Skip to content

Commit d160f87

Browse files
faikitfaikit
committed
.
1 parent a3543fb commit d160f87

File tree

1 file changed

+188
-26
lines changed

1 file changed

+188
-26
lines changed

‎public/SimScale/index.html‎

Lines changed: 188 additions & 26 deletions
Original file line numberDiff line numberDiff line change
@@ -215,11 +215,14 @@ <h1 class="title is-1 publication-title">Learning to Drive via Real-World Simula
215215
<div class="message-body"
216216
style=" background-color: transparent; color: inherit; border: none !important;; ">
217217

218-
<strong>Scaling Up End-to-End Planners by Simulation.</strong> <br>
219-
(a) We construct large-scale simulation data by perturbing ego trajectories, generating corresponding pseudo-expert demonstrations, and rendering multi-view observations in reactive environments.
218+
<h4><strong>TL;DR: Scaling Up End-to-End Planners by Simulation.</strong></h3>
219+
🏗️ A scalable simulation pipeline that synthesizes diverse and high-fidelity reactive driving scenarios with pseudo-expert. <br>
220+
🚀 An effective sim-real co-training strategy that improves robustness and generalization across end-to-end planners. <br>
221+
🔬 A comprehensive recipe that reveals crucial insights into the underlying scaling properties of sim-real learning systems.<br>
222+
<!-- (a) We construct large-scale simulation data by perturbing ego trajectories, generating corresponding pseudo-expert demonstrations, and rendering multi-view observations in reactive environments.
220223
Combined with real-world data, this enables broad coverage of out-of-distribution states and supports sim–real co-training for any end-to-end planner.<br>
221224
(b) Across three representative planner families, including regression, diffusion, and vocabulary scoring,
222-
sim-real co-training consistently produces synergistic improvements in robustness and generalization, demonstrating clear and predictable simulation scaling trends.
225+
sim-real co-training consistently produces synergistic improvements in robustness and generalization, demonstrating clear and predictable simulation scaling trends. -->
223226
</div>
224227
</article>
225228
<!-- </h2> -->
@@ -334,47 +337,144 @@ <h2 class="title is-2">Leaderboard Results</h2>
334337
<div class="columns is-centered has-text-centered">
335338
<div class="column is-four-fifths ">
336339
<div class="content has-text-justified">
340+
337341
<h2 class="subtitle is-3" style="text-align: left;margin-bottom: 15px">&nbspNAVSIMv2 navhard</h2>
338-
<tr>
339-
<td>
342+
343+
<div class="columns is-centered has-text-centered">
344+
<div class="column">
345+
<div class="content has-text-justified ">
346+
347+
<p style="font-size: 18px; margin-top: -10px; margin-bottom: -15px;">
348+
&nbsp*: pseudo-expert supervision; †: reward scoring; S.: per-stage EPDM score.
349+
</p>
350+
</div>
351+
</div>
352+
</div>
353+
340354
<div class="row">
341355
<div class="col">
342356
<div align="center">
343357
<img src="https://ik.imagekit.io/StarBurger/SimScale/navhard.png" alt="Image description" width="100%">
344358
</div>
345359
</div>
346360
</div>
361+
362+
363+
<h2 class="subtitle is-3" style="text-align: left; margin-bottom: 15px">&nbspNAVSIMv2 navtest</h2>
364+
365+
<div class="columns is-centered has-text-centered">
366+
<div class="column">
367+
<div class="content has-text-justified ">
368+
369+
<p style="font-size: 18px; margin-top: -10px; margin-bottom: -15px;">
370+
&nbsp*: pseudo-expert supervision; †: reward scoring.
371+
</p>
372+
</div>
373+
</div>
374+
</div>
375+
376+
<div class="row">
377+
<div class="col">
378+
<div align="center">
379+
<img src="https://ik.imagekit.io/StarBurger/SimScale/navtest.png" alt="Image description" width="100%">
380+
</div>
381+
</div>
382+
</div>
383+
384+
385+
<details style="margin-bottom: 1.5rem;">
386+
<summary class="subtitle is-3" style="text-align: left; margin-bottom: 15px;margin-top: 20px; font-weight:600">🏆 #1 NAVSIMv2 Official Leaderboard</summary>
387+
347388
<div class="columns is-centered has-text-centered">
348389
<div class="column">
349-
<!-- <div class="content has-text-justified interpolation-panel"> -->
350390
<div class="content has-text-justified ">
351391

352392
<p style="font-size: 18px">
353-
*: pseudo-expert supervision; †: reward scoring; S.: per-stage EPDM score.
393+
<a target="_blank" href="https://huggingface.co/spaces/AGC2025/e2e-driving-navhard">NAVSIMv2 Official Leaderboard</a>
354394
</p>
355395
</div>
356396
</div>
357397
</div>
358398

