765\u6e90\u4e0e\u7406\u7531<\/tr>\n\n\n\n\n\n\n\n\n\n
                  \u8bad\u7ec3\u9891\u7387<\/td>\n50 Hz \u91c7\u96c6 \u2192 25 Hz \u8bad\u7ec3<\/td>\n\u9ad8\u9891\u5173\u8282\u6296\u52a8\u5bfc\u81f4\u8f93\u51fa\u4e0d\u7a33\u5b9a（\u57513）<\/td>\n<\/tr>\n
                  \u6f14\u793a\u91cf<\/td>\n\u6bcf\u4efb\u52a1 50 \u6761<\/td>\nACT \u539f\u914d\u7f6e<\/td>\n<\/tr>\n
                  \u4f18\u5316<\/td>\nAdamW, \u03bb=1e-5, 5000 epochs, batch 8<\/td>\n\u8bba\u6587 III-B<\/td>\n<\/tr>\n
                  \u98ce\u683c\u5efa\u6a21<\/td>\nCVAE + \u9690\u53d8\u91cf z<\/td>\n\u7ee7\u627f ACT<\/td>\n<\/tr>\n
                  \u8f93\u51fa<\/td>\n\u52a8\u4f5c\u5206\u5757（k \u6b65\u5e8f\u5217）+ \u65f6\u5e8f\u96c6\u6210<\/td>\n\u7ee7\u627f ACT<\/td>\n<\/tr>\n
                  \u89c6\u89c9\u4e3b\u5e72<\/td>\nResNet18（4 \u8def\u76f8\u673a）<\/td>\n\u7ee7\u627f ACT<\/td>\n<\/tr>\n
                  \u534f\u8c03\u673a\u5236<\/td>\nIACE（\u72ec\u7acb\u534f\u8c03\u7f16\u7801\u5668）<\/td>\n\u672c\u6587\u8d21\u732e<\/td>\n<\/tr>\n
                  \u89e3\u7801\u5668<\/td>\n\u5f02\u6b65\u2192\u5206\u88c2 \/ \u540c\u6b65\u2192\u5355<\/td>\n\u672c\u6587\u5b9e\u9a8c\u7ed3\u8bba<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
                  \n

                  \u4e03\u3001\u6548\u679c\u5bf9\u6bd4<\/h3>\n

                  7.1 \u516b\u4e2a\u771f\u673a\u4efb\u52a1 \u00d7 25 episodes（Table I）<\/h4>\n

                  \u6ce8：\u5f02\u6b65\u4efb\u52a1\u7684\u539f\u59cb\u8868\u683c\u5728 PDF \u63d0\u53d6\u4e2d\u5217\u5e8f\u6709\u9519\u4f4d，\u4e0b\u8868\u7684\u884c\u5747\u503c\u7ecf\u9010\u884c\u6c42\u548c\u6838\u7b97（\u4e0e\u8bba\u6587"\u5e73\u5747\u6bd4 ACT \u9ad8 8-9%"\u7684\u58f0\u660e\u4ea4\u53c9\u9a8c\u8bc1\u4e00\u81f4），\u4e2a\u522b\u5355\u5143\u683c\u5f52\u5c5e\u4ee5\u6b63\u6587\u660e\u786e\u53d9\u8ff0\u7684\u6570\u5b57\u4e3a\u51c6\u3002<\/p>\n

                  \u5f02\u6b65\u4efb\u52a1（4 \u9879，9 \u4e2a\u5b50\u6307\u6807）\u5e73\u5747\u6210\u529f\u7387：<\/p>\n\n\u65b9\u6cd5\u5e73\u5747\u6210\u529f\u7387<\/tr>\n\n\n\n\n\n\n
                  ACT<\/td>\n82.2%<\/td>\n<\/tr>\n
                  Ours（\u5355\u89e3\u7801\u5668）<\/td>\n85.8%<\/td>\n<\/tr>\n
                  Ours（\u5206\u88c2\u89e3\u7801\u5668）<\/td>\n89.3%<\/td>\n<\/tr>\n
                  Ours（\u5355\u89e3\u7801\u5668）w\/o IACE<\/td>\n60.4%<\/td>\n<\/tr>\n
                  Ours（\u5206\u88c2\u89e3\u7801\u5668）w\/o IACE<\/td>\n74.7%<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

