Add Kinex pose checkpoint results with original RoboTwin native instructions
Browse files- README.md +9 -0
- agent-versions.json +8 -0
- app.js +15 -2
- data-source.json +5 -3
- index.html +1 -1
- manifest.json +97 -93
- native-pose-publication.json +59 -0
- results/data.json +0 -0
- results/sources.json +7 -0
README.md
CHANGED
|
@@ -21,3 +21,12 @@ Mode 1 uses instructions and images with pose conditioning disabled. Mode 2 adds
|
|
| 21 |
Historical Kinex 82% and Kinex + VLAct 62% retain their RLE instructions and are shown separately. The former standalone RLE result, 11/50 (22%), remains archived. The original 32% result shares 46 of 50 scene seeds with the new campaign and is not substituted for its fresh baseline.
|
| 22 |
|
| 23 |
See the on-page protocol, native instruction inventory and pinned sources for evidence and comparison limits.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
Historical Kinex 82% and Kinex + VLAct 62% retain their RLE instructions and are shown separately. The former standalone RLE result, 11/50 (22%), remains archived. The original 32% result shares 46 of 50 scene seeds with the new campaign and is not substituted for its fresh baseline.
|
| 22 |
|
| 23 |
See the on-page protocol, native instruction inventory and pinned sources for evidence and comparison limits.
|
| 24 |
+
|
| 25 |
+
## Pose-conditioned native-instruction evaluation
|
| 26 |
+
|
| 27 |
+
| Condition | Native success | Coverage |
|
| 28 |
+
| --- | ---: | ---: |
|
| 29 |
+
| Kinex + VLAct TCP-PoseState | 0 / 0 | 0 / 50 |
|
| 30 |
+
| Kinex + progVLA pose_base | 0 / 0 | 0 / 50 |
|
| 31 |
+
|
| 32 |
+
Both pose checkpoints use original native unseen RoboTwin instructions, GPT-6 Astra/high, all 50 Hard tasks, the unchanged authored initialization seeds, and native budgets of 400–1,700 control frames at 25 Hz. Progress heads are disabled. Kinex chooses TCP subgoals and executes each returned four-tick command sequence. Native instructions are preserved in the recorded model requests. All 50 tasks are scheduled for both conditions; pending episodes are excluded from success rates. Engineering checks are excluded from scores. Other published protocols remain individually labeled.
|
agent-versions.json
CHANGED
|
@@ -23,6 +23,14 @@
|
|
| 23 |
"revision": "dcf0f1d7239b28ae81a039c8d05bdfcd1d4792a5",
|
| 24 |
"detail": "Released Clean 50k checkpoint; official standalone with current RLE task13 instructions, evaluated 2026-11-13."
|
| 25 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
}
|
| 27 |
}
|
| 28 |
}
|
|
|
|
| 23 |
"revision": "dcf0f1d7239b28ae81a039c8d05bdfcd1d4792a5",
|
| 24 |
"detail": "Released Clean 50k checkpoint; official standalone with current RLE task13 instructions, evaluated 2026-11-13."
|
| 25 |
}
|
| 26 |
+
},
|
| 27 |
+
"research/kinex-pose-hard-native-20261113/source-identities.json": {
|
| 28 |
+
"kinex": {
|
| 29 |
+
"label": "Kinex v1.1.0",
|
| 30 |
+
"version": "1.1.0",
|
| 31 |
+
"revision": "f1102c0f6adb55859018c26c1a3aab0130bc39d3",
|
| 32 |
+
"detail": "Frozen runtime; exact workspace hashes retained."
|
| 33 |
+
}
|
| 34 |
}
|
| 35 |
}
|
| 36 |
}
|
app.js
CHANGED
|
@@ -26,7 +26,7 @@ const pct = v => v == null ? '—' : (v*100).toFixed(1)+'%';
|
|
| 26 |
const elapsed = v => v == null ? '—' : v >= 3600 ? (v/3600).toFixed(1)+' h' : v >= 60 ? (v/60).toFixed(1)+' min' : v.toFixed(1)+' s';
|
| 27 |
const arrow = '<span aria-hidden="true">↗</span>';
|
| 28 |
// Text-target Clean 50k presentation.
|
| 29 |
-
const colors = {'native-baseline':'#5a6b9c','pose-base-mode1':'#308f85','pose-base-mode2':'#20685f','jy-mode1':'#b77b42','jy-mode2':'#865325','kinex-vlact50k-input-required-text':'#7a6940','kinex-vlact50k-autonomous':'#4674a3','kinex-vlact50k-input-required':'#a96682','kinex-vla-input-required':'#b35349','kinex-vla-autonomous':'#367d96','kinex-vla-input-guided':'#96703b','standalone-oft-all140k':'#547a36','standalone-vlact-clean':'#a75f7f','standalone-vlact-all':'#bd733b','standalone-official':'#39819c','standalone':'#39819c','kinex_policy':'#756ee4','kinex':'#27847a','codex-astra-spectator':'#a76a34','codex-astra':'#39819c','kinex-astra':'#756ee4','cc-opus':'#b48566','kinex-opus':'#958bc8','cc-opus-vertex-high':'#c9875a','cc-opus-vertex-xhigh':'#8c5a3a','kinex-opus-vertex-high':'#5b8bd8','kinex-opus-vertex-xhigh':'#3a5ba6','cc-opus-vertex-high-run2':'#e0a37a','cc-opus-vertex-xhigh-run2':'#a8704a','kinex-opus-vertex-high-run2':'#86aae6','kinex-opus-vertex-xhigh-run2':'#5876c0'};
|
| 30 |
let data, epoch = 0, current = {}, matrixMetric = 'success';
|
| 31 |
let filters = {query:'',family:'',status:'',variant:'standalone-vlact-clean',sort:'catalog'};
|
| 32 |
let sessionState = {filter:'all',query:'',count:40,view:'timeline'};
|
|
@@ -51,6 +51,7 @@ const methodLabel = v => ({
|
|
| 51 |
}[v.id] || v.label || v.agent);
|
| 52 |
const protocolLabel = v => v.id==='kinex-vlact50k-input-required-text' ? 'Protocol: input-required · text-targets-v1' : v.id==='kinex-vlact50k-input-required' ? 'Protocol: input-required · numeric-target-text-v1' : v.id==='kinex-vlact50k-autonomous' ? 'Protocol: autonomous · verbatim-text-v1' : v.id==='kinex-vla-input-required' ? 'Protocol: five JSON fields · nullable targets' : '';
