index.xml

<?xml version="1.0" encoding="utf-8" standalone="yes" ?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Stefanos Charalambous</title>
    <link>https://scharalambous3.github.io/</link>
      <atom:link href="https://scharalambous3.github.io/index.xml" rel="self" type="application/rss+xml" />
    <description>Stefanos Charalambous</description>
    <generator>Wowchemy (https://wowchemy.com)</generator><language>en-us</language>
    <image>
      <url>https://scharalambous3.github.io/media/icon_hub7a1c745800879165a56661b27a6324f_21626_512x512_fill_lanczos_center_3.png</url>
      <title>Stefanos Charalambous</title>
      <link>https://scharalambous3.github.io/</link>
    </image>
    
    <item>
      <title>Reinforcement Learning-Guided MPC for Whole-Body Loco-Manipulation</title>
      <link>https://scharalambous3.github.io/project/hrlmpc/</link>
      <pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate>
      <guid>https://scharalambous3.github.io/project/hrlmpc/</guid>
      <description>&lt;h2 id=&#34;framework&#34;&gt;Framework&lt;/h2&gt;
&lt;p&gt;We introduce the first framework that combines RL and whole-body MPC, where the MPC is embedded in training. The high-level planner is trained using RL to provide task-space references to a low-level MPC. We are thus able to leverage the benefits of both learning- and optimization-based methods including safety guarantees, great generalization, minimum sim2real gap and data-efficient training.&lt;/p&gt;
&lt;p&gt;















&lt;figure  &gt;
  &lt;div class=&#34;d-flex justify-content-center&#34;&gt;
    &lt;div class=&#34;w-100&#34; &gt;&lt;img alt=&#34;Framework&#34;
           src=&#34;https://scharalambous3.github.io/project/hrlmpc/static/image/ThreeStageFramework.gif&#34;
           loading=&#34;lazy&#34; data-zoomable /&gt;&lt;/div&gt;
  &lt;/div&gt;&lt;/figure&gt;

&lt;em&gt;During a pre-training stage, we use a simplified kinematic model for fast simulation. When training with MPC, we use the MPC solution as our simulator to avoid the additional cost of physics simulation. During deployment, the RL is deployed in closed-loop with the MPC.&lt;/em&gt;&lt;/p&gt;
&lt;div class=&#34;row&#34;&gt;
&lt;img src=&#34;static/image/MPCGetsStuckFaster.gif&#34; alt=&#34;Github stars&#34; width=&#34;350&#34;/&gt;
&lt;img src=&#34;static/image/RLHelpsMPCGetUnstuck2.gif&#34; alt=&#34;Language&#34; width=&#34;350&#34;/&gt;
&lt;/div&gt;
&lt;p&gt;&lt;em&gt;Deploying MPC with the planner trained using RL (Right) alleviates the issue of getting stuck in a local minimum  when deploying MPC alone (Left) due to its shortsightedness.&lt;/em&gt;&lt;/p&gt;
&lt;h2 id=&#34;results&#34;&gt;Results&lt;/h2&gt;
&lt;p&gt;We are currently working to apply our framework in whole-body navigation to target end-effector positions in cluttered scenes in the real-world.&lt;/p&gt;
&lt;p&gt;















&lt;figure  &gt;
  &lt;div class=&#34;d-flex justify-content-center&#34;&gt;
    &lt;div class=&#34;w-100&#34; &gt;&lt;img alt=&#34;BoxBoxWallScene&#34;
           src=&#34;https://scharalambous3.github.io/project/hrlmpc/static/image/SceneOptimized.gif&#34;
           loading=&#34;lazy&#34; data-zoomable /&gt;&lt;/div&gt;
  &lt;/div&gt;&lt;/figure&gt;
&lt;/p&gt;
</description>
    </item>
    
    <item>
      <title></title>
      <link>https://scharalambous3.github.io/admin/config.yml</link>
      <pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate>
      <guid>https://scharalambous3.github.io/admin/config.yml</guid>
      <description></description>
    </item>
    