                  \u540c\u6b65\u4efb\u52a1（4 \u9879，5 \u4e2a\u5b50\u6307\u6807）\u5e73\u5747\u6210\u529f\u7387：<\/p>\n\n\u65b9\u6cd5\u5e73\u5747\u6210\u529f\u7387<\/tr>\n\n\n\n\n\n\n
                  ACT<\/td>\n80.8%<\/td>\n<\/tr>\n
                  Ours（\u5355\u89e3\u7801\u5668）<\/td>\n98.4%<\/td>\n<\/tr>\n
                  Ours（\u5206\u88c2\u89e3\u7801\u5668）<\/td>\n90.4%<\/td>\n<\/tr>\n
                  Ours（\u5355\u89e3\u7801\u5668）w\/o IACE<\/td>\n56.8%<\/td>\n<\/tr>\n
                  Ours（\u5206\u88c2\u89e3\u7801\u5668）w\/o IACE<\/td>\n57.6%<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

                  \u5168\u90e8 14 \u4e2a\u5b50\u6307\u6807\u603b\u5e73\u5747： ACT 81.7% \u2192 Ours \u5355\u89e3\u7801\u5668 90.3% \/ \u5206\u88c2\u89e3\u7801\u5668 89.7%\u2014\u2014\u7ea6 +8.6 \u4e2a\u767e\u5206\u70b9，\u4e0e\u8bba\u6587\u6458\u8981"8-9% higher success rate than ACT"\u5b8c\u5168\u543b\u5408\u3002<\/p>\n

                  \u6b63\u6587\u70b9\u540d\u7684\u9ad8\u4eae\u5bf9\u6bd4：<\/p>\n

                    \n
                  • \u6298\u6bdb\u5dfe：\u5206\u88c2\u89e3\u7801\u5668\u6361\u653e\u5168 96%，ACT \u6361 92% \/ \u653e 88%；<\/li>\n
                  • \u9012\u94f6\u5305：ACT \u548c\u5206\u88c2\u89e3\u7801\u5668\u53cc\u53cc 100%（\u6293\u53d6 + \u4ea4\u63a5）；<\/li>\n
                  • \u62ac\u653e\u6bdb\u5dfe（\u540c\u6b65）：ACT \u7684"\u653e"\u5b50\u6307\u6807\u4ec5 4%，Ours \u5355\u89e3\u7801\u5668 96%\u2014\u2014\u540c\u6b65\u534f\u8c03\u5dee\u8ddd\u6700\u523a\u773c\u7684\u4e00\u4e2a\u6570\u5b57\u3002<\/li>\n<\/ul>\n

                    7.2 IACE \u6d88\u878d：\u5750\u6807\u6700\u91cd\u8981\u7684\u53d1\u73b0<\/h4>\n\n\u67b6\u6784\u5f02\u6b65\u5747\u503c\u540c\u6b65\u5747\u503c\u603b\u5747\u503c<\/tr>\n\n\n\n\n\n
                    Ours（\u5206\u88c2）+ IACE<\/td>\n89.3%<\/td>\n90.4%<\/td>\n89.7%<\/td>\n<\/tr>\n
                    Ours（\u5206\u88c2）w\/o IACE<\/td>\n74.7%（\u219314.6）<\/td>\n57.6%（\u219332.8）<\/td>\n74.7%<\/td>\n<\/tr>\n
                    Ours（\u5355）+ IACE<\/td>\n85.8%<\/td>\n98.4%<\/td>\n90.3%<\/td>\n<\/tr>\n
                    Ours（\u5355）w\/o IACE<\/td>\n60.4%（\u219325.4）<\/td>\n56.8%（\u219341.6）<\/td>\n60.4%<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

