new_paper

Author: Guanya Shi

people.html

@@ -360,12 +360,15 @@
             <p align="center" style="font-size: 18px; font-weight: 700; margin-top: 0; margin-bottom: 0">Haotian Lin</p>
             <p align="center" style="margin-top: 0; margin-bottom: 0">RI MSR (Fall 2024)</p>
             <div class="social-icons-small">
-                <a href="https://www.linkedin.com/in/haotian-lin-9b29b7324/" target="_blank" aria-label="Homepage">
+                <a href="https://darthutopian.github.io/" target="_blank" aria-label="Homepage">
                     <i class="fas fa-home"></i>
                 </a>
                 <a href="mailto:vlin3@andrew.cmu.edu" target="_blank" aria-label="Email">
                     <i class="fas fa-envelope"></i>
                 </a>
+                <a href="https://x.com/DarthUtopian" target="_blank" aria-label="Twitter">
+                    <i class="fab fa-twitter"></i>
+                </a>
             </div>
         </td>
 

publications.html

@@ -74,6 +74,70 @@
     </tr>
     </table>
 
+    <table width="880" border="0" align="center" cellspacing="0" cellpadding="0">
+    <tr style="background-color: var(--highlight-color)">
+        <td style="width:35%; vertical-align:middle; padding-right: 20px;">
+            <div class="image-container">
+                <img src='publications/2025_PLD.gif' width="90%">
+            </div>
+        </td>
+        <td style="width:65%; vertical-align:middle">
+            <papertitle>Self-Improving Vision-Language-Action Models with Data Generation via Residual RL</papertitle>
+            <br>
+            Wenli Xiao<sup>*</sup>, Haotian Lin<sup>*</sup>, Andy Peng, Haoru Xue, Tairan He, Yuqi Xie, Fengyuan Hu, Jimmy Wu, Zhengyi Luo, Linxi "Jim" Fan, Guanya Shi, Yuke Zhu
+            <br>
+            <a href="https://arxiv.org/abs/2511.00091" target="_blank"><i class="far fa-file"></i> paper</a> &nbsp
+            <a href="https://www.wenlixiao.com/self-improve-VLA-PLD" target="_blank"><i class="fas fa-globe"></i> website</a> 
+            <p style="margin-top: 5px"><i class="fas fa-comment-dots"></i> TL;DR: Probe, Learn, Distill (PLD): On-policy probing from a base VLA model + off-policy residual RL + distillation for VLA post-training.
+        </td>
+    </tr>
+    </table>
+
+    <br>
+
+    <table width="880" border="0" align="center" cellspacing="0" cellpadding="0">
+    <tr>
+        <td style="width:35%; vertical-align:middle; padding-right: 20px;">
+            <div class="image-container">
+                <img src='publications/2025_TWIST2.gif' width="90%">
+            </div>
+        </td>
+        <td style="width:65%; vertical-align:middle">
+            <papertitle>TWIST2: Scalable, Portable, and Holistic Humanoid Data Collection System</papertitle>
+            <br>
+            Yanjie Ze, Siheng Zhao, Weizhuo Wang, Angjoo Kanazawa<sup>&dagger;</sup>, Rocky Duan<sup>&dagger;</sup>, Pieter Abbeel<sup>&dagger;</sup>, Guanya Shi<sup>&dagger;</sup>, Jiajun Wu<sup>&dagger;</sup>, C. Karen Liu<sup>&dagger;</sup>
+            <br>
+            <a href="https://arxiv.org/abs/2511.02832" target="_blank"><i class="far fa-file"></i> paper</a> &nbsp
+            <a href="https://yanjieze.com/TWIST2/" target="_blank"><i class="fas fa-globe"></i> website</a> &nbsp
+            <a href="https://github.com/amazon-far/TWIST2" target="_blank"><i class="fas fa-code"></i> code</a> 
+            <p style="margin-top: 5px"><i class="fas fa-comment-dots"></i> TL;DR: TWIST2 is a portable whole-body humanoid teleoperation system that enables scalable data collection.
+        </td>
+    </tr>
+    </table>
+
+    <br>
+
+    <table width="880" border="0" align="center" cellspacing="0" cellpadding="0">
+    <tr>
+        <td style="width:35%; vertical-align:middle; padding-right: 20px;">
+            <div class="image-container">
+                <img src='publications/2025_ResMimic.gif' width="90%">
+            </div>
+        </td>
+        <td style="width:65%; vertical-align:middle">
+            <papertitle>ResMimic: From General Motion Tracking to Humanoid Whole-body Loco-Manipulation via Residual Learning</papertitle>
+            <br>
+            Siheng Zhao, Yanjie Ze, Yue Wang, C. Karen Liu<sup>&dagger;</sup>, Pieter Abbeel<sup>&dagger;</sup>, Guanya Shi<sup>&dagger;</sup>, Rocky Duan<sup>&dagger;</sup>
+            <br>
+            <a href="https://www.arxiv.org/abs/2510.05070" target="_blank"><i class="far fa-file"></i> paper</a> &nbsp
+            <a href="https://resmimic.github.io/" target="_blank"><i class="fas fa-globe"></i> website</a> 
+            <p style="margin-top: 5px"><i class="fas fa-comment-dots"></i> TL;DR: A two-stage humanoid skill learning framework that learns a residual policy on top of a pre-trained general motion tracking policy.
+        </td>
+    </tr>
+    </table>
+
+    <br>
+
     <table width="880" border="0" align="center" cellspacing="0" cellpadding="0">
     <tr style="background-color: var(--highlight-color)">
         <td style="width:35%; vertical-align:middle; padding-right: 20px;">