|
| 53 |
function checkpointDetails(v) {
|
|
|
|
| 54 |
if(!v.checkpoint_repo)return '';
|
| 55 |
return `<aside class="native-note" data-checkpoint-card><span class="note-symbol">i</span><div><strong>VLAct Clean 50k · ${esc(v.agent_protocol)}</strong><p>${esc(v.checkpoint_repo)} · step ${number(v.checkpoint_step)} · ${esc(v.training_distribution)}. Agent: GPT-6 Astra / high.</p><p>${esc(protocolLabel(v))}. VLA invocation remains optional.</p><div class="inline-links">${external(v.model_card,'Model card & protocol')}${external(v.checkpoint,'Pinned checkpoint & hashes')}</div></div></aside>`;
|
| 56 |
}
|
|
@@ -88,8 +89,9 @@ function usageTiles(s,includeCost=true) {
|
|
| 88 |
function modelGroups() {
|
| 89 |
return [
|
| 90 |
{key:'baselines', model:'Native instructions · paired 50k evaluation', note:'Same scenes and instructions · 50 tasks per condition', variants:data.variants.filter(v=>v.paired_campaign)},
|
|
|
|
| 91 |
{key:'historical-native', model:'Restored native baseline', note:'Original 16/50 result · historical seeds retained', variants:data.variants.filter(v=>v.historical_reference)},
|
| 92 |
-
{key:'historical-rle', model:'Historical RLE agent results', note:'Different instruction and execution protocol', variants:data.variants.filter(v=>!v.paired_campaign&&!v.historical_reference)}
|
| 93 |
];
|
| 94 |
}
|
| 95 |
let expandedModels=null;
|
|
@@ -524,6 +526,7 @@ async function loadPolicy(ep,token) {
|
|
| 524 |
function jsonBlock(title,value,open=false) {return `<details class="tool-block" ${open?'open':''}><summary>${esc(title)}</summary><pre>${esc(JSON.stringify(value,null,2))}</pre></details>`;}
|
| 525 |
function renderPolicy(index) {
|
| 526 |
const record=policyRecord,c=record.calls[index],ep=current.ep;
|
|
|
|
| 527 |
if(ep.provider==='robotwin-official')return renderOfficialPolicy(index);
|
| 528 |
if(!c){$('#policy-body').innerHTML='<div class="empty-state">No VLA inference was made in this episode.</div>';return;}
|
| 529 |
const images=c.images.map(im=>`<figure><a href="${esc(im.model_image)}" target="_blank"><img src="${esc(im.model_image)}" alt="${esc(im.camera)} model input, call ${index+1}"></a><figcaption>${esc(im.camera)} · ${im.model_size.join(' × ')}<br><a href="${esc(im.recorded)}" target="_blank">Recorded RGB · ${im.recorded_size.join(' × ')} ↗</a></figcaption></figure>`).join('');
|
|
@@ -566,3 +569,13 @@ function renderOfficialPolicy(index) {
|
|
| 566 |
$('#policy-body').innerHTML=`<label>Inference <select class="policy-select" id="policy-call" aria-label="VLA inference">${r.calls.map((x,i)=>`<option value="${i}" ${i===index?'selected':''}>${i+1} / ${r.calls.length} · action ${x.native_step}</option>`).join('')}</select></label><div class="inline-links"><code>${esc(c.call_id)}</code><span>${c.service_response.inference_s.toFixed(3)} s · ${(c.service_response.waypoints||c.service_response.actions).length} × ${(c.service_response.waypoints||c.service_response.actions)[0].length} actions</span></div><h3>Task instruction</h3><div class="instruction-text">${esc(c.request.task)}</div><div class="policy-images">${ims}</div><p class="subtle">Exact transmitted RGB pixels. The recorded mode contract determines whether measured current and reference target TCP poses accompany the images.</p>${jsonBlock('Complete model request · image arrays linked as PNGs',c.request.wire_payload)}<h3>Native action chunk</h3><p class="subtle">Reordered model output consumed by the official evaluator. Absolute joint radians; gripper opening: 0 closed, 1 open.</p><div class="policy-table"><table><thead><tr><th>Action</th>${r.contract.output_layout.map(x=>`<th>${esc(x.replace('left_','L ').replace('right_','R '))}</th>`).join('')}</tr></thead><tbody>${(c.service_response.waypoints||c.service_response.actions).map((row,i)=>`<tr><th>${i+1}</th>${row.map(x=>`<td title="${x}">${x.toFixed(5)}</td>`).join('')}</tr>`).join('')}</tbody></table></div>${jsonBlock('Complete recorded StarVLA response · full precision',c.raw_response)}${jsonBlock('Reordered native actions · full precision',c.service_response.actions)}<p class="subtle">Standalone execution; no language-agent session or tool call. Token and API costs cover the language agent only; local GPU cost is not reported.</p>${jsonBlock('Recording coverage & model contract',r.contract)}<div class="inline-links">${external(r.contract.model_metadata,'Model configuration')}${external(r.contract.executed_sources,'Source identities')}${external(r.contract.adapter_source||'official/adapters.py','Policy adapter')}${external(r.contract.integration_source||'official/integrations.py','Official evaluator integration')}</div>`;
|
| 567 |
$('#policy-call').onchange=e=>renderPolicy(Number(e.target.value));
|
| 568 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
const elapsed = v => v == null ? '—' : v >= 3600 ? (v/3600).toFixed(1)+' h' : v >= 60 ? (v/60).toFixed(1)+' min' : v.toFixed(1)+' s';
|
| 27 |
const arrow = '<span aria-hidden="true">↗</span>';
|
| 28 |
// Text-target Clean 50k presentation.
