{"id":68469,"date":"2024-04-24T14:55:44","date_gmt":"2024-04-24T06:55:44","guid":{"rendered":"https:\/\/inventec2.mjitec.tw\/?page_id=68469"},"modified":"2024-04-24T14:55:44","modified_gmt":"2024-04-24T06:55:44","slug":"expert-composer-policy-scalable-skill-repertoire-for-quadruped-robots","status":"publish","type":"page","link":"https:\/\/inventec2.mjitec.tw\/zh-hans\/ai\/expert-composer-policy-scalable-skill-repertoire-for-quadruped-robots\/","title":{"rendered":"Expanding Versatility of Agile Locomotion"},"content":{"rendered":"<div class=\"wpb-content-wrapper\"><p>[vc_row full_width=&#8221;stretch_row&#8221;][vc_column]<div id=\"rs-space-69e10d8f216b2\" class=\"rs-space\">\r\n                <div class=\"rs-space-data\" data-conf=\"{&quot;uqid&quot;:&quot;69e10d8f216b2&quot;,&quot;space_lg&quot;:&quot;150&quot;,&quot;space_md&quot;:&quot;80&quot;,&quot;space_sm&quot;:&quot;60&quot;,&quot;space_xs&quot;:&quot;60&quot;}\"><\/div>\t\t\t\r\n\t\t\t<\/div>[vc_row_inner el_class=&#8221;md-full-col&#8221;][vc_column_inner el_class=&#8221;m_p&#8221; width=&#8221;1\/2&#8243;]\n        <div class=\"rs-heading    \">\n        \t<div class=\"title-inner\"  data-border-color=\"\">\n        \t\t\n\t            \n\t            <h2 class=\"title \" style=\"color: #333333\">Expert Composer Policy: Scalable Skill Repertoire for Quadruped Robots <\/h2>\n\t        <\/div><\/div>[vc_column_text css=&#8221;.vc_custom_1713835880408{margin-bottom: 20px !important;}&#8221;]<\/p>\n<div>\n<p>International Conference on Robotics and Automation 2024 (ICRA 2024)<\/p>\n<\/div>\n<p>[\/vc_column_text][vc_column_text css=&#8221;.vc_custom_1660542172270{margin-bottom: 5px !important;}&#8221;]<\/p>\n<div>\n<h6>\u4f5c\u8005<\/h6>\n<\/div>\n<p>[\/vc_column_text][vc_column_text css=&#8221;.vc_custom_1713835894268{margin-bottom: 20px !important;}&#8221;]<\/p>\n<div>\n<p>Guilherme Christmann*, Ying-Sheng Luo*, Wei-Chao Chen<\/p>\n<\/div>\n<p>[\/vc_column_text][vc_column_text css=&#8221;.vc_custom_1689317028937{margin-bottom: 5px !important;}&#8221;]<\/p>\n<div>\n<h6>\u53d1\u8868\u65e5\u671f<\/h6>\n<\/div>\n<p>[\/vc_column_text][vc_column_text]<\/p>\n<div>\n<p>2024\/3\/18<\/p>\n<\/div>\n<p>[\/vc_column_text][\/vc_column_inner][vc_column_inner el_class=&#8221;m_p&#8221; width=&#8221;1\/2&#8243;][vc_single_image image=&#8221;68460&#8243; img_size=&#8221;full&#8221;][\/vc_column_inner][\/vc_row_inner][\/vc_column][\/vc_row][vc_row][vc_column]<div id=\"rs-space-69e10d8f217a4\" class=\"rs-space\">\r\n                <div class=\"rs-space-data\" data-conf=\"{&quot;uqid&quot;:&quot;69e10d8f217a4&quot;,&quot;space_lg&quot;:&quot;150&quot;,&quot;space_md&quot;:&quot;80&quot;,&quot;space_sm&quot;:&quot;60&quot;,&quot;space_xs&quot;:&quot;60&quot;}\"><\/div>\t\t\t\r\n\t\t\t<\/div>[\/vc_column][\/vc_row][vc_row full_width=&#8221;stretch_row&#8221;][vc_column][vc_row_inner content_placement=&#8221;top&#8221; css=&#8221;.vc_custom_1657794580528{margin-bottom: 20px !important;}&#8221;][vc_column_inner el_class=&#8221;m_p paragraph_title&#8221; width=&#8221;1\/3&#8243;]\n        <div class=\"rs-heading   vc_custom_1657008747808  \">\n        \t<div class=\"title-inner\"  data-border-color=\"\">\n        \t\t\n\t            \n\t            <h2 class=\"title \" style=\"color: #333333\">\u6982\u8981 <\/h2>\n\t        <\/div><\/div>[\/vc_column_inner][vc_column_inner el_class=&#8221;m_p&#8221; width=&#8221;2\/3&#8243;][vc_column_text]We propose the expert composer policy, a framework to reliably expand the skill repertoire of quadruped agents. The composer policy links pair of experts via transitions to a sampled target state, allowing experts to be composed sequentially. Each expert specializes in a single skill, such as a locomotion gait or a jumping motion. Instead of a hierarchical or mixture-of-experts architecture, we train a single composer policy in an independent process that is not conditioned on the other expert policies. By reusing the same composer policy, our approach enables adding new experts without affecting existing ones, enabling incremental repertoire expansion and preserving original motion quality. We measured the transition success rate of 72 transition pairs and achieved an average success rate of 99.99\\%, which is over 10\\% higher than the baseline random approach, and outperforms other state-of-the-art methods. Using domain randomization during training we ensure a successful transfer to the real world, where we achieve an average transition success rate of 97.22\\% (N=360) in our experiments.[\/vc_column_text][\/vc_column_inner][\/vc_row_inner][vc_row_inner content_placement=&#8221;middle&#8221; el_class=&#8221;md-full-col video_wrap&#8221;][vc_column_inner el_class=&#8221;m_p&#8221;][vc_raw_html]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[\/vc_raw_html][\/vc_column_inner][\/vc_row_inner][\/vc_column][\/vc_row][vc_row][vc_column]<div id=\"rs-space-69e10d8f21859\" class=\"rs-space\">\r\n                <div class=\"rs-space-data\" data-conf=\"{&quot;uqid&quot;:&quot;69e10d8f21859&quot;,&quot;space_lg&quot;:&quot;80&quot;,&quot;space_md&quot;:&quot;80&quot;,&quot;space_sm&quot;:&quot;60&quot;,&quot;space_xs&quot;:&quot;60&quot;}\"><\/div>\t\t\t\r\n\t\t\t<\/div>[\/vc_column][\/vc_row][vc_row][vc_column width=&#8221;1\/3&#8243; el_class=&#8221;m_p keyword_title&#8221;][vc_column_text]<\/p>\n<h2>\u5173\u952e\u5b57<\/h2>\n<p>[\/vc_column_text][\/vc_column][vc_column width=&#8221;2\/3&#8243; el_class=&#8221;m_p keyword&#8221;][vc_row_inner content_placement=&#8221;middle&#8221;][vc_column_inner width=&#8221;1\/3&#8243;][vc_raw_html]JTNDdWwlMjBjbGFzcyUzRCUyMnN0eWxlbGlzdGluZyUyMiUzRSUwQSUyMCUwOSUzQ2xpJTIwc3R5bGUlM0QlMjJsaW5lLWhlaWdodCUzQTM0cHglM0IlMjIlM0VSZWluZm9yY2VtZW50JTIwTGVhcm5pbmclM0MlMkZsaSUzRSUwQSUyMCUwOSUzQ2xpJTIwc3R5bGUlM0QlMjJsaW5lLWhlaWdodCUzQTM0cHglM0IlMjIlM0VDb250cm9sJTNDJTJGbGklM0UlMEElM0MlMkZ1bCUzRQ==[\/vc_raw_html][\/vc_column_inner][vc_column_inner width=&#8221;1\/3&#8243;][vc_raw_html]JTNDdWwlMjBjbGFzcyUzRCUyMnN0eWxlbGlzdGluZyUyMiUzRSUwQSUyMCUwOSUzQ2xpJTIwc3R5bGUlM0QlMjJsaW5lLWhlaWdodCUzQTM0cHglM0IlMjIlM0VSb2JvdGljcyUzQyUyRmxpJTNFJTBBJTNDJTJGdWwlM0U=[\/vc_raw_html][\/vc_column_inner][vc_column_inner