    <item>
      <title>Contact-Implicit Planning for Athletic, Contact-Rich Tasks</title>
      <link>https://scharalambous3.github.io/project/cito/</link>
      <pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate>
      <guid>https://scharalambous3.github.io/project/cito/</guid>
      <description>&lt;h2 id=&#34;framework&#34;&gt;Framework&lt;/h2&gt;
&lt;p&gt;We present a continuation method for contact-implicit trajectory optimization that works with minimal guidance i.e. without a handcrafted initialization nor references. We are able to automatically discover motions and contact schedules given only a terminal goal specified by the user. The generated motion plans are then tracked by a whole-body MPC.&lt;/p&gt;
&lt;p&gt;















&lt;figure  &gt;
  &lt;div class=&#34;d-flex justify-content-center&#34;&gt;
    &lt;div class=&#34;w-100&#34; &gt;&lt;img alt=&#34;Framework&#34; srcset=&#34;
               /project/cito/static/image/Framework_hu105e8e2be2e30de96f5da3257add3649_93742_51b5f66460dbc6c87a15357ba17599f1.webp 400w,
               /project/cito/static/image/Framework_hu105e8e2be2e30de96f5da3257add3649_93742_d9dcc7b5b2993d9a2a7d06e0a6a4a05b.webp 760w,
               /project/cito/static/image/Framework_hu105e8e2be2e30de96f5da3257add3649_93742_1200x1200_fit_q75_h2_lanczos_3.webp 1200w&#34;
               src=&#34;https://scharalambous3.github.io/project/cito/static/image/Framework_hu105e8e2be2e30de96f5da3257add3649_93742_51b5f66460dbc6c87a15357ba17599f1.webp&#34;
               width=&#34;760&#34;
               height=&#34;356&#34;
               loading=&#34;lazy&#34; data-zoomable /&gt;&lt;/div&gt;
  &lt;/div&gt;&lt;/figure&gt;
&lt;/p&gt;
&lt;p&gt;Given only a commanded position (red sphere) and orientation (green arrow) the whole-body motion plan is optimized including gait, foothold positions and base pose references.&lt;/p&gt;
&lt;div class=&#34;row&#34;&gt;
&lt;img src=&#34;static/image/obstacle_xfurther.gif&#34; alt=&#34;Github stars&#34; width=&#34;400&#34;/&gt;
&lt;img src=&#34;static/image/obstacle_xfurther2.gif&#34; alt=&#34;Language&#34; width=&#34;250&#34;/&gt;
&lt;/div&gt;
&lt;h2 id=&#34;results-on-hardware&#34;&gt;Results on hardware&lt;/h2&gt;
&lt;p&gt;Our novel contact-implicit trajectory optimization formulation in the motion refinement stage gives us a direct handle on the contact modes and is a crucial tool in overcoming the sim2real gap. In addition, through the use of a continuation method the generated motion plan has strict physical feasibility.To prove both of these claims, we tested the motion plans generated by our framework on hardware for a complex, under-actuated legged system.&lt;/p&gt;
&lt;p&gt;















&lt;figure  &gt;
  &lt;div class=&#34;d-flex justify-content-center&#34;&gt;
    &lt;div class=&#34;w-100&#34; &gt;&lt;img alt=&#34;HWTest_Highlight&#34;
           src=&#34;https://scharalambous3.github.io/project/cito/static/image/HWTest_Highlight.gif&#34;
           loading=&#34;lazy&#34; data-zoomable /&gt;&lt;/div&gt;
  &lt;/div&gt;&lt;/figure&gt;












  





&lt;video controls  &gt;
  &lt;source src=&#34;https://scharalambous3.github.io/project/cito/static/image/4.mp4&#34; type=&#34;video/mp4&#34;&gt;
&lt;/video&gt;
&lt;/p&gt;
&lt;h3 id=&#34;report&#34;&gt;Report&lt;/h3&gt;
&lt;script type=&#34;text/javascript&#34; src= &#39;/js/pdf-js/build/pdf.js&#39;&gt;&lt;/script&gt;