|
| 29 |
+
const colors = {'kinex-tcp-posestate':'#7069ac','kinex-pose-base':'#ad7840','native-baseline':'#5a6b9c','pose-base-mode1':'#308f85','pose-base-mode2':'#20685f','jy-mode1':'#b77b42','jy-mode2':'#865325','kinex-vlact50k-input-required-text':'#7a6940','kinex-vlact50k-autonomous':'#4674a3','kinex-vlact50k-input-required':'#a96682','kinex-vla-input-required':'#b35349','kinex-vla-autonomous':'#367d96','kinex-vla-input-guided':'#96703b','standalone-oft-all140k':'#547a36','standalone-vlact-clean':'#a75f7f','standalone-vlact-all':'#bd733b','standalone-official':'#39819c','standalone':'#39819c','kinex_policy':'#756ee4','kinex':'#27847a','codex-astra-spectator':'#a76a34','codex-astra':'#39819c','kinex-astra':'#756ee4','cc-opus':'#b48566','kinex-opus':'#958bc8','cc-opus-vertex-high':'#c9875a','cc-opus-vertex-xhigh':'#8c5a3a','kinex-opus-vertex-high':'#5b8bd8','kinex-opus-vertex-xhigh':'#3a5ba6','cc-opus-vertex-high-run2':'#e0a37a','cc-opus-vertex-xhigh-run2':'#a8704a','kinex-opus-vertex-high-run2':'#86aae6','kinex-opus-vertex-xhigh-run2':'#5876c0'};
|
| 30 |
let data, epoch = 0, current = {}, matrixMetric = 'success';
|
| 31 |
let filters = {query:'',family:'',status:'',variant:'standalone-vlact-clean',sort:'catalog'};
|
| 32 |
let sessionState = {filter:'all',query:'',count:40,view:'timeline'};
|
|
|
|
| 51 |
}[v.id] || v.label || v.agent);
|
| 52 |
const protocolLabel = v => v.id==='kinex-vlact50k-input-required-text' ? 'Protocol: input-required · text-targets-v1' : v.id==='kinex-vlact50k-input-required' ? 'Protocol: input-required · numeric-target-text-v1' : v.id==='kinex-vlact50k-autonomous' ? 'Protocol: autonomous · verbatim-text-v1' : v.id==='kinex-vla-input-required' ? 'Protocol: five JSON fields · nullable targets' : '';
|
| 53 |
function checkpointDetails(v) {
|
| 54 |
+
if(v.input_contract==='images-current-target-tcp-xyzw-v1') return `<aside class="native-note" data-checkpoint-card><span class="note-symbol">i</span><div><strong>${esc(v.vla_model)} · native instruction</strong><p>${esc(v.checkpoint_repo)} / ${esc(v.checkpoint_filename)}. Agent: GPT-6 Astra / high.</p><p>Original RoboTwin unseen goal; Kinex plans numeric TCP targets. Progress head disabled. Fifty tasks, one fixed seed each. Model invocation remains optional.</p><div class="inline-links">${external(v.model_card,'Protocol')}${external(v.checkpoint,'Pinned checkpoint & hashes')}</div></div></aside>`;
|
| 55 |
if(!v.checkpoint_repo)return '';
|
| 56 |
return `<aside class="native-note" data-checkpoint-card><span class="note-symbol">i</span><div><strong>VLAct Clean 50k · ${esc(v.agent_protocol)}</strong><p>${esc(v.checkpoint_repo)} · step ${number(v.checkpoint_step)} · ${esc(v.training_distribution)}. Agent: GPT-6 Astra / high.</p><p>${esc(protocolLabel(v))}. VLA invocation remains optional.</p><div class="inline-links">${external(v.model_card,'Model card & protocol')}${external(v.checkpoint,'Pinned checkpoint & hashes')}</div></div></aside>`;
|
| 57 |
}
|
|
|
|
| 89 |
function modelGroups() {
|
| 90 |
return [
|
| 91 |
{key:'baselines', model:'Native instructions · paired 50k evaluation', note:'Same scenes and instructions · 50 tasks per condition', variants:data.variants.filter(v=>v.paired_campaign)},
|
| 92 |
+
{key:'native-pose', model:'Kinex + pose models · native instructions', note:'GPT-6 Astra / high · agent-chosen TCP targets · progress off · 50 tasks per checkpoint', variants:data.variants.filter(v=>v.native_pose_campaign)},
|
| 93 |
{key:'historical-native', model:'Restored native baseline', note:'Original 16/50 result · historical seeds retained', variants:data.variants.filter(v=>v.historical_reference)},
|
| 94 |
+
{key:'historical-rle', model:'Historical RLE agent results', note:'Different instruction and execution protocol', variants:data.variants.filter(v=>!v.paired_campaign&&!v.historical_reference&&!v.native_pose_campaign)}
|
| 95 |
];
|
| 96 |
}
|
| 97 |
let expandedModels=null;
|
|
|
|
| 526 |
function jsonBlock(title,value,open=false) {return `<details class="tool-block" ${open?'open':''}><summary>${esc(title)}</summary><pre>${esc(JSON.stringify(value,null,2))}</pre></details>`;}
|
| 527 |
function renderPolicy(index) {