                    \u4e24\u4e2a\u8bfb\u6cd5：<\/p>\n

                  • \u540c\u6b65\u4efb\u52a1\u5bf9 IACE \u7684\u4f9d\u8d56\u662f\u6bc1\u706d\u6027\u7684\u2014\u2014\u53bb\u6389\u540e\u65e0\u8bba\u54ea\u79cd\u89e3\u7801\u5668\u90fd\u6389\u5230 57% \u9644\u8fd1，\u56e0\u4e3a"\u540c\u65f6\u62ac\/\u540c\u65f6\u5f00"\u8fd9\u7c7b\u4efb\u52a1\u6ca1\u6709\u534f\u8c03\u901a\u8def\u5c31\u7269\u7406\u4e0a\u4e0d\u53ef\u80fd\u5b8c\u6210（\u62ac\u6bdb\u5dfe\u4e00\u53ea\u624b\u665a 0.5 \u79d2，\u6bdb\u5dfe\u5c31\u6389）；<\/li>\n
                  • \u5355\u89e3\u7801\u5668\u6bd4\u5206\u88c2\u89e3\u7801\u5668\u66f4\u4f9d\u8d56 IACE（-41.6 vs -32.8）\u2014\u2014\u5355\u89e3\u7801\u5668\u628a\u534f\u8c03\u5b8c\u5168\u4ea4\u7ed9 IACE \u5355\u70b9\u8d1f\u8d23，\u5206\u88c2\u89e3\u7801\u5668\u81f3\u5c11\u8fd8\u6709\u5404\u81c2\u72ec\u7acb\u89e3\u7801\u7684\u515c\u5e95\u3002\u53cd\u8fc7\u6765\u8bf4，IACE + \u5355\u89e3\u7801\u5668\u662f\u540c\u6b65\u4efb\u52a1\u7684\u6700\u4f18\u7ec4\u5408（98.4%）\u3002<\/li>\n

                    7.3 \u4e00\u53e5\u8bdd\u603b\u7ed3\u5bf9\u6bd4\u8868<\/h4>\n

                    \u5f02\u6b65\u4efb\u52a1 \u540c\u6b65\u4efb\u52a1 \u603b\u5747\u503c
                    \n ACT \u57fa\u7ebf 82.2% 80.8% 81.7%
                    \n Ours(\u5355) 85.8% 98.4% \u2605 90.3%
                    \n Ours(\u5206\u88c2) 89.3% \u2605 90.4% 89.7%
                    \n \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500
                    \n \u53bb\u6389 IACE: 60~75% (\u5f02\u6b65) 57% (\u540c\u6b65) \u2190 \u67b6\u6784\u6838\u5fc3\u662f IACE, \u4e0d\u662f\u89e3\u7801\u5668\u6446\u6cd5<\/p>\n

                    \n

                    \u516b\u3001\u603b\u7ed3<\/h3>\n

                    8.1 \u6838\u5fc3\u601d\u60f3\u4e00\u53e5\u8bdd<\/h4>\n

                    \u53cc\u81c2\u7b56\u7565\u7684\u67b6\u6784\u4e0d\u8be5\u662f"\u7ef4\u5ea6\u62fc\u63a5"\u7684\u9ed1\u7bb1，\u800c\u5e94\u8be5\u662f"\u6bcf\u81c2\u4e00\u6761\u6df1\u5b66\u901a\u8def + \u4e00\u6761\u663e\u5f0f\u534f\u8c03\u901a\u8def（IACE）"\u7684\u5206\u5c42\u8bbe\u8ba1\u2014\u2014\u4e14\u89e3\u7801\u5668\u6309\u4efb\u52a1\u8026\u5408\u5ea6\u9009\u578b：\u5f02\u6b65\u7528\u5206\u88c2，\u540c\u6b65\u7528\u5355\u3002<\/p>\n

