| Method | ${group.headers.map(h=>`${h} | `).join('')}
|---|---|
| ${esc(name)}${group.indices[0]===8&&name==='Fast-WAM'?'†':''} | ${views.map((view,c)=>{const val=view.rows[r][1],formatted=val.toFixed(view.decimals);return `${val===ranks[c][0]?`${formatted}`:val===ranks[c][1]?`${formatted}`:formatted} | `;}).join('')}
Best in bold; second best underlined. † Fast-WAM on RoboCasa–GR1 is our reproduction.
`; const componentViews=datasets.components.views; const componentMax=componentViews.map(v=>Math.max(...v.rows.map(r=>r[1]))); const configurations=[[false,false],[true,false],[false,true],[true,true]]; const mark=enabled=>`${enabled?'✓':'✗'}`; document.querySelector('#component-table-panel').innerHTML=`| Components | LIBERO | Real-world | |||||
|---|---|---|---|---|---|---|---|
| WING-LAM | Spectral guidance | Spatial | Object | Goal | Long | Standard | Gen. |
| ${mark(lam)} | ${mark(dct)} | ${componentViews.map((view,c)=>{const value=view.rows[r][1];return `${value===componentMax[c]?`${value.toFixed(1)}`:value.toFixed(1)} | `;}).join('')}|||||
Together, the two components improve real-world success by +7.5 pp in standard settings and +9.0 pp under generalization, compared with using neither.
`; const pretraining=document.querySelector('#pretraining-comparison'); const preMin=[90,95,90,80,55,25]; const preMax=[100,100,100,100,80,55]; const preNames=['Hand-pose','Video-only','Latent action','Latent action','WING','WING']; pretraining.innerHTML=`200 hours of egocentric pretraining per strategy · Success rate (%)
| Pretraining strategy | ${indices.map(i=>`${suites[i].split(' · ')[1]}${i<4?'LIBERO':'Real-world'} · ${preMin[i]}–${preMax[i]}% | `).join('')}
|---|---|
| ${name}${r===2||r===4?'without debiasing':''} | ${indices.map(i=>{const v=datasets.pretraining.views[i].rows[r][1];return ``;}).join('')} |
| ${indices.map(i=>` | ${preMin[i]}%${(preMin[i]+preMax[i])/2}%${preMax[i]}% | `).join('')}
Axes are truncated to show differences; each panel’s range is labeled. Hover, focus, or tap a bar for details.
No debiasing: DreamDojo replaces WING-LAM. Pretraining targets: latent actions vs. spectral guidance (WING).
`; const preBars=[...pretraining.querySelectorAll('.pretraining-bar')]; preBars.forEach(button=>{ const activate=()=>{preBars.forEach(b=>b.classList.toggle('is-selected',b===button));pretraining.querySelector('.pretraining-readout').textContent=button.dataset.detail;}; button.addEventListener('pointerenter',activate);button.addEventListener('focus',activate);button.addEventListener('click',activate); button.addEventListener('pointerleave',()=>button.classList.remove('is-selected'));button.addEventListener('blur',()=>button.classList.remove('is-selected')); }); const realPanel=document.querySelector('#real-world-panel'); const realColors=['#c5cac0','#a7b09f','#7d8b77','#347eab']; realPanel.innerHTML=`${metric?'Progress score · 0–10':'Success rate · %'}
Swipe horizontally to see both settings →
Four tasks · Four methods · Hover, focus, or tap a bar for details.
Success measures full task completion; progress credits partial completion on a 0–10 scale. Generalization uses the precise averages from Table 12.
`; const realBars=[...realPanel.querySelectorAll('.real-bar')]; realBars.forEach(button=>{ const activate=()=>{realBars.forEach(b=>b.classList.toggle('is-selected',b===button));realPanel.querySelector('.real-detail').textContent=button.dataset.detail;}; button.addEventListener('pointerenter',activate);button.addEventListener('focus',activate);button.addEventListener('click',activate); const clear=()=>{button.classList.remove('is-selected');};button.addEventListener('pointerleave',clear);button.addEventListener('blur',clear); }); document.querySelectorAll('[data-chart]').forEach((card) => { const key=card.dataset.chart, data=datasets[key], id='chart-view-'+key+(card.dataset.viewIndex!==undefined?'-'+card.dataset.viewIndex:''); card.innerHTML=`Swipe horizontally to compare all methods →
${data.note}
`; const select=card.querySelector('select'), plot=card.querySelector('.result-plot'), readout=card.querySelector('.result-readout'); if (card.dataset.viewIndex !== undefined) { select.value=card.dataset.viewIndex; card.querySelector('.result-heading h4').textContent=data.views[Number(select.value)].label.replace('LIBERO · ', ''); } function render() { const view=data.views[Number(select.value)], fmt=v=>v.toFixed(view.decimals)+view.unit; const min=view.min ?? 0, range=view.max-min; const ticks=view.ticks ?? [0,.2,.4,.6,.8,1].map(f=>min+f*range); const focus=view.highlight || (key==='semantics'?'WING-LAM':'WING'); card.querySelector('.result-view-label').textContent=view.label; plot.style.setProperty('--chart-columns',view.rows.length); plot.style.minWidth=card.dataset.viewIndex!