width=&#8221;1\/3&#8243;][vc_raw_html]JTNDdWwlMjBjbGFzcyUzRCUyMnN0eWxlbGlzdGluZyUyMiUzRSUwQSUyMCUwOSUzQ2xpJTIwc3R5bGUlM0QlMjJsaW5lLWhlaWdodCUzQTM0cHglM0IlMjIlM0VMZWdnZWQlMjBSb2JvdHMlM0MlMkZsaSUzRSUwQSUzQyUyRnVsJTNF[\/vc_raw_html][\/vc_column_inner][\/vc_row_inner][\/vc_column][\/vc_row][vc_row][vc_column]<div id=\"rs-space-69e10d8f21894\" class=\"rs-space\">\r\n                <div class=\"rs-space-data\" data-conf=\"{&quot;uqid&quot;:&quot;69e10d8f21894&quot;,&quot;space_lg&quot;:&quot;80&quot;,&quot;space_md&quot;:&quot;80&quot;,&quot;space_sm&quot;:&quot;60&quot;,&quot;space_xs&quot;:&quot;60&quot;}\"><\/div>\t\t\t\r\n\t\t\t<\/div>[\/vc_column][\/vc_row][vc_row full_width=&#8221;stretch_row&#8221; el_class=&#8221;bg&#8221; css=&#8221;.vc_custom_1657248474326{padding-top: 50px !important;padding-bottom: 50px !important;}&#8221;][vc_column][vc_column_text css=&#8221;.vc_custom_1689317065018{margin-bottom: 20px !important;}&#8221;]<\/p>\n<h3 style=\"text-align: center; color: #fff;\">\u4e0b\u8f7d\u4e0e\u5206\u4eab<\/h3>\n<p>[\/vc_column_text][vc_row_inner content_placement=&#8221;middle&#8221;][vc_column_inner el_class=&#8221;download_btn_wrap&#8221;][vc_btn title=&#8221;PDF&#8221; style=&#8221;flat&#8221; color=&#8221;white&#8221; align=&#8221;center&#8221; link=&#8221;url:https%3A%2F%2Farxiv.org%2Fabs%2F2403.11412|target:_blank&#8221; el_class=&#8221;download_btn&#8221; css=&#8221;.vc_custom_1713837476921{padding-right: 20px !important;padding-left: 20px !important;}&#8221;][\/vc_column_inner][\/vc_row_inner][\/vc_column][\/vc_row]<\/p>\n<\/div>","protected":false},"excerpt":{"rendered":"<p>[vc_row full_width=&#8221;stretch_row&#8221;][vc_column&#8230;<\/p>\n","protected":false},"author":4,"featured_media":0,"parent":4976,"menu_order":0,"comment_status":"closed","ping_status":"closed","template":"","meta":{"footnotes":""},"class_list":["post-68469","page","type-page","status-publish","hentry"],"_links":{"self":[{"href":"https:\/\/inventec2.mjitec.tw\/zh-hans\/wp-json\/wp\/v2\/pages\/68469","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/inventec2.mjitec.tw\/zh-hans\/wp-json\/wp\/v2\/pages"}],"about":[{"href":"https:\/\/inventec2.mjitec.tw\/zh-hans\/wp-json\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"https:\/\/inventec2.mjitec.tw\/zh-hans\/wp-json\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"https:\/\/inventec2.mjitec.tw\/zh-hans\/wp-json\/wp\/v2\/comments?post=68469"}],"version-history":[{"count":4,"href":"https:\/\/inventec2.mjitec.tw\/zh-hans\/wp-json\/wp\/v2\/pages\/68469\/revisions"}],"predecessor-version":[{"id":68499,"href":"https:\/\/inventec2.mjitec.tw\/zh-hans\/wp-json\/wp\/v2\/pages\/68469\/revisions\/68499"}],"up":[{"embeddable":true,"href":"https:\/\/inventec2.mjitec.tw\/zh-hans\/wp-json\/wp\/v2\/pages\/4976"}],"wp:attachment":[{"href":"https:\/\/inventec2.mjitec.tw\/zh-hans\/wp-json\/wp\/v2\/media?parent=68469"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}