{"id":131,"date":"2026-03-18T14:07:32","date_gmt":"2026-03-18T14:07:32","guid":{"rendered":"https:\/\/lemursproject.udg.edu\/?p=131"},"modified":"2026-03-18T14:07:32","modified_gmt":"2026-03-18T14:07:32","slug":"virtual-environment-implementation-progress","status":"publish","type":"post","link":"https:\/\/lemursproject.udg.edu\/index.php\/2026\/03\/18\/virtual-environment-implementation-progress\/","title":{"rendered":"Virtual Environment Implementation Progress"},"content":{"rendered":"<p>One of the key challenges in the development of the <strong data-start=\"189\" data-end=\"207\">LEMURS project<\/strong> is the availability of an <strong data-start=\"234\" data-end=\"283\">accurate and efficient simulation environment<\/strong>. Such a simulator is essential to enable learning and validation of robotic manipulation tasks in a virtual setting before transferring them to real robotic platforms, reducing risks, costs, and development time.<\/p>\n<p data-start=\"525\" data-end=\"743\">Task <strong data-start=\"530\" data-end=\"575\">T3.3 \u2013 Virtual Environment Implementation<\/strong> focuses on the development of a realistic simulation framework capable of supporting the evaluation and optimization of machine learning-based manipulation algorithms. In this task, a virtual environment has to be implemented using the <strong data-start=\"806\" data-end=\"829\">Stonefish simulator<\/strong>, to simulate the upgraded <strong data-start=\"858\" data-end=\"878\">Girona 500 I-AUV<\/strong>. The environment must be designed to:<\/p>\n<ul data-start=\"957\" data-end=\"1386\">\n<li data-section-id=\"1fpa822\" data-start=\"957\" data-end=\"1023\">\n<p data-start=\"959\" data-end=\"1023\">Support the evaluation and optimization of ML-based algorithms<\/p>\n<\/li>\n<li data-section-id=\"1brclpt\" data-start=\"1024\" data-end=\"1108\">\n<p data-start=\"1026\" data-end=\"1108\">Define appropriate reward functions (sparse and\/or dense) for manipulation tasks<\/p>\n<\/li>\n<li data-section-id=\"18h9a6r\" data-start=\"1109\" data-end=\"1205\">\n<p data-start=\"1111\" data-end=\"1205\">Be calibrated through standard benchmark tasks such as reaching, pushing, and pick-and-place<\/p>\n<\/li>\n<li data-section-id=\"bs11k6\" data-start=\"1206\" data-end=\"1277\">\n<p data-start=\"1208\" data-end=\"1277\">Ensure sufficient realism to enable effective <strong data-start=\"1254\" data-end=\"1275\">Sim2Real transfer<\/strong><\/p>\n<\/li>\n<\/ul>\n<p data-start=\"1423\" data-end=\"1539\">\n<p data-start=\"1423\" data-end=\"1539\">To address the challenges of realism and computational efficiency, a <strong data-start=\"1492\" data-end=\"1521\">dual development approach<\/strong> has been adopted:<\/p>\n<p><strong>1. Stonefish-Based Simulation (High Realism)<\/strong><\/p>\n<p data-start=\"1594\" data-end=\"1809\">We are extending the highly realistic <strong data-start=\"1632\" data-end=\"1666\">Stonefish underwater simulator<\/strong> to support reinforcement learning workflows. This includes adapting the simulator for training and evaluating learning-based control policies (see <a class=\"decorated-link\" href=\"https:\/\/github.com\/narcispr\/stonefish_rl\" target=\"_new\" rel=\"noopener\" data-start=\"1828\" data-end=\"1868\">https:\/\/github.com\/narcispr\/stonefish_rl)<\/a>.<\/p>\n<p><strong>2. MuJoCo-Based Simulation (High Performance)<\/strong><\/p>\n<p data-start=\"1926\" data-end=\"2161\">In parallel, we are leveraging the widely used <strong data-start=\"1973\" data-end=\"1993\">MuJoCo simulator<\/strong>, known for its speed and efficiency in machine learning applications. A preliminary underwater model inspired by the <strong data-start=\"2111\" data-end=\"2131\">Girona 500 I-AUV<\/strong> has already been implemented. Using this framework, we have successfully trained policies for a <strong data-start=\"2229\" data-end=\"2277\">free-floating dual manipulation control task<\/strong>, demonstrating the feasibility of learning complex behaviors in simulation.