|
| 528 |
const record=policyRecord,c=record.calls[index],ep=current.ep;
|
| 529 |
+
if(record.schema==='programvla/model-device-io/1')return renderPosePolicy(index);
|
| 530 |
if(ep.provider==='robotwin-official')return renderOfficialPolicy(index);
|
| 531 |
if(!c){$('#policy-body').innerHTML='<div class="empty-state">No VLA inference was made in this episode.</div>';return;}
|
| 532 |
const images=c.images.map(im=>`<figure><a href="${esc(im.model_image)}" target="_blank"><img src="${esc(im.model_image)}" alt="${esc(im.camera)} model input, call ${index+1}"></a><figcaption>${esc(im.camera)} · ${im.model_size.join(' × ')}<br><a href="${esc(im.recorded)}" target="_blank">Recorded RGB · ${im.recorded_size.join(' × ')} ↗</a></figcaption></figure>`).join('');
|
|
|
|
| 569 |
$('#policy-body').innerHTML=`<label>Inference <select class="policy-select" id="policy-call" aria-label="VLA inference">${r.calls.map((x,i)=>`<option value="${i}" ${i===index?'selected':''}>${i+1} / ${r.calls.length} · action ${x.native_step}</option>`).join('')}</select></label><div class="inline-links"><code>${esc(c.call_id)}</code><span>${c.service_response.inference_s.toFixed(3)} s · ${(c.service_response.waypoints||c.service_response.actions).length} × ${(c.service_response.waypoints||c.service_response.actions)[0].length} actions</span></div><h3>Task instruction</h3><div class="instruction-text">${esc(c.request.task)}</div><div class="policy-images">${ims}</div><p class="subtle">Exact transmitted RGB pixels. The recorded mode contract determines whether measured current and reference target TCP poses accompany the images.</p>${jsonBlock('Complete model request · image arrays linked as PNGs',c.request.wire_payload)}<h3>Native action chunk</h3><p class="subtle">Reordered model output consumed by the official evaluator. Absolute joint radians; gripper opening: 0 closed, 1 open.</p><div class="policy-table"><table><thead><tr><th>Action</th>${r.contract.output_layout.map(x=>`<th>${esc(x.replace('left_','L ').replace('right_','R '))}</th>`).join('')}</tr></thead><tbody>${(c.service_response.waypoints||c.service_response.actions).map((row,i)=>`<tr><th>${i+1}</th>${row.map(x=>`<td title="${x}">${x.toFixed(5)}</td>`).join('')}</tr>`).join('')}</tbody></table></div>${jsonBlock('Complete recorded StarVLA response · full precision',c.raw_response)}${jsonBlock('Reordered native actions · full precision',c.service_response.actions)}<p class="subtle">Standalone execution; no language-agent session or tool call. Token and API costs cover the language agent only; local GPU cost is not reported.</p>${jsonBlock('Recording coverage & model contract',r.contract)}<div class="inline-links">${external(r.contract.model_metadata,'Model configuration')}${external(r.contract.executed_sources,'Source identities')}${external(r.contract.adapter_source||'official/adapters.py','Policy adapter')}${external(r.contract.integration_source||'official/integrations.py','Official evaluator integration')}</div>`;
|
| 570 |
$('#policy-call').onchange=e=>renderPolicy(Number(e.target.value));
|
| 571 |
}
|
| 572 |
+
|
| 573 |
+
function renderPosePolicy(index) {
|
| 574 |
+
const record=policyRecord,c=record.calls[index],ep=current.ep;
|
| 575 |
+
if(!c){$('#policy-body').innerHTML='<div class="empty-state">No VLA inference was made in this episode.</div>';return;}
|
| 576 |
+
const images=c.images.map(im=>`<figure><a href="${esc(im.recorded)}" target="_blank"><img src="${esc(im.recorded)}" alt="${esc(im.camera)} input, call ${index+1}"></a><figcaption>${esc(im.camera)} · ${im.recorded_size.join(' × ')}</figcaption></figure>`).join('');
|
| 577 |
+
const cols=record.contract.output_layout;
|
| 578 |
+
const poses=['left','right'].map(side=>`<tr><th>${side}</th><td>${esc(JSON.stringify(c.request.current_tcp_pose[side]))}</td><td>${esc(JSON.stringify(c.request.target_tcp_pose[side]))}</td></tr>`).join('');
|
| 579 |
+