                    8.2 \u8fb9\u754c<\/h4>\n

                    \u8bba\u6587\u77ed\u5c0f，\u8fb9\u754c\u4e5f\u660e\u6446\u7740：<\/p>\n

                      \n
                    • \u57fa\u7ebf\u53ea\u6709 ACT \u4e00\u4e2a，\u6ca1\u6709\u9a8c\u8bc1 IACE \u601d\u60f3\u80fd\u5426\u8fc1\u79fb\u5230 Diffusion Policy\u3001SARNN \u7b49\u5176\u4ed6\u7b97\u6cd5（\u4f5c\u8005\u81ea\u5df1\u5728\u7ed3\u8bba\u91cc\u627f\u8ba4\u8fd9\u662f\u5f85\u529e）；<\/li>\n
                    • \u4efb\u52a1\u5168\u90e8\u662f\u684c\u9762\u53cc\u81c2，\u6ca1\u6709\u79fb\u52a8\u3001\u6ca1\u6709\u5168\u8eab\u534f\u8c03（\u5bf9\u6bd4 BRS \u7684 21-DoF \u5168\u8eab\u4efb\u52a1），\u534f\u8c03\u95ee\u9898\u88ab\u7b80\u5316\u5728"\u4e24\u53ea\u624b"\u7684\u5c3a\u5ea6；<\/li>\n
                    • \u6bcf task 50 \u6761\u6f14\u793a\u3001\u56fa\u5b9a 8 \u4e2a\u4efb\u52a1\u2014\u2014\u67b6\u6784\u7ed3\u8bba\u7684\u5916\u63a8\u8303\u56f4（\u66f4\u591a\u4efb\u52a1\u3001\u66f4\u591a\u6570\u636e\u91cf\u3001\u5176\u4ed6\u673a\u5668\u4eba\u5e73\u53f0）\u672a\u7ecf\u68c0\u9a8c；<\/li>\n
                    • \u63d0\u5347\u5e45\u5ea6\u5728\u90e8\u5206\u7b80\u5355\u4efb\u52a1\u4e0a\u4e0d\u663e\u8457（\u4f5c\u8005\u81ea\u5df1\u627f\u8ba4），IACE \u7684\u4ef7\u503c\u96c6\u4e2d\u5728\u534f\u8c03\u654f\u611f\u578b\u4efb\u52a1\u3002<\/li>\n<\/ul>\n

                      8.3 \u4e0e\u7cfb\u5217\u524d\u4e09\u7bc7\u7684\u8fde\u7ebf<\/h4>\n

                      \u56db\u7bc7\u8fde\u8bfb，"\u534f\u8c03"\u8fd9\u4e2a\u4e3b\u9898\u51fa\u73b0\u4e86\u4e09\u79cd\u622a\u7136\u4e0d\u540c\u7684\u5efa\u6a21\u8303\u5f0f，\u975e\u5e38\u503c\u5f97\u5bf9\u7167：<\/p>\n\n\u7ef4\u5ea6iDP3BRS (WB-VIMA)LINGO\u672c\u6587 (IACE)<\/tr>\n\n\n\n\n\n
                      \u534f\u8c03\u5bf9\u8c61<\/td>\n\u5168\u8eab 25-DoF<\/td>\n\u5e95\u76d8-\u8eaf\u5e72-\u53cc\u81c2<\/td>\n\u591a\u9636\u6bb5\u52a8\u4f5c<\/td>\n\u5de6\u53f3\u81c2<\/td>\n<\/tr>\n
                      \u534f\u8c03\u673a\u5236<\/td>\n\u9690\u5f0f（\u7aef\u5230\u7aef\u6269\u6563）<\/td>\n\u81ea\u56de\u5f52\u6761\u4ef6\u94fe（\u4e0a\u6e38\u2192\u4e0b\u6e38）<\/td>\n\u8c03\u5ea6\u5668\u5207\u6362\u9636\u6bb5<\/td>\n\u5bf9\u79f0\u63d2\u5165\u7f16\u7801\u5668<\/td>\n<\/tr>\n
                      \u534f\u8c03\u7ed3\u6784<\/td>\n\u65e0\u663e\u5f0f\u7ed3\u6784<\/td>\n\u6709\u5411\u5c42\u7ea7（\u8fd0\u52a8\u5b66\u5e8f）<\/td>\n\u65f6\u95f4\u5e8f<\/td>\n\u5bf9\u79f0\u53cc\u5411（\u5de6\u53f3\u5bf9\u7b49）<\/td>\n<\/tr>\n
                      \u4efb\u52a1\u6027\u8d28<\/td>\n\u64cd\u4f5c<\/td>\n\u5168\u8eab\u64cd\u4f5c<\/td>\n\u884c\u8d70+\u4ea4\u4e92<\/td>\n\u53cc\u81c2\u64cd\u4f5c<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

