{"id":1028,"date":"2025-09-16T19:54:41","date_gmt":"2025-09-16T19:54:41","guid":{"rendered":"https:\/\/biconnect.psicologia.unimib.it\/?p=1028"},"modified":"2025-12-03T14:04:00","modified_gmt":"2025-12-03T14:04:00","slug":"ieee-icdl-2025-prague-part-i-robot-learning-theory-of-mind-through-self-observation","status":"publish","type":"post","link":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/ieee-icdl-2025-prague-part-i-robot-learning-theory-of-mind-through-self-observation\/","title":{"rendered":"IEEE ICDL 2025 (Prague) Part I: Robot Learning Theory of Mind through Self-Observation"},"content":{"rendered":"\n<p>We\u2019ll be at the <strong>2025 IEEE International Conference on Development and Learning (ICDL)<\/strong>, hosted by <strong>CTU in Prague<\/strong> on <strong>Sept 16\u201319<\/strong>.<\/p>\n\n\n\n<figure class=\"wp-block-image size-large is-resized\"><img fetchpriority=\"high\" decoding=\"async\" width=\"1024\" height=\"609\" src=\"https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/pexels-photo-753337-1024x609.jpeg\" alt=\"\" class=\"wp-image-1040\" style=\"width:839px;height:auto\" srcset=\"https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/pexels-photo-753337-1024x609.jpeg 1024w, https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/pexels-photo-753337-300x178.jpeg 300w, https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/pexels-photo-753337-768x457.jpeg 768w, https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/pexels-photo-753337-1536x914.jpeg 1536w, https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/pexels-photo-753337-2048x1219.jpeg 2048w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\">What we\u2019re showing<\/h2>\n\n\n\n<p><strong>Robot Learning Theory of Mind through Self-Observation: Exploiting the Intentions\u2013Beliefs Synergy<\/strong><br><strong>Francesca Bianco &amp; Dimitri Ognibene<\/strong> \u2014 presented as a <strong>spotlight + poster<\/strong>. The short paper distills our approach: teach a robot to read others\u2019 goals better by <strong>jointly learning intentions and beliefs<\/strong>, using the robot\u2019s <strong>own control\/belief traces as supervision<\/strong> (\u201cself-observation\u201d).<\/p>\n\n\n\n<figure class=\"wp-block-image size-large is-resized\"><img decoding=\"async\" width=\"1024\" height=\"791\" src=\"https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/Question_cover_bianco_ognibene_2025_page-0001-1024x791.jpg\" alt=\"\" class=\"wp-image-1032\" style=\"width:691px;height:auto\" srcset=\"https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/Question_cover_bianco_ognibene_2025_page-0001-1024x791.jpg 1024w, https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/Question_cover_bianco_ognibene_2025_page-0001-300x232.jpg 300w, https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/Question_cover_bianco_ognibene_2025_page-0001-768x593.jpg 768w, https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/Question_cover_bianco_ognibene_2025_page-0001-1536x1187.jpg 1536w, https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/Question_cover_bianco_ognibene_2025_page-0001-2048x1583.jpg 2048w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\">The gist<\/h2>\n\n\n\n<p>When targets are hidden or scenes are cluttered, behavior is driven by <strong>beliefs<\/strong> (what the agent thinks is true). If our model learns to <strong>predict beliefs and intentions together<\/strong>, it becomes better at <strong>guessing the goal<\/strong> from a short trajectory. In our earlier experiments this yielded:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>a small but consistent <strong>overall boost<\/strong> in target prediction (about <strong>+1.9 pp<\/strong> at mid-size training sets),<\/li>\n\n\n\n<li>a bigger jump when the <strong>target isn\u2019t visible<\/strong> (about <strong>+7 pp<\/strong>), and<\/li>\n\n\n\n<li>the <strong>largest gains in hard clutter<\/strong> (aligned distractors: up to <strong>+12.8 pp<\/strong>, ~30% relative).