$('#policy-body').innerHTML=`<label>Inference <select class="policy-select" id="policy-call" aria-label="VLA inference">${record.calls.map((x,i)=>`<option value="${i}" ${i===index?'selected':''}>${i+1} / ${record.calls.length} · ${x.call_id.slice(0,12)}</option>`).join('')}</select></label><div class="inline-links"><code>${esc(c.call_id)}</code><span>${c.response.inference_s.toFixed(3)} s · ${c.actions.length} × ${cols.length} actions · progress off</span></div><h3>Native task instruction</h3><div class="instruction-text">${esc(c.request.instruction)}</div><p class="subtle">${c.instruction_matches_task?'Exactly matches the original native instruction.':'The recorded model request differs from the original native instruction; this outcome is retained.'}</p><div class="policy-images">${images}</div><h3>Measured current and agent-chosen target poses</h3><p class="subtle">world:base · metres · [x, y, z, qx, qy, qz, qw]</p><div class="policy-table"><table><thead><tr><th>Arm</th><th>Current TCP pose</th><th>Target TCP pose</th></tr></thead><tbody>${poses}</tbody></table></div>${jsonBlock('Complete recorded model request',c.request)}<h3>Model output</h3><p class="subtle">Absolute joint radians; gripper opening: 0 closed, 1 open. The device returns joint trajectories and four gripper ticks at 40 ms each. The Kinex session records their execution.</p><div class="policy-table"><table><thead><tr><th>Action</th>${cols.map(x=>`<th>${esc(x.replace('left_','L ').replace('right_','R '))}</th>`).join('')}</tr></thead><tbody>${c.actions.map((row,i)=>`<tr><th>${i+1}</th>${row.map(x=>`<td title="${x}">${x.toFixed(5)}</td>`).join('')}</tr>`).join('')}</tbody></table></div>${jsonBlock('Full model action chunk',c.actions)}${jsonBlock('Returned trajectories and gripper ticks',c.response,true)}<div class="inline-links"><a href="${taskRoute(ep.task_key,ep.variant,'session')}">Kinex session ↗</a>${external(record.contract.model_metadata,'Model configuration')}${external(record.contract.executed_sources,'Executed source hashes')}</div>${jsonBlock('Recording coverage and model contract',record.contract)}`;
|
| 580 |
+
$('#policy-call').onchange=e=>renderPolicy(Number(e.target.value));
|
| 581 |
+
}
|
data-source.json
CHANGED
|
@@ -1,8 +1,8 @@
|
|
| 1 |
{
|
| 2 |
"schema": "programvla/dataset-source/1",
|
| 3 |
"dataset": "RLE-Bench/programmable-vla",
|
| 4 |
-
"revision": "
|
| 5 |
-
"evidence_root": "https://huggingface.co/datasets/RLE-Bench/programmable-vla/resolve/
|
| 6 |
"files": {
|
| 7 |
"data.json": {
|
| 8 |
"url": "./results/data.json"
|
|
@@ -36,6 +36,8 @@
|
|
| 36 |
"jy-mode1",
|
| 37 |
"jy-mode2",
|
| 38 |
"kinex",
|
| 39 |
-
"kinex-vlact50k-autonomous"
|
|
|
|
|
|
|
| 40 |
]
|
| 41 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"schema": "programvla/dataset-source/1",
|
| 3 |
"dataset": "RLE-Bench/programmable-vla",
|
| 4 |
+
"revision": "90ff5cc6e0d326e5073933bd87ecf284700f4698",
|
| 5 |
+
"evidence_root": "https://huggingface.co/datasets/RLE-Bench/programmable-vla/resolve/90ff5cc6e0d326e5073933bd87ecf284700f4698/",
|
| 6 |
"files": {
|
| 7 |
"data.json": {
|
| 8 |
"url": "./results/data.json"
|
|
|
|
| 36 |
"jy-mode1",
|
| 37 |
"jy-mode2",
|
| 38 |
"kinex",
|
| 39 |
+
"kinex-vlact50k-autonomous",
|
| 40 |
+
"kinex-tcp-posestate",
|
| 41 |
+
"kinex-pose-base"
|
| 42 |
]
|
| 43 |
}
|
index.html
CHANGED
|
@@ -11,7 +11,7 @@
|
|
| 11 |
<script src="usage-metrics.js?v=330e92f8eb96" defer></script>
|
| 12 |
<script src="instruction-audit.js?v=ab1836bf6474" defer></script>
|
| 13 |
<script src="model-cards.js?v=7e115b1b03ae" defer></script>
|
| 14 |
-
<script src="app.js?v=
|
| 15 |
<link rel="stylesheet" href="policy.css">
|
| 16 |
<link rel="stylesheet" href="instruction-audit.css?v=43163fc879ed">
|
| 17 |
<link rel="stylesheet" href="comparison-layout.css?v=359bdb6df354">
|
|
|
|
| 11 |
<script src="usage-metrics.js?v=330e92f8eb96" defer></script>
|
| 12 |
<script src="instruction-audit.js?v=ab1836bf6474" defer></script>
|
| 13 |
<script src="model-cards.js?v=7e115b1b03ae" defer></script>
|
| 14 |
+
<script src="app.js?v=0c812377279b" defer></script>
|
| 15 |
<link rel="stylesheet" href="policy.css">
|
| 16 |
<link rel="stylesheet" href="instruction-audit.css?v=43163fc879ed">
|
| 17 |
<link rel="stylesheet" href="comparison-layout.css?v=359bdb6df354">
|
manifest.json
CHANGED
|
@@ -1,145 +1,149 @@
|
|
| 1 |
{
|
| 2 |
"schema": "programvla/space-files/1",
|
| 3 |
"files": {
|
| 4 |