<\/p>\n<p><iframe title=\"Preliminary version of DRL controller for underwater vehicle dual reach task\" width=\"800\" height=\"450\" src=\"https:\/\/www.youtube.com\/embed\/MBbpDSHgarc?feature=oembed\" frameborder=\"0\" allow=\"accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share\" referrerpolicy=\"strict-origin-when-cross-origin\" allowfullscreen><\/iframe><\/p>\n<p data-start=\"2376\" data-end=\"2414\">Future work within T3.3 will focus on:<\/p>\n<ul data-start=\"2416\" data-end=\"2711\">\n<li data-section-id=\"3ggndx\" data-start=\"2416\" data-end=\"2515\">\n<p data-start=\"2418\" data-end=\"2515\">Developing a more <strong data-start=\"2436\" data-end=\"2462\">accurate dynamic model<\/strong> of the Girona 500 I-AUV to reduce the Sim2Real gap<\/p>\n<\/li>\n<li data-section-id=\"1n87j37\" data-start=\"2516\" data-end=\"2592\">\n<p data-start=\"2518\" data-end=\"2592\">Enhancing simulation fidelity while maintaining computational efficiency<\/p>\n<\/li>\n<li data-section-id=\"kbetw0\" data-start=\"2593\" data-end=\"2711\">\n<p data-start=\"2595\" data-end=\"2711\">Transitioning from standard MuJoCo to <strong data-start=\"2633\" data-end=\"2653\">MuJoCo XLA (MJX)<\/strong> to significantly improve training speed and scalability<\/p>\n<\/li>\n<\/ul>\n<p data-start=\"2744\" data-end=\"2973\">This dual simulation strategy strengthens the LEMURS pipeline by combining <strong data-start=\"2819\" data-end=\"2846\">realism and performance<\/strong>, enabling faster iteration of learning algorithms while ensuring their applicability in real-world underwater robotic systems.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>One of the key challenges in the development of the LEMURS project is the availability of an accurate and efficient simulation environment. Such a simulator is essential to enable learning and validation of robotic manipulation tasks in a virtual setting before transferring them to real robotic platforms, reducing risks, costs, and development time. Task T3.3 [&hellip;]<\/p>\n","protected":false},"author":2,"featured_media":132,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[6],"tags":[],"class_list":["post-131","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news"],"_links":{"self":[{"href":"https:\/\/lemursproject.udg.edu\/index.php\/wp-json\/wp\/v2\/posts\/131","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/lemursproject.udg.edu\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/lemursproject.udg.edu\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/lemursproject.udg.edu\/index.php\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/lemursproject.udg.edu\/index.php\/wp-json\/wp\/v2\/comments?post=131"}],"version-history":[{"count":1,"href":"https:\/\/lemursproject.udg.edu\/index.php\/wp-json\/wp\/v2\/posts\/131\/revisions"}],"predecessor-version":[{"id":133,"href":"https:\/\/lemursproject.udg.edu\/index.php\/wp-json\/wp\/v2\/posts\/131\/revisions\/133"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/lemursproject.udg.edu\/index.php\/wp-json\/wp\/v2\/media\/132"}],"wp:attachment":[{"href":"https:\/\/lemursproject.udg.edu\/index.php\/wp-json\/wp\/v2\/media?parent=131"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/lemursproject.udg.edu\/index.php\/wp-json\/wp\/v2\/categories?post=131"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/lemursproject.udg.edu\/index.php\/wp-json\/wp\/v2\/tags?post=131"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}