&lt;style&gt;
  #embed-pdf-container {
    position: relative;
    width: 100%;
    height: auto;
    min-height: 20vh;
     
  }
  
  .pdf-canvas {
    border: 1px solid black;
    direction: ltr;
    width: 100%;
    height: auto;
    display: none;
  }
  
  #the-canvas {
    border: 1px solid black;
    direction: ltr;
    width: 100%;
    height: auto;
    display: none;
  }
  
  
  .pdf-loadingWrapper {
    display: none;
    justify-content: center;
    align-items: center;
    width: 100%;
    height: 350px;
  }
  
  .pdf-loading {
    display: inline-block;
    width: 50px;
    height: 50px;
    border: 3px solid #d2d0d0;;
    border-radius: 50%;
    border-top-color: #383838;
    animation: spin 1s ease-in-out infinite;
    -webkit-animation: spin 1s ease-in-out infinite;
  }
  
  
  
  
  
  #overlayText {
    word-wrap: break-word;
    display: grid;
    justify-content: end;
  }
  
  #overlayText a {
    position: relative;
    top: 10px;
    right: 4px;
    color: #000;
    margin: auto;
    background-color: #eeeeee;
    padding: 0.3em 1em;
    border: solid 2px;
    border-radius: 12px;
    border-color: #00000030;
    text-decoration: none;
  }
  
  #overlayText svg {
    height: clamp(1em, 2vw, 1.4em);
    width:  clamp(1em, 2vw, 1.4em);
  }
  
  
  
  @keyframes spin {
    to { -webkit-transform: rotate(360deg); }
  }
  @-webkit-keyframes spin {
    to { -webkit-transform: rotate(360deg); }
  }
  &lt;/style&gt;&lt;div class=&#34;embed-pdf-container&#34; id=&#34;embed-pdf-container-3a71ce23&#34;&gt;
    &lt;div class=&#34;pdf-loadingWrapper&#34; id=&#34;pdf-loadingWrapper-3a71ce23&#34;&gt;
        &lt;div class=&#34;pdf-loading&#34; id=&#34;pdf-loading-3a71ce23&#34;&gt;&lt;/div&gt;
    &lt;/div&gt;
    &lt;div id=&#34;overlayText&#34;&gt;
      &lt;a href=&#34;static/image/CITO_Paper.pdf&#34; aria-label=&#34;Download&#34; download&gt;
        &lt;svg aria-hidden=&#34;true&#34; xmlns=&#34;http://www.w3.org/2000/svg&#34; viewBox=&#34;0 0 18 18&#34;&gt;
            &lt;path d=&#34;M9 13c.3 0 .5-.1.7-.3L15.4 7 14 5.6l-4 4V1H8v8.6l-4-4L2.6 7l5.7 5.7c.2.2.4.3.7.3zm-7 2h14v2H2z&#34; /&gt;
        &lt;/svg&gt;
      &lt;/a&gt;
    &lt;/div&gt;
    &lt;canvas class=&#34;pdf-canvas&#34; id=&#34;pdf-canvas-3a71ce23&#34;&gt;&lt;/canvas&gt;
&lt;/div&gt;

&lt;div class=&#34;pdf-paginator&#34; id=&#34;pdf-paginator-3a71ce23&#34;&gt;
    &lt;button id=&#34;pdf-prev-3a71ce23&#34;&gt;Previous&lt;/button&gt;
    &lt;button id=&#34;pdf-next-3a71ce23&#34;&gt;Next&lt;/button&gt; &amp;nbsp; &amp;nbsp;
    &lt;span&gt;
      &lt;span class=&#34;pdf-pagenum&#34; id=&#34;pdf-pagenum-3a71ce23&#34;&gt;&lt;/span&gt; / &lt;span class=&#34;pdf-pagecount&#34; id=&#34;pdf-pagecount-3a71ce23&#34;&gt;&lt;/span&gt;
    &lt;/span&gt;
    &lt;a class=&#34;pdf-source&#34; id=&#34;pdf-source-3a71ce23&#34; href=&#34;static/image/CITO_Paper.pdf&#34;&gt;[pdf]&lt;/a&gt;
&lt;/div&gt;