<\/li>\n<\/ul>\n\n\n\n<p>Under the hood, it\u2019s a <strong>Social Perception System (SPS)<\/strong>: a shared network with heads for next action\/state\/target\u2014and a <strong>belief-map head<\/strong> trained from self-observation; multi-tasking acts like a helpful regularizer.<\/p>\n\n\n\n<figure class=\"wp-block-image size-large is-resized\"><img decoding=\"async\" width=\"1024\" height=\"791\" src=\"https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/System_bianco_ognibene_2025_page-0001-1024x791.jpg\" alt=\"\" class=\"wp-image-1033\" style=\"width:730px;height:auto\" srcset=\"https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/System_bianco_ognibene_2025_page-0001-1024x791.jpg 1024w, https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/System_bianco_ognibene_2025_page-0001-300x232.jpg 300w, https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/System_bianco_ognibene_2025_page-0001-768x593.jpg 768w, https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/System_bianco_ognibene_2025_page-0001-1536x1187.jpg 1536w, https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/System_bianco_ognibene_2025_page-0001-2048x1583.jpg 2048w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\">Why it matters (the one-minute version)<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Beliefs help intention prediction.<\/strong> When we add a belief-prediction head to the network, overall <strong>target-prediction accuracy<\/strong> improves consistently once there\u2019s a modest amount of data (e.g., +1.9% at 25 maps; peaks around +1.8\u20131.9% across mid-size datasets). <\/li>\n\n\n\n<li><strong>Big gains in the hard cases.<\/strong> Under <strong>hidden-target<\/strong> conditions the joint model improves accuracy by <strong>~7 percentage points<\/strong>; with <strong>distractors aligned with the future path<\/strong> we see improvements <strong>up to ~12.8 points<\/strong> (\u224830% relative in that setting). <\/li>\n\n\n\n<li><strong>Trains on self, generalizes to others.<\/strong> Models trained from an agent\u2019s <strong>own belief\/decision traces<\/strong> transfer to agents with <strong>different decision policies<\/strong> while preserving the joint-learning benefit. <\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\">Come find us<\/h2>\n\n\n\n<p>ICDL schedules <strong>poster spotlights<\/strong> followed by <strong>poster sessions<\/strong> each afternoon\u2014check the program on site for exact slots and location at CTU-FEL. We\u2019d love to talk! (<a href=\"https:\/\/icdl2025.fel.cvut.cz\/program\/?utm_source=chatgpt.com\" target=\"_blank\" rel=\"noopener\">icdl2025.fel.cvut.cz<\/a>)<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Additional Material<\/h2>\n\n\n\n<p>Full paper <a href=\"https:\/\/arxiv.org\/abs\/2210.09435\" target=\"_blank\" rel=\"noopener\">[2210.09435] Robot Learning Theory of Mind through Self-Observation: Exploiting the Intentions-Beliefs Synergy<\/a><\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Slides<\/h2>\n\n\n\n<div class=\"wp-block-embedpress-embedpress-pdf\"><div class=\"embedpress-document-embed ep-doc-embedpress-pdf-1758102790079   ep-percentage-width\" style=\"width:600%;max-width:100%\" id=\"ep-doc-undefined\" data-source-id=\"source-undefined\" data-embed-type=\"PDF\"><div class=\"ep-embed-content-wraper\"><div class=\"position-right-wraper gutenberg-pdf-wraper\"><div class=\"main-content-wraper\"><iframe title=\"\" style=\"height:600px;width:100%\" class=\"embedpress-embed-document-pdf embedpress-pdf-1758102790079\" data-emid=\"embedpress-pdf-1758102790079\" src=\"https:\/\/biconnect.psicologia.unimib.it\/wp-admin\/admin-ajax.php?action=get_viewer&amp;scrolling=-1&amp;selection_tool=0&amp;spreads=0&amp;file=https%3A%2F%2Fbiconnect.psicologia.unimib.it%2Fwp-content%2Fuploads%2F2025%2F09%2FBIANCO_OGNIBENE_ICDL2025SPOTLIGHT.pdf#key=dGhlbWVNb2RlPWRlZmF1bHQmcHJlc2VudGF0aW9uPXRydWUmbGF6eUxvYWQ9ZmFsc2UmcG9zaXRpb249dG9wJmZsaXBib29rX3Rvb2xiYXJfcG9zaXRpb249Ym90dG9tJmRvd25sb2FkPXRydWUmdG9vbGJhcj10cnVlJmNvcHlfdGV4dD10cnVlJmFkZF90ZXh0PXRydWUmZHJhdz1mYWxzZSZkb2NfZGV0YWlscz10cnVlJmRvY19yb3RhdGlvbj10cnVlJmFkZF9pbWFnZT10cnVlJnpvb21faW49dHJ1ZSZ6b29tX291dD10cnVlJmZpdF92aWV3PXRydWUmYm9va21hcms9dHJ1ZSZzZWxlY3Rpb25fdG9vbD0wJnNjcm9sbGluZz0tMSZzcHJlYWRzPTA=\"><\/iframe><\/div><p