Recent interesting robotics papershttps://broadcastchannel-6vg.pages.devhttps://broadcastchannel-6vg.pages.dev/posts/32https://broadcastchannel-6vg.pages.dev/posts/32Mon, 12 May 2025 23:24:59 GMTVisuomotor Policies to <mark>Grasp</mark> Anything with Dexterous Hands<br /><br />- <mark>grasping</mark> based on stero input<br />- engineering example of teacher-student, sim2real, and geometric fabrics<br /><br /><a href="/search/%23arxiv">#arxiv</a> <a href="/search/%23dexterous">#dexterous</a> <a href="/search/%23grasp">#grasp</a> <a href="/search/%23sim2real">#sim2real</a><br /><br /><a href="https://fixupx.com/ritvik_singh9/status/1887582921453936928" target="_blank">X Overview</a>, <a href="https://dextrah-rgb.github.io/" target="_blank">Project Website</a>, <a href="https://dextreme.org/fgp.html" target="_blank">Geometric Fabrics</a><a href="https://fixupx.com/ritvik_singh9/status/1887582921453936928" target="_blank"> <div><i><b>🧵</b></i> Thread • FixupX</div> <div> <i></i> <div> <video src="https://cdn4.telesco.pe/file/b230a060dd.mp4?token=h__KYE0dRoFSiQ_UjwSNapH6cZ45FnuC0Hstof5qd6eAwhtEXNdHCFY9Rt79tthfXUZxxOjDFkC-8FWNcNeLf8N3heC3OoDlWCoLo6fWOmwIK0ilvqirMNXwM_kLrBANayrNz7p2vVoHWx2gicntNAzaAjU13k6IdcQRvWVzXcJHf-st96BrA0Y2TAfi6g2ODSRyFYlz3-fClkRHcz5cpNUsVfIgrOrvd5RamMFLq8-JnZrq5Fo5XDcsLzqOXEu7UHRdRo3L_OrrNyVRSXAp18B6pYUBm0Mx_RIA-mlpuTyrSn6vviUykIhMyksR33JtRt5-Xvq0iFZm54mjWVfNMA" width="100%" height="100%"></video> </div> </div> <div>Ritvik Singh (@ritvik_singh9)</div> <div>Excited to announce our latest work: DextrAH-RGB where we successfully train visuomotor policies in sim to perform dexterous <mark>grasping</mark> of arbitrary objects end-to-end from RGB input!</div> </a>https://broadcastchannel-6vg.pages.dev/posts/31https://broadcastchannel-6vg.pages.dev/posts/31Mon, 12 May 2025 22:53:25 GMTGraspVLA: a <mark>Grasping</mark> Foundation Model Pre-trained on Billion-scale Synthetic Action Data<br /><br />- combination of sim and real could help train robust <mark>grasping</mark> policy<br />- engineering example of pi-0 policy (VLM + flow matching)<br /><br /><a href="/search/%23arxiv">#arxiv</a> <a href="/search/%23grasp">#grasp</a> <a href="/search/%23simulation">#simulation</a> <a href="/search/%23sim2real">#sim2real</a> <a href="/search/%23flow_matching">#flow_matching</a><br /><br /><a href="https://fixupx.com/TuoLiuRobot/status/1878060780400324938" target="_blank">X Overview</a>, <a href="https://pku-epic.github.io/GraspVLA-web/" target="_blank">Project Website</a>, <a href="https://huggingface.co/datasets/nvidia/PhysicalAI-Robotics-GraspGen" target="_blank">NVidia GraspGen</a><a href="https://fixupx.com/TuoLiuRobot/status/1878060780400324938" target="_blank"> <div><i><b>🧵</b></i> Thread • FixupX</div> <img class="link_preview_image" alt="Tuo Liu (@TuoLiuRobot)" src="/static/https://cdn4.telesco.pe/file/snQZznpLEXDCivTCTlPQF8A1Bt_CNMLbNMUJ9HD9a71SuF7jRHKQM4bfLdgszyjSDCGlGdwbXAI9gvlCuVqwhybhtYsgQALNyiwCb5rTUDA53Gds1sKDewK2m9Wb0w5WBbLbA4wpyhPbjKZWfQJBOfysDojnMDj4tXmmKlfNHjFOw2CKyPG2UUzcQhTgZQRZZHKts_pzSaQe_AHsqADilY2eE5hIZhdYtq13wotBUiQH-bisrC1cuTab8JKBsjQ5Lra2TkV1qjPE0wWAjXUgqYMqyVG1tt5R09bUxfh7qOj5EHYYrEhonm1pqqTW0aZBagfq_uJPBAeU50EPcc0_bA.jpg" loading="lazy" /> <div>Tuo Liu (@TuoLiuRobot)</div> <div>Galbot（银河通用）has just dropped GraspVLA, the world’s first end-to-end embodied <mark>grasping</mark> model! Trained on over a billion “visual-language-action” pairs, GraspVLA is ready to handle a wide range of tasks, even the unexpected ones. <br /><br />And with a little post-training…</div> </a>https://broadcastchannel-6vg.pages.dev/posts/30https://broadcastchannel-6vg.pages.dev/posts/30Mon, 12 May 2025 22:22:37 GMTFunGrasp: Functional <mark>Grasping</mark> for Diverse Dexterous Hands<br /><br />- detailed explanation of engineering<br />- demonstrate the importance of contact signal<br /><br /><a href="/search/%23IROS2025">#IROS2025</a> <a href="/search/%23dexterous">#dexterous</a> <a href="/search/%23grasp">#grasp</a><br /><br /><a href="https://fixupx.com/Hui_Zhang_eth/status/1861698268473815414" target="_blank">X Overview</a>, <a href="https://hly-123.github.io/FunGrasp/" target="_blank">Project Website</a><a href="https://fixupx.com/Hui_Zhang_eth/status/1861698268473815414" target="_blank"> <div><i><b>🧵</b></i> Thread • FixupX</div> <div> <i></i> <div> <video src="https://cdn4.telesco.pe/file/02f28c3dcd.mp4?token=tiDgDjgitkDtJGxCH3mX_M9XbUUlu9fZarx7vKo0vq-DXWj13w8_vE9QvUYFjiPaeHVMbDwd6BbVsjI_h5BujF8XVrr86-J0TILGdnH5VHa20fVZptupCZMVGawA6Xl71l2S_Xu7WQn91w0VvB3ih9FrYvIATmEwkGXWMS_W16j4lHveDJNrcPn9J-ATBGNfCAI_tWhYsbkZ2wuB6aseb9ud3Ik0-lXEdIWYbXo11pdk7iH0mQScJ5uX1GJEuFmXD8jpOc9g1wXqIAhbvZzNNNLjwis2uVpMx7uLao_JZWirYdyducUe1SVABac-06vGLEsFRsL2uqw1FE9BELH-Zg" width="100%" height="100%"></video> </div> </div> <div>Hui Zhang (@Hui_Zhang_eth)</div> <div>People <mark>grasp</mark> objects with different poses for different tasks. Can dexterous robots possess similar functional <mark>grasping</mark> capabilities? Our recent work FunGrasp enables this for various unseen objects, diverse task-specific poses, and different robot hands.…</div> </a>