-
"instruction-audit.css": {
|
| 5 |
-
"sha256": "43163fc879edcc6dcc6e3c6bc1faa058d1ac3fd914fcb05302c5ca08640ef7d5",
|
| 6 |
-
"bytes": 1958
|
| 7 |
-
},
|
| 8 |
".gitattributes": {
|
| 9 |
"sha256": "34a4d6e437537e78bc242418805a3af3d71b43980280482fd8fc6bfda91f79f8",
|
| 10 |
"bytes": 56
|
| 11 |
},
|
| 12 |
-
"
|
| 13 |
-
"sha256": "
|
| 14 |
-
"bytes":
|
| 15 |
},
|
| 16 |
-
"
|
| 17 |
-
"sha256": "
|
| 18 |
-
"bytes":
|
| 19 |
},
|
| 20 |
-
"
|
| 21 |
-
"sha256": "
|
| 22 |
-
"bytes":
|
| 23 |
},
|
| 24 |
-
"
|
| 25 |
-
"sha256": "
|
| 26 |
-
"bytes":
|
| 27 |
},
|
| 28 |
-
"
|
| 29 |
-
"sha256": "
|
| 30 |
-
"bytes":
|
| 31 |
},
|
| 32 |
-
"
|
| 33 |
-
"sha256": "
|
| 34 |
-
"bytes":
|
| 35 |
},
|
| 36 |
-
"
|
| 37 |
-
"sha256": "
|
| 38 |
-
"bytes":
|
| 39 |
},
|
| 40 |
-
"
|
| 41 |
-
"sha256": "
|
| 42 |
-
"bytes":
|
| 43 |
},
|
| 44 |
-
"
|
| 45 |
-
"sha256": "
|
| 46 |
-
"bytes":
|
| 47 |
},
|
| 48 |
-
"
|
| 49 |
-
"sha256": "
|
| 50 |
-
"bytes":
|
| 51 |
},
|
| 52 |
"comparison-layout.css": {
|
| 53 |
"sha256": "359bdb6df354c6ff81e167720d5e1898724483e28a84dcfd6d49dd90380f2b7e",
|
| 54 |
"bytes": 4387
|
| 55 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
| 56 |
"icon.svg": {
|
| 57 |
"sha256": "b27d9f4bfb9557513913b43b95108d995c7a18241629631a39889a2989b744c7",
|
| 58 |
"bytes": 725
|
| 59 |
},
|
| 60 |
-
"
|
| 61 |
-
"sha256": "
|
| 62 |
-
"bytes":
|
| 63 |
},
|
| 64 |
-
"
|
| 65 |
-
"sha256": "
|
| 66 |
-
"bytes":
|
| 67 |
},
|
| 68 |
-
"
|
| 69 |
-
"sha256": "
|
| 70 |
-
"bytes":
|
| 71 |
},
|
| 72 |
-
"
|
| 73 |
-
"sha256": "
|
| 74 |
-
"bytes":
|
| 75 |
},
|
| 76 |
-
"
|
| 77 |
-
"sha256": "
|
| 78 |
-
"bytes":
|
| 79 |
},
|
| 80 |
-
"
|
| 81 |
-
"sha256": "
|
| 82 |
-
"bytes":
|
| 83 |
},
|
| 84 |
-
"
|
| 85 |
-
"sha256": "
|
| 86 |
-
"bytes":
|
| 87 |
},
|
| 88 |
-
"
|
| 89 |
-
"sha256": "
|
| 90 |
-
"bytes":
|
| 91 |
},
|
| 92 |
-
"
|
| 93 |
-
"sha256": "
|
| 94 |
-
"bytes":
|
| 95 |
},
|
| 96 |
-
"
|
| 97 |
-
"sha256": "
|
| 98 |
-
"bytes":
|
| 99 |
},
|
| 100 |
-
"
|
| 101 |
-
"sha256": "
|
| 102 |
-
"bytes":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 103 |
},
|
| 104 |
"results/cost-data.json": {
|
| 105 |
"sha256": "244fecb37e750bc84326b3589885be32fad515da82cf1c60dba57e67a798f242",
|
| 106 |
"bytes": 217225
|
| 107 |
},
|
| 108 |
-
"results/
|
| 109 |
-
"sha256": "
|
| 110 |
-
"bytes":
|
| 111 |
-
},
|
| 112 |
-
"results/native-protocol.json": {
|
| 113 |
-
"sha256": "e6646677c010b53371d58d8f52d583dd032b104be4638af912d276422b64a196",
|
| 114 |
-
"bytes": 14647
|
| 115 |
-
},
|
| 116 |
-
"results/native-instructions.json": {
|
| 117 |
-
"sha256": "45540253f590c871a75f18667444d3c52f7e24b5b7fb30b9f38753b83cb4d48a",
|
| 118 |
-
"bytes": 33277
|
| 119 |
},
|
| 120 |
"results/instruction-audit.json": {
|
| 121 |
"sha256": "b26c2303fa9e98eab10293d8d3b4d8543ac40b2ba6aba46bfc306ce191463ec4",
|
| 122 |
"bytes": 282
|
| 123 |
},
|
| 124 |
-
"results/
|
| 125 |
-
"sha256": "
|
| 126 |
-
"bytes":
|
| 127 |
},
|
| 128 |
-
"
|
| 129 |
-
"sha256": "
|
| 130 |
-
"bytes":
|
| 131 |
},
|
| 132 |
-
"
|
| 133 |
-
"sha256": "
|
| 134 |
-
"bytes":
|
| 135 |
},
|
| 136 |
-
"
|
| 137 |
-
"sha256": "
|
| 138 |
-
"bytes":
|
| 139 |
},
|
| 140 |
-
"
|
| 141 |
-
"sha256": "
|
| 142 |
-
"bytes":
|
|
|
|
|
|
|
|
|
|
|
|
|
| 143 |
}
|
| 144 |
}
|
| 145 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"schema": "programvla/space-files/1",
|
| 3 |
"files": {
|
|
|
|
|
|
|
|
|
|
|
|
|
| 4 |
".gitattributes": {
|
| 5 |
"sha256": "34a4d6e437537e78bc242418805a3af3d71b43980280482fd8fc6bfda91f79f8",
|
| 6 |
"bytes": 56
|
| 7 |
},
|
| 8 |
+
"README.md": {
|
| 9 |
+
"sha256": "8ce4346c6543fd9cb0233233f3308c6971cf4f5e71e43c341ad40c5af968963e",
|
| 10 |
+
"bytes": 2240
|
| 11 |
},
|
| 12 |
+
"agent-versions.json": {
|
| 13 |
+
"sha256": "bb2e44cc64b715cebf56c98a1332754407840f3c0f8012771dcb9dd4457118bf",
|
| 14 |
+
"bytes": 1560
|
| 15 |
},
|
| 16 |
+
"app.js": {
|
| 17 |
+
"sha256": "0c812377279b78e73d39b13fe46ce575de9da0a78c6550334ec5640e690401f1",
|
| 18 |
+
"bytes": 100690
|
| 19 |
},
|
| 20 |
+
"archive/attempt-history-before-native-restoration.json": {
|
| 21 |
+
"sha256": "7cbda46c2994948c07fa3faa55f59f9fbbc9753f2f727ee57fe21c07e630ff60",