                      IACE \u7684\u534f\u8c03\u662f\u5bf9\u79f0\u7684（\u5de6\u53f3\u81c2\u5e73\u7b49\u4e92\u67e5），BRS \u7684\u534f\u8c03\u662f\u6709\u5411\u7684（\u5e95\u76d8\u2192\u8eaf\u5e72\u2192\u81c2，\u6309\u8fd0\u52a8\u5b66\u56e0\u679c\u5e8f），\u8fd9\u4e2a\u5dee\u5f02\u672c\u8eab\u5c31\u662f\u8bbe\u8ba1\u6d1e\u5bdf：\u5f53\u534f\u8c03\u53cc\u65b9\u6ca1\u6709\u5929\u7136\u7684\u56e0\u679c\u5c42\u7ea7\u65f6（\u53cc\u81c2\u5bf9\u7b49），\u5bf9\u79f0\u63d2\u5165\u5f0f\u534f\u8c03\u66f4\u5408\u9002；\u5f53\u5b58\u5728\u660e\u786e\u7684\u56e0\u679c\u94fe\u65f6（\u5e95\u76d8\u8bef\u5dee\u653e\u5927\u672b\u7aef\u8bef\u5dee），\u81ea\u56de\u5f52\u6761\u4ef6\u94fe\u66f4\u5408\u9002\u3002<\/p>\n

                      8.4 \u5ef6\u4f38<\/h4>\n

                      \u7528\u7f16\u7801\u7684\u7c7b\u6bd4\u6536\u5c3e：\u8fd9\u7bc7\u8bba\u6587\u672c\u8d28\u4e0a\u662f\u7ed9"\u591a\u901a\u9053\u4fe1\u53f7\u8054\u5408\u5efa\u6a21"\u505a\u4e86\u4e00\u6b21\u7f16\u7801\u7ed3\u6784\u6d88\u878d\u2014\u2014\u5206\u88c2\u89e3\u7801\u5668 = \u5404\u901a\u9053\u72ec\u7acb\u7f16\u7801 + \u516c\u5171\u53c2\u8003\u6d41（IACE），\u5355\u89e3\u7801\u5668 = \u8054\u5408\u7f16\u7801，\u800c"\u5f02\u6b65 vs \u540c\u6b65\u4efb\u52a1"\u5c31\u662f\u901a\u9053\u95f4\u76f8\u5173\u6027\u7684\u9ad8\u4e0e\u4f4e\u3002\u4fe1\u53f7\u5904\u7406\u91cc\u65e9\u6709\u7ed3\u8bba：\u901a\u9053\u76f8\u5173\u6027\u4f4e\u65f6\u72ec\u7acb\u7f16\u7801\u66f4\u4f18（\u53bb\u9664\u5197\u4f59\u5f00\u9500），\u76f8\u5173\u6027\u9ad8\u65f6\u8054\u5408\u7f16\u7801\u66f4\u4f18（\u5229\u7528\u76f8\u5173\u6027）\u2014\u2014\u673a\u5668\u4eba\u67b6\u6784\u8bbe\u8ba1\u6b63\u5728\u91cd\u65b0\u53d1\u73b0\u8fd9\u4e9b\u7ecf\u5178\u6743\u8861，\u800c IACE \u7684\u89d2\u8272\u6070\u597d\u662f\u90a3\u6761"\u516c\u5171\u53c2\u8003\u6d41"：\u4e0d\u7ba1\u4f60\u600e\u4e48\u7f16\u7801\u5404\u901a\u9053，\u65f6\u5e8f\u5bf9\u9f50\u7684\u4fe1\u606f\u603b\u5f97\u6709\u4e00\u6761\u4e13\u95e8\u7684\u901a\u8def\u6765\u4f20\u3002<\/p>\n