==undefined?'0':`${Math.max(360,view.rows.length*(card.closest('.result-pair')?56:91)+44)}px`; plot.innerHTML=``; const defaultText=key==='semantics'?'Accuracy (%). Hover, focus, or tap a bar for details.':key==='bandwidth'?'SR (%). Hover, focus, or tap a bar for its exact value.':`${view.unit==='%'?(key==='semantics'?'Classification accuracy (%)':'SR (%)'):'Semantic similarity'} · ${min}–${view.max} scale${min>0?' (truncated)':''}. Hover, focus, or tap a bar for its exact value.`; readout.textContent=defaultText; const bars=[...plot.querySelectorAll('.result-bar')]; function highlight(button,i){bars.forEach(b=>b.classList.toggle('is-selected',b===button));readout.textContent=`${view.rows[i][0]} — ${fmt(view.rows[i][1])} · ${view.label}`;} bars.forEach((button,i)=>{button.addEventListener('pointerenter',()=>highlight(button,i));button.addEventListener('focus',()=>highlight(button,i));button.addEventListener('click',()=>highlight(button,i));button.addEventListener('pointerleave',()=>{if(document.activeElement!==button){button.classList.remove('is-selected');readout.textContent=defaultText;}});button.addEventListener('blur',()=>{button.classList.remove('is-selected');readout.textContent=defaultText;});}); } select.addEventListener('change',render);render(); }); })(); // Figure 3(b): exact means and standard deviations from graph/camera_sensitivity_with_std.py. const cameraSensitivityData = {"Ours": {"mean": [0.0025, 0.0053, 0.0117, 0.0288, 0.0655], "std": [0.0004, 0.0009, 0.002, 0.0043, 0.0072]}, "CD-lam": {"mean": [0.0314, 0.0617, 0.1186, 0.218, 0.3782], "std": [0.0054, 0.01, 0.0185, 0.0287, 0.0358]}, "LAPA": {"mean": [0.0982, 0.1628, 0.3108, 0.516, 0.6933], "std": [0.0168, 0.0306, 0.0517, 0.0643, 0.0781]}, "UniVLA": {"mean": [0.0236, 0.0792, 0.1534, 0.3043, 0.5808], "std": [0.0108, 0.0197, 0.028, 0.0542, 0.0763]}, "DreamDojo": {"mean": [0.0335, 0.0649, 0.1233, 0.2218, 0.3797], "std": [0.0061, 0.011, 0.0197, 0.0304, 0.0369]}, "villa-X": {"mean": [0.085, 0.1587, 0.2869, 0.4887, 0.7512], "std": [0.0152, 0.022, 0.0308, 0.0421, 0.0616]}}; (() => { const motion=document.querySelector('#motion-decodability'); const points=[['WING teacher',117,54,126,41,true],['WING student',146,97,153,88,true],['DreamDojo',299,172,246,158,false],['CD-LAM',317,194,270,225,false],['UniVLA',132,200,99,184,false],['LAPA',112,216,70,237,false],['VILLA-X',163,242,166,264,false]]; motion.innerHTML=`Motion decodability · MLP probes
WING retains interaction information with less camera-motion information.
Preserve interaction motion; suppress camera motion.
`; motion.querySelectorAll('.diagnostic-point').forEach(point=>{ const show=()=>{motion.querySelectorAll('.diagnostic-point').forEach(p=>p.classList.toggle('is-selected',p===point));motion.querySelector('.result-readout').textContent=point.dataset.method+' · Camera-motion readout on the horizontal axis; interaction-motion readout on the vertical axis.';}; point.addEventListener('pointerenter',show);point.addEventListener('focus',show);point.addEventListener('click',show); point.addEventListener('pointerleave',()=>point.classList.remove('is-selected'));point.addEventListener('blur',()=>point.classList.remove('is-selected')); }); const card=document.querySelector('#camera-sensitivity'); const labels={'Ours':'WING','CD-lam':'CD-LAM','villa-X':'VILLA-X'}; const entries=Object.entries(cameraSensitivityData); const colors=['#347eab','#505b50','#758371','#939f89','#a9afa4','#bcc2b5']; const x=i=>60+i*72, y=v=>263-v/0.85*225; card.innerHTML=`Camera sensitivity · Lower is better
Select a legend entry or point to inspect a method.
Lines: mean · Shaded bands: ±1 SD.
`; function selectSeries(index){card.querySelectorAll('.camera-series').forEach(g=>g.style.opacity=index===null||g.dataset.series===index?'1':'.2');card.querySelectorAll('.camera-legend button').forEach(b=>b.setAttribute('aria-pressed',String(b.dataset.series===index)));} card.querySelectorAll('.camera-legend button').forEach(button=>button.addEventListener('click',()=>{const next=button.getAttribute('aria-pressed')==='true'?null:button.dataset.series;selectSeries(next);card.querySelector('.result-readout').textContent=next===null?'All methods · Mean ±1 standard deviation':button.textContent+' · Select a point for its exact mean and standard deviation.';})); card.querySelectorAll('.camera-point').forEach(point=>{const show=()=>{selectSeries(point.parentElement.dataset.series);card.querySelector('.result-readout').textContent=point.dataset.readout;};point.addEventListener('pointerenter',show);point.addEventListener('focus',show);point.addEventListener('click',show);}); })();