|
| 22 |
+
"bytes": 1149449
|
| 23 |
},
|
| 24 |
+
"archive/rle-instruction-panel.json": {
|
| 25 |
+
"sha256": "8f9768b07ca2f40409d8699aaa744ac33d759c71c85422b52815c4a62e2969a7",
|
| 26 |
+
"bytes": 3709044
|
| 27 |
},
|
| 28 |
+
"assets/anthropic-logo.svg": {
|
| 29 |
+
"sha256": "a53108594a23b740fbb08623f6fffdda1d73608b552988a9ffda9f0944ae66a1",
|
| 30 |
+
"bytes": 352
|
| 31 |
},
|
| 32 |
+
"assets/kinex-logo-night.svg": {
|
| 33 |
+
"sha256": "7ed5de35505694eb892fc3b601d92e1d2fe864b1b0863fc0afff7c873fcfcc7c",
|
| 34 |
+
"bytes": 725
|
| 35 |
},
|
| 36 |
+
"assets/kinex-logo.svg": {
|
| 37 |
+
"sha256": "b27d9f4bfb9557513913b43b95108d995c7a18241629631a39889a2989b744c7",
|
| 38 |
+
"bytes": 725
|
| 39 |
},
|
| 40 |
+
"assets/openai-logo.svg": {
|
| 41 |
+
"sha256": "8e1b976ba47e927ac2928303fd5362fcadfefd68743506510596f107df676e58",
|
| 42 |
+
"bytes": 2571
|
| 43 |
},
|
| 44 |
+
"assets/vla-logo.svg": {
|
| 45 |
+
"sha256": "4731ad16fd0b9b8f7dc02e05de3c1de0dd70223b93056d8345c8175bd9cad287",
|
| 46 |
+
"bytes": 287
|
| 47 |
},
|
| 48 |
"comparison-layout.css": {
|
| 49 |
"sha256": "359bdb6df354c6ff81e167720d5e1898724483e28a84dcfd6d49dd90380f2b7e",
|
| 50 |
"bytes": 4387
|
| 51 |
},
|
| 52 |
+
"data-source.json": {
|
| 53 |
+
"sha256": "50b826448854f0ca6b1ab619543d4f6751c16c2cf8b4cc859c15bcd8d42378f2",
|
| 54 |
+
"bytes": 1165
|
| 55 |
+
},
|
| 56 |
"icon.svg": {
|
| 57 |
"sha256": "b27d9f4bfb9557513913b43b95108d995c7a18241629631a39889a2989b744c7",
|
| 58 |
"bytes": 725
|
| 59 |
},
|
| 60 |
+
"index.html": {
|
| 61 |
+
"sha256": "07c7610e1cb16d68807821005332eddb556f74e1faf28c56ff21f525867c901c",
|
| 62 |
+
"bytes": 3456
|
| 63 |
},
|
| 64 |
+
"instruction-audit/hard.json": {
|
| 65 |
+
"sha256": "065ad767b70ad57a8c0da1bdca8c6660aeb8834de6dc793687081375a6b95e73",
|
| 66 |
+
"bytes": 262598
|
| 67 |
},
|
| 68 |
+
"instruction-audit/sources/task13-catalog.json": {
|
| 69 |
+
"sha256": "340d808fc366701c1f5ce892625c99f4a94a7a137b74e26a9ae28f8ef644ec31",
|
| 70 |
+
"bytes": 62227
|
| 71 |
},
|
| 72 |
+
"instruction-audit/sources/task13-manifest.yaml": {
|
| 73 |
+
"sha256": "2f3a1f347bb855e8046857555f98003fd207316a2ad9abe908a3cc791d7176eb",
|
| 74 |
+
"bytes": 7558
|
| 75 |
},
|
| 76 |
+
"instruction-audit/sources.json": {
|
| 77 |
+
"sha256": "a4948988bc31bc5213a39f4cdd7ccbb1074ffee13a1d9770e811ebcf57e4fe1c",
|
| 78 |
+
"bytes": 9188
|
| 79 |
},
|
| 80 |
+
"instruction-audit.css": {
|
| 81 |
+
"sha256": "43163fc879edcc6dcc6e3c6bc1faa058d1ac3fd914fcb05302c5ca08640ef7d5",
|
| 82 |
+
"bytes": 1958
|
| 83 |
},
|
| 84 |
+
"instruction-audit.js": {
|
| 85 |
+
"sha256": "ab1836bf64741abceda5dfb11aa8ce0bc6efb78959c53d6ea5189958d4dd10ec",
|
| 86 |
+
"bytes": 13004
|
| 87 |
},
|
| 88 |
+
"migration.json": {
|
| 89 |
+
"sha256": "179ad601177499499cef65e7fff05a099334633e81cfdb0ac628fd8c106b8f29",
|
| 90 |
+
"bytes": 942
|
| 91 |
},
|
| 92 |
+
"model-cards.js": {
|
| 93 |
+
"sha256": "7e115b1b03ae5c273e579aa93603199db23060d7b8af764ade90199bf47dfc4a",
|
| 94 |
+
"bytes": 3400
|
| 95 |
},
|
| 96 |
+
"native-pose-publication.json": {
|
| 97 |
+
"sha256": "9da5c7868ea4c1ed0ba0b0133d0b8a087d7bd794a2eb84f99129f682b791e998",
|
| 98 |
+
"bytes": 1426
|
| 99 |
},
|
| 100 |
+
"policy.css": {
|
| 101 |
+
"sha256": "e564132eb242d994b3a0388da20dd9d1cef28d849447cceca674bb4beef32660",
|
| 102 |
+
"bytes": 1067
|
| 103 |
+
},
|
| 104 |
+
"publication.json": {
|
| 105 |
+
"sha256": "52fde578b8e54b9baef504c051de5cfed505849402c5cf331c284b1ff74ce31b",
|
| 106 |
+
"bytes": 417
|
| 107 |
+
},
|
| 108 |
+
"results/attempt-history.json": {
|
| 109 |
+
"sha256": "9593881d06181e57d01743de9c1c29ae81cb495f6cc814f1d50cf920fedbf99f",
|
| 110 |
+
"bytes": 1166469
|
| 111 |
},
|
| 112 |
"results/cost-data.json": {
|
| 113 |
"sha256": "244fecb37e750bc84326b3589885be32fad515da82cf1c60dba57e67a798f242",
|
| 114 |
"bytes": 217225
|
| 115 |
},
|
| 116 |
+
"results/data.json": {
|
| 117 |
+
"sha256": "b4a206499e10c4704fe1f41d03d8fcf7e9aef57ceeb079f21a0640600fac2dc9",
|
| 118 |
+
"bytes": 4087038
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 119 |
},
|
| 120 |
"results/instruction-audit.json": {
|
| 121 |