                      \n

                      \u535a\u5ba2\u57fa\u4e8e arXiv:2503.13916v1 (2025-03-18) \u64b0\u5199\u3002\u6ce8：\u8bba\u6587 Table I \u7684\u5f02\u6b65\u4efb\u52a1\u90e8\u5206\u5728 PDF \u6587\u672c\u63d0\u53d6\u4e2d\u5217\u5e8f\u5b58\u5728\u9519\u4f4d，\u535a\u5ba2\u4e2d\u5404\u884c\u5747\u503c\u7ecf\u9010\u884c\u6c42\u548c\u6838\u7b97\u5e76\u4e0e\u8bba\u6587\u6458\u8981\u58f0\u660e（+8-9%）\u4ea4\u53c9\u9a8c\u8bc1；\u6b63\u6587\u660e\u786e\u53d9\u8ff0\u7684\u5177\u4f53\u6570\u5b57（\u6298\u6bdb\u5dfe 96% vs 92\/88%\u3001\u94f6\u5305 100%\u3001\u62ac\u653e\u6bdb\u5dfe ACT \u653e\u7f6e 4%）\u5747\u76f4\u63a5\u5f15\u81ea\u539f\u6587\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

                      \u8bba\u6587\u6807\u9898: Learning Bimanual Manipulation via Action Chunking and Inter-Arm Coordination with Transformers
                      \n\u4f5c\u8005: Tomohiro Motoda, Ryo Hanai, Ryoichi Nakajo, Masaki Murooka, Floris Erich, Yukiyasu Domae
                      \n\u673a\u6784: \u4ea7\u4e1a\u6280\u672f\u7efc\u5408\u7814\u7a76\u6240 (AIST), \u65e5\u672c
                      \n\u53d1\u8868: arXiv:2503.13916<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[12236,12233,12235,12234,841],"topic":[],"class_list":["post-107605","post","type-post","status-publish","format-standard","hentry","category-server","tag-12236","tag-12233","tag-12235","tag-12234","tag-transformer"],"yoast_head":"\nIACE \u6df1\u5ea6\u89e3\u6790\uff1a\u53cc\u81c2\u673a\u5668\u4eba\u8be5\u600e\u4e48\u201c\u5206\u5de5\u201c\uff1fAIST \u7528 14 \u7ec4\u771f\u673a\u5b9e\u9a8c\u7ed9\u51fa\u67b6\u6784\u9009\u578b\u6307\u5357 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/107605.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"IACE \u6df1\u5ea6\u89e3\u6790\uff1a\u53cc\u81c2\u673a\u5668\u4eba\u8be5\u600e\u4e48\u201c\u5206\u5de5\u201c\uff1fAIST \u7528 14 \u7ec4\u771f\u673a\u5b9e\u9a8c\u7ed9\u51fa\u67b6\u6784\u9009\u578b\u6307\u5357 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u8bba\u6587\u6807\u9898: Learning Bimanual Manipulation via Action Chunking and Inter-Arm Coordination with Transformers \u4f5c\u8005: Tomohiro Motoda, Ryo Hanai, Ryoichi Nakajo, Masaki Murooka, Floris Erich, Yukiyasu Domae \u673a\u6784: \u4ea7\u4e1a\u6280\u672f\u7efc\u5408\u7814\u7a76\u6240 (AIST), \u65e5\u672c \u53d1\u8868: arXiv:2503.13916\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/107605.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-09-19T13:40:34+00:00\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"8 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/107605.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/107605.html\",\"name\":\"IACE \u6df1\u5ea6\u89e3\u6790\uff1a\u53cc\u81c2\u673a\u5668\u4eba\u8be5\u600e\u4e48\u201c\u5206\u5de5\u201c\uff1fAIST \u7528 14 \u7ec4\u771f\u673a\u5b9e\u9a8c\u7ed9\u51fa\u67b6\u6784\u9009\u578b\u6307\u5357 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-09-19T13:40:34+00:00\",\"dateModified\":\"2026-09-19T13:40:34+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/107605.