class=\"embedpress-el-powered\">Powered By EmbedPress<\/p><\/div><\/div><\/div><\/div>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>Poster<\/strong><\/h2>\n\n\n\n<div class=\"wp-block-embedpress-embedpress-pdf\"><div class=\"embedpress-document-embed ep-doc-embedpress-pdf-1758102827919   ep-percentage-width\" style=\"width:600%;max-width:100%\" id=\"ep-doc-undefined\" data-source-id=\"source-undefined\" data-embed-type=\"PDF\"><div class=\"ep-embed-content-wraper\"><div class=\"position-right-wraper gutenberg-pdf-wraper\"><div class=\"main-content-wraper\"><iframe title=\"\" style=\"height:600px;width:100%\" class=\"embedpress-embed-document-pdf embedpress-pdf-1758102827919\" data-emid=\"embedpress-pdf-1758102827919\" src=\"https:\/\/biconnect.psicologia.unimib.it\/wp-admin\/admin-ajax.php?action=get_viewer&amp;scrolling=-1&amp;selection_tool=0&amp;spreads=0&amp;file=https%3A%2F%2Fbiconnect.psicologia.unimib.it%2Fwp-content%2Fuploads%2F2025%2F09%2FBiConnect-Lab-University-of-Milano-Bicocca-6.pdf#key=dGhlbWVNb2RlPWRlZmF1bHQmcHJlc2VudGF0aW9uPXRydWUmbGF6eUxvYWQ9ZmFsc2UmcG9zaXRpb249dG9wJmZsaXBib29rX3Rvb2xiYXJfcG9zaXRpb249Ym90dG9tJmRvd25sb2FkPXRydWUmdG9vbGJhcj10cnVlJmNvcHlfdGV4dD10cnVlJmFkZF90ZXh0PXRydWUmZHJhdz1mYWxzZSZkb2NfZGV0YWlscz10cnVlJmRvY19yb3RhdGlvbj10cnVlJmFkZF9pbWFnZT10cnVlJnpvb21faW49dHJ1ZSZ6b29tX291dD10cnVlJmZpdF92aWV3PXRydWUmYm9va21hcms9dHJ1ZSZzZWxlY3Rpb25fdG9vbD0wJnNjcm9sbGluZz0tMSZzcHJlYWRzPTA=\"><\/iframe><\/div><p class=\"embedpress-el-powered\">Powered By EmbedPress<\/p><\/div><\/div><\/div><\/div>\n\n\n\n<h2 class=\"wp-block-heading\">Acknowledgments<\/h2>\n\n\n\n<p>Supported by the <strong>Volkswagen Foundation<\/strong> project <strong>\u201cDeveloping an Artificial Social Childhood (ASC)\u201d<\/strong>.<\/p>\n\n\n\n<p><\/p>\n","protected":false},"excerpt":{"rendered":"<p>We\u2019ll be at the 2025 IEEE International Conference on Development and Learning (ICDL), hosted by CTU in Prague on Sept 16\u201319. What we\u2019re showing Robot Learning Theory of Mind through Self-Observation: Exploiting the Intentions\u2013Beliefs SynergyFrancesca Bianco &amp; Dimitri Ognibene \u2014 presented as a spotlight + poster. The short paper distills our approach: teach a robot &#8230; <a title=\"IEEE ICDL 2025 (Prague) Part I: Robot Learning Theory of Mind through Self-Observation\" class=\"read-more\" href=\"https:\/\/biconnect.psicologia.unimib.it\/index.php\/ieee-icdl-2025-prague-part-i-robot-learning-theory-of-mind-through-self-observation\/\" aria-label=\"Read more about IEEE ICDL 2025 (Prague) Part I: Robot Learning Theory of Mind through Self-Observation\">Read more<\/a><\/p>\n","protected":false},"author":1,"featured_media":1030,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[38],"tags":[],"class_list":["post-1028","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-events"],"acf":[],"_links":{"self":[{"href":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/wp-json\/wp\/v2\/posts\/1028","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/wp-json\/wp\/v2\/comments?post=1028"}],"version-history":[{"count":16,"href":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/wp-json\/wp\/v2\/posts\/1028\/revisions"}],"predecessor-version":[{"id":1121,"href":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/wp-json\/wp\/v2\/posts\/1028\/revisions\/1121"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/wp-json\/wp\/v2\/media\/1030"}],"wp:attachment":[{"href":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/wp-json\/wp\/v2\/media?parent=1028"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/wp-json\/wp\/v2\/categories?post=1028"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/wp-json\/wp\/v2\/tags?post=1028"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}