359-
<h2 class="subtitle is-3" style="text-align: left; margin-bottom: 15px;margin-top: -10px">&nbspNAVSIMv2 navtest</h2>
399+
<div class="row">
400+
<div class="col">
401+
<div align="center">
402+
<img src="https://ik.imagekit.io/StarBurger/SimScale/leaderboard.png" style="margin-top: -30px;" alt="Image description" width="100%">
403+
</div>
404+
</div>
405+
</div>
406+
407+
</details>
408+
409+
410+
<details style="margin-bottom: 1.5rem;">
411+
<summary class="subtitle is-3" style="text-align: left; margin-bottom: 15px;margin-top: -10px; font-weight:600">More Results</summary>
360412

413+
<h2 class="subtitle is-4" style="text-align: left; margin-bottom: 0px;margin-top:20px">&nbspThe Effect of Multi-Expert Ensemble</h2>
414+
415+
<div class="columns is-centered has-text-centered">
416+
<div class="column">
417+
<div class="content has-text-justified ">
418+
<p style="font-size: 18px;">
419+
&nbspS1/2: Per-stage score; recovery/planner: recovery-based/planner-based expert; reward: reward scoring only; *: ensemble
420+
</p>
421+
</div>
422+
</div>
423+
</div>
424+
361425
<div class="row">
362426
<div class="col">
363427
<div align="center">
364-
<img src="https://ik.imagekit.io/StarBurger/SimScale/navtest.png" alt="Image description" width="100%">
428+
<img src="https://ik.imagekit.io/StarBurger/SimScale/ensemble.png" style="margin-top: -30px;" alt="Image description" width="50%">
365429
</div>
366430
</div>
367431
</div>
368-
<div class="columns is-centered has-text-centered">
369-
<div class="column">
370-
<div class="content has-text-justified ">
371432

372-
<p style="font-size: 18px">
373-
*: pseudo-expert supervision; †: reward scoring.
374-
</p>
433+
434+
<h2 class="subtitle is-4" style="text-align: left; margin-bottom: 0px;margin-top:20px">&nbspThe Effect of Simulated Reward Scoring</h2>
435+
436+
<div class="columns is-centered has-text-centered">
437+
<div class="column">
438+
<div class="content has-text-justified ">
439+
<p style="font-size: 18px;">
440+
&nbspS1/2: per-stage score.
441+
</p>
442+
</div>
443+
</div>
444+
</div>
445+
446+
<div class="row">
447+
<div class="col">
448+
<div align="center">
449+
<img src="https://ik.imagekit.io/StarBurger/SimScale/scoring.png" style="margin-top: -30px;" alt="Image description" width="50%">
450+
</div>
375451
</div>
376452
</div>
453+
454+
<h2 class="subtitle is-4" style="text-align: left; margin-bottom: 0px;margin-top:20px">&nbspThe Effect between Non-reactive vs. Reactive Data Simualtion</h2>
455+
456+
457+
<div class="columns is-centered has-text-centered">
458+
<div class="column">
459+
<div class="content has-text-justified ">
460+
<p style="font-size: 18px; ">
461+
&nbsp#Round: sampling rounds; #Sim.: simulation data number.
462+
</p>
463+
</div>
464+
</div>
377465
</div>
466+
467+
<div class="row">
468+
<div class="col">
469+
<div align="center">
470+
<img src="https://ik.imagekit.io/StarBurger/SimScale/reaction.png" style="margin-top: -30px;" alt="Image description" width="50%">
471+
</div>
472+
</div>
473+
</div>
474+
475+
</details>
476+
477+
378478
</td>
379479
</tr>
380480
</div>
@@ -406,24 +506,86 @@ <h2 class="title is-2">Scaling Property Analysis</h2>
406506
<div class="content has-text-justified">
407507
<tr>
408508
<td>
509+
<h2 class="subtitle is-3" style="text-align: left;margin-bottom: 0px">&nbspScaling Simulation with Fixed Real-World Corpus</h2>
510+
511+
<div class="columns is-centered has-text-centered">
512+
<div class="column">
513+
<div class="content has-text-justified ">
514+
515+
<p style="font-size: 18px; margin-top: 0px;margin-bottom: -10px;">
516+
&nbspScaling Dynamics with Varying Simulation Data aross End-to-End Planners and Pseudo-Experts.
517+
</p>
518+
</div>
519+
</div>
520+
</div>
521+
409522
<div class="row">
410523
<div class="col">
411524
<div align="center">
412-
<img src="https://ik.imagekit.io/StarBurger/SimScale/curve_scaling.svg" alt="Image description" width="100%">
525+
<img src="https://ik.imagekit.io/StarBurger/SimScale/curve_scaling.svg" style="margin-bottom: 20px" alt="Image description" width="100%">
413526
</div>
414527
</div>
415528
</div>
529+
530+
531+
<div style="display: flex; justify-content: center; gap: 10px; flex-wrap: wrap;">
532+
533+
<div class="content has-text-justified interpolation-panel"
534+
style="display: inline-block; margin: 5px 0;padding: 10px 7px;">
535+
<p style="text-align: center; font-size: 18px; margin: 0;">
536+
<strong>💡 Pseudo-Expert Should Be Exploratory.</strong>
537+
</p>
538+
</div>
539+
540+
<div class="content has-text-justified interpolation-panel"
541+
style="display: inline-block; margin: 5px 0; padding: 10px 7px;">
542+
<p style="text-align: center; font-size: 18px; margin: 0;">
543+
<strong>💡 Multi-Modality Modeling Sparks Scaling.</strong>
544+
</p>
545+
</div>
546+
547+
<div class="content has-text-justified interpolation-panel"
548+
style="display: inline-block; margin: 5px 0; padding: 10px 7px;">
549+
<p style="text-align: center; font-size: 18px; margin: 0;">
550+
<strong>💡 Reward is All You Need.</strong>
551+
</p>
552+
</div>
553+
554+
</div>
555+
556+
557+
558+
<h2 class="subtitle is-3" style="text-align: left; margin-bottom: 0px; margin-top: 40px">&nbspScaling Simulation with Fixed Sim-Real Ratio</h2>
416559
<div class="columns is-centered has-text-centered">
417560
<div class="column">
418561
<div class="content has-text-justified ">
419562