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/107605.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/107605.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"IACE \u6df1\u5ea6\u89e3\u6790\uff1a\u53cc\u81c2\u673a\u5668\u4eba\u8be5\u600e\u4e48\u201c\u5206\u5de5\u201c\uff1fAIST \u7528 14 \u7ec4\u771f\u673a\u5b9e\u9a8c\u7ed9\u51fa\u67b6\u6784\u9009\u578b\u6307\u5357\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"IACE \u6df1\u5ea6\u89e3\u6790\uff1a\u53cc\u81c2\u673a\u5668\u4eba\u8be5\u600e\u4e48\u201c\u5206\u5de5\u201c\uff1fAIST \u7528 14 \u7ec4\u771f\u673a\u5b9e\u9a8c\u7ed9\u51fa\u67b6\u6784\u9009\u578b\u6307\u5357 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/107605.html","og_locale":"zh_CN","og_type":"article","og_title":"IACE \u6df1\u5ea6\u89e3\u6790\uff1a\u53cc\u81c2\u673a\u5668\u4eba\u8be5\u600e\u4e48\u201c\u5206\u5de5\u201c\uff1fAIST \u7528 14 \u7ec4\u771f\u673a\u5b9e\u9a8c\u7ed9\u51fa\u67b6\u6784\u9009\u578b\u6307\u5357 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u8bba\u6587\u6807\u9898: Learning Bimanual Manipulation via Action Chunking and Inter-Arm Coordination with Transformers \u4f5c\u8005: Tomohiro Motoda, Ryo Hanai, Ryoichi Nakajo, Masaki Murooka, Floris Erich, Yukiyasu Domae \u673a\u6784: \u4ea7\u4e1a\u6280\u672f\u7efc\u5408\u7814\u7a76\u6240 (AIST), \u65e5\u672c \u53d1\u8868: arXiv:2503.13916","og_url":"https:\/\/www.wsisp.com\/helps\/107605.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-09-19T13:40:34+00:00","author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"8 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/107605.html","url":"https:\/\/www.wsisp.com\/helps\/107605.html","name":"IACE \u6df1\u5ea6\u89e3\u6790\uff1a\u53cc\u81c2\u673a\u5668\u4eba\u8be5\u600e\u4e48\u201c\u5206\u5de5\u201c\uff1fAIST \u7528 14 \u7ec4\u771f\u673a\u5b9e\u9a8c\u7ed9\u51fa\u67b6\u6784\u9009\u578b\u6307\u5357 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-09-19T13:40:34+00:00","dateModified":"2026-09-19T13:40:34+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/107605.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/107605.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/107605.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"IACE \u6df1\u5ea6\u89e3\u6790\uff1a\u53cc\u81c2\u673a\u5668\u4eba\u8be5\u600e\u4e48\u201c\u5206\u5de5\u201c\uff1fAIST \u7528 14 \u7ec4\u771f\u673a\u5b9e\u9a8c\u7ed9\u51fa\u67b6\u6784\u9009\u578b\u6307\u5357"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/107605","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=107605"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/107605\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=107605"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=107605"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=107605"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=107605"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}