&lt;noscript&gt;
View the PDF file &lt;a class=&#34;pdf-source&#34; id=&#34;pdf-source-noscript-3a71ce23&#34; href=&#34;static/image/CITO_Paper.pdf&#34;&gt;here&lt;/a&gt;.
&lt;/noscript&gt;

&lt;script type=&#34;text/javascript&#34;&gt;
    (function(){
    var url = &#39;static\/image\/CITO_Paper.pdf&#39;;

    var hidePaginator = &#34;&#34; === &#34;true&#34;;
    var hideLoader = &#34;&#34; === &#34;true&#34;;
    var selectedPageNum = parseInt(&#34;&#34;) || 1;

    
    var pdfjsLib = window[&#39;pdfjs-dist/build/pdf&#39;];

    
    if (pdfjsLib.GlobalWorkerOptions.workerSrc == &#39;&#39;)
      pdfjsLib.GlobalWorkerOptions.workerSrc = &#34;https:\/\/scharalambous3.github.io\/&#34; + &#39;js/pdf-js/build/pdf.worker.js&#39;;

    
    var pdfDoc = null,
        pageNum = selectedPageNum,
        pageRendering = false,
        pageNumPending = null,
        scale = 3,
        canvas = document.getElementById(&#39;pdf-canvas-3a71ce23&#39;),
        ctx = canvas.getContext(&#39;2d&#39;),
        paginator = document.getElementById(&#34;pdf-paginator-3a71ce23&#34;),
        loadingWrapper = document.getElementById(&#39;pdf-loadingWrapper-3a71ce23&#39;);


    
    showPaginator();
    showLoader();

    

    function renderPage(num) {
      pageRendering = true;
      
      pdfDoc.getPage(num).then(function(page) {
        var viewport = page.getViewport({scale: scale});
        canvas.height = viewport.height;
        canvas.width = viewport.width;

        
        var renderContext = {
          canvasContext: ctx,
          viewport: viewport
        };
        var renderTask = page.render(renderContext);

        
        renderTask.promise.then(function() {
          pageRendering = false;
          showContent();

          if (pageNumPending !== null) {
            
            renderPage(pageNumPending);
            pageNumPending = null;
          }
        });
      });

      
      document.getElementById(&#39;pdf-pagenum-3a71ce23&#39;).textContent = num;
    }

    

    function showContent() {
      loadingWrapper.style.display = &#39;none&#39;;
      canvas.style.display = &#39;block&#39;;
    }

    

    function showLoader() {
      if(hideLoader) return
      loadingWrapper.style.display = &#39;flex&#39;;
      canvas.style.display = &#39;none&#39;;
    }

    

    function showPaginator() {
      if(hidePaginator) return
      paginator.style.display = &#39;block&#39;;
    }

    

    function queueRenderPage(num) {
      if (pageRendering) {
        pageNumPending = num;
      } else {
        renderPage(num);
      }
    }

    

    function onPrevPage() {
      if (pageNum &lt;= 1) {
        return;
      }
      pageNum--;
      queueRenderPage(pageNum);
    }
    document.getElementById(&#39;pdf-prev-3a71ce23&#39;).addEventListener(&#39;click&#39;, onPrevPage);

    

    function onNextPage() {
      if (pageNum &gt;= pdfDoc.numPages) {
        return;
      }
      pageNum++;
      queueRenderPage(pageNum);
    }
    document.getElementById(&#39;pdf-next-3a71ce23&#39;).addEventListener(&#39;click&#39;, onNextPage);

    

    pdfjsLib.getDocument(url).promise.then(function(pdfDoc_) {
      pdfDoc = pdfDoc_;
      var numPages = pdfDoc.numPages;
      document.getElementById(&#39;pdf-pagecount-3a71ce23&#39;).textContent = numPages;

      
      if(pageNum &gt; numPages) {
        pageNum = numPages
      }

      
      renderPage(pageNum);
    });
    })();
&lt;/script&gt;

</description>
    </item>
    
  </channel>
</rss>