420-
<p style="font-size: 18px; margin-top: 10px;">
421-
<strong>Scaling Dynamics with Different Planners and Pseudo-Expert Trajectories.</strong>
422-
Visualization how simulation data scale and supervision signals influence the driving performance of various planners, where the infection point indicates learning plateau
563+
<p style="font-size: 18px; margin-top: 0px;margin-bottom: -10px;">
564+
&nbspScaling Dynamics with Varying Real-World Data across End-to-End Planners.
423565
</p>
424566
</div>
567+
</div>
425568
</div>
426-
</div>
569+
570+
<div class="row">
571+
<div class="col">
572+
<div align="center">
573+
<img src="https://ik.imagekit.io/StarBurger/SimScale/curve_scaling_real.svg" alt="Image description" width="50%">
574+
</div>
575+
</div>
576+
</div>
577+
578+
<div style="display: flex; justify-content: center; gap: 10px; flex-wrap: wrap;">
579+
580+
<div class="content has-text-justified interpolation-panel"
581+
style="display: inline-block; margin: 5px 0;padding: 10px 7px;">
582+
<p style="text-align: center; font-size: 18px; margin: 0;">
583+
<strong>💡 Sustained Simulation Gains Across Real Data Scales.</strong>
584+
</p>
585+
</div>
586+
587+
</div>
588+
427589
</td>
428590
</tr>
429591
</div>
@@ -455,7 +617,7 @@ <h2 class="title is-2">Qaulitative Results</h2>
455617
<div class="columns is-centered has-text-centered">
456618
<div class="column is-four-fifths ">
457619
<div class="content has-text-justified">
458-
<h2 class="subtitle is-3" style="text-align: left;margin-bottom: 15px">&nbspPseudo-Expert Scene Simulation</h2>
620+
<h2 class="subtitle is-3" style="text-align: left; margin-bottom: 15px">Pseudo-Expert Scene Simulation</h2>
459621
<tr>
460622
<td>
461623
<div class="row">
@@ -490,7 +652,7 @@ <h2 class="subtitle is-3" style="text-align: left;margin-bottom: 15px">&nbspPseu
490652
</td>
491653
</tr>
492654

493-
<h2 class="subtitle is-3" style="text-align: left; margin-bottom: 15px;margin-top: px">&nbspSimulated OOD Scenes</h2>
655+
<h2 class="subtitle is-3" style="text-align: left; margin-bottom: 15px; margin-top: px">&nbspSimulated OOD Scenes</h2>
494656

495657
<div class="row">
496658
<div class="col">
@@ -501,10 +663,10 @@ <h2 class="subtitle is-3" style="text-align: left; margin-bottom: 15px;margin-to
501663
</div>
502664

503665
<details style="margin-bottom: 1.5rem;">
504-
<summary style="font-size:1.5rem; font-weight:700; cursor:pointer;">
666+
<!-- <summary style="font-size:1.5rem; font-weight:700; cursor:pointer;">
505667
More Results
506-
</summary>
507-
668+
</summary> -->
669+
<summary class="subtitle is-3" style="text-align: left; margin-bottom: 15px; margin-top: px; font-weight:600">More Results</summary>
508670
<div style="margin-top:1rem;">
509671
<div class="row">
510672
<div class="col">
@@ -571,7 +733,7 @@ <h2 class="title is-2">Ecosystem</h2>
571733
</div>
572734
</section>
573735

574-
<section class="section" id="BibTeX">
736+
<section class="section" id="BibTeX" style="margin-top: -60px;">
575737
<div class="container is-max-desktop content">
576738
<h2 class="titile">BibTeX</h2>
577739
If you find the project helpful for your research, please consider citing our paper:

0 commit comments

Comments
 (0)