"sha256": "b26c2303fa9e98eab10293d8d3b4d8543ac40b2ba6aba46bfc306ce191463ec4",
|
| 122 |
"bytes": 282
|
| 123 |
},
|
| 124 |
+
"results/measured-instructions.json": {
|
| 125 |
+
"sha256": "1d12bba4a4533487c9cac2753f90cd0a22d1c12bb388b0ce040d1025bea60e26",
|
| 126 |
+
"bytes": 105583
|
| 127 |
},
|
| 128 |
+
"results/native-instructions.json": {
|
| 129 |
+
"sha256": "45540253f590c871a75f18667444d3c52f7e24b5b7fb30b9f38753b83cb4d48a",
|
| 130 |
+
"bytes": 33277
|
| 131 |
},
|
| 132 |
+
"results/native-protocol.json": {
|
| 133 |
+
"sha256": "e6646677c010b53371d58d8f52d583dd032b104be4638af912d276422b64a196",
|
| 134 |
+
"bytes": 14647
|
| 135 |
},
|
| 136 |
+
"results/sources.json": {
|
| 137 |
+
"sha256": "e668de72fe58d7115ffb6cc2fe5ee8672d4fbbc8dc0e84011c1c4e2d563bfe79",
|
| 138 |
+
"bytes": 2751
|
| 139 |
},
|
| 140 |
+
"style.css": {
|
| 141 |
+
"sha256": "bd1a59147e1d5e6c4fc5abd0871fb7b2e5e282a9ee3d15ca4df8a219509af69c",
|
| 142 |
+
"bytes": 60419
|
| 143 |
+
},
|
| 144 |
+
"usage-metrics.js": {
|
| 145 |
+
"sha256": "330e92f8eb9670cb506f2909a2a47ee8f0bd569199898df3bfa044239ae64161",
|
| 146 |
+
"bytes": 5238
|
| 147 |
}
|
| 148 |
}
|
| 149 |
}
|
native-pose-publication.json
ADDED
|
@@ -0,0 +1,59 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"time": "2026-11-13T21:49:56.455415+00:00",
|
| 3 |
+
"scope": "Two pose-conditioned checkpoints, all 50 Hard tasks, original native unseen instructions",
|
| 4 |
+
"dataset_revision": "90ff5cc6e0d326e5073933bd87ecf284700f4698",
|
| 5 |
+
"previous_space_revision": "86dd9be62dd2ca9960479788beba834ce77ff376",
|
| 6 |
+
"scores": {
|
| 7 |
+
"standalone-vlact-clean": {
|
| 8 |
+
"success": 16,
|
| 9 |
+
"evaluated": 50
|
| 10 |
+
},
|
| 11 |
+
"native-baseline": {
|
| 12 |
+
"success": 0,
|
| 13 |
+
"evaluated": 0
|
| 14 |
+
},
|
| 15 |
+
"pose-base-mode1": {
|
| 16 |
+
"success": 0,
|
| 17 |
+
"evaluated": 0
|
| 18 |
+
},
|
| 19 |
+
"pose-base-mode2": {
|
| 20 |
+
"success": 0,
|
| 21 |
+
"evaluated": 0
|
| 22 |
+
},
|
| 23 |
+
"jy-mode1": {
|
| 24 |
+
"success": 0,
|
| 25 |
+
"evaluated": 0
|
| 26 |
+
},
|
| 27 |
+
"jy-mode2": {
|
| 28 |
+
"success": 0,
|
| 29 |
+
"evaluated": 0
|
| 30 |
+
},
|
| 31 |
+
"kinex": {
|
| 32 |
+
"success": 41,
|
| 33 |
+
"evaluated": 50
|
| 34 |
+
},
|
| 35 |
+
"kinex-vlact50k-autonomous": {
|
| 36 |
+
"success": 31,
|
| 37 |
+
"evaluated": 50
|
| 38 |
+
},
|
| 39 |
+
"kinex-tcp-posestate": {
|
| 40 |
+
"success": 0,
|
| 41 |
+
"evaluated": 0
|
| 42 |
+
},
|
| 43 |
+
"kinex-pose-base": {
|
| 44 |
+
"success": 0,
|
| 45 |
+
"evaluated": 0
|
| 46 |
+
}
|
| 47 |
+
},
|
| 48 |
+
"merge": {
|
| 49 |
+
"retained_existing_episodes": 150,
|
| 50 |
+
"new_native_episodes": 0,
|
| 51 |
+
"all_existing_episode_records_unchanged": true,
|
| 52 |
+
"native_instruction_protocol": true,
|
| 53 |
+
"space_revision": {
|
| 54 |
+
"repo_id": "RLE-Bench/programmableVLA-robotwin2.0-hard",
|
| 55 |
+
"revision": "86dd9be62dd2ca9960479788beba834ce77ff376",
|
| 56 |
+
"time": "2026-11-13T21:45:39.425258+00:00"
|
| 57 |
+
}
|
| 58 |
+
}
|
| 59 |
+
}
|
results/data.json
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|
results/sources.json
CHANGED
|
@@ -40,5 +40,12 @@
|
|
| 40 |
"protocol": "https://huggingface.co/datasets/RLE-Bench/programmable-vla/resolve/bc9bcd12aaee09df3e685f80c02788a49bc2e340/research/robotwin-hard-20261101/protocol.json",
|
| 41 |
"successes": 16,
|
| 42 |
"valid": 50
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 43 |
}
|
| 44 |
}
|
|
|
|
| 40 |
"protocol": "https://huggingface.co/datasets/RLE-Bench/programmable-vla/resolve/bc9bcd12aaee09df3e685f80c02788a49bc2e340/research/robotwin-hard-20261101/protocol.json",
|
| 41 |
"successes": 16,
|
| 42 |
"valid": 50
|
| 43 |
+
},
|
| 44 |
+
"native_pose_evaluation": {
|
| 45 |
+
"protocol": "research/kinex-pose-hard-native-20261113/PROTOCOL.md",
|
| 46 |
+
"source_identities": "research/kinex-pose-hard-native-20261113/source-identities.json",
|
| 47 |
+
"results": "research/kinex-pose-hard-native-20261113/summary.json",
|
| 48 |
+
"validation": "research/kinex-pose-hard-native-20261113/validation.json",
|
| 49 |
+
"report": "research/kinex-pose-hard-native-20261113/REPORT.md"
|
| 50 |
}
|
| 51 |
}
|