-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
781 lines (707 loc) · 22.9 KB
/
Copy pathindex.html
File metadata and controls
781 lines (707 loc) · 22.9 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1" />
<meta name="description" content="A2World project page: Learning Transferable Dynamics Priors from Action to World Modeling." />
<title>A2World | Learning Transferable Dynamics Priors from Action to World Modeling</title>
<style>
:root {
--ink: #171717;
--muted: #5f6368;
--line: #dadce0;
--paper: #ffffff;
--soft: #f5f6f3;
--accent: #2d6a4f;
--accent-2: #9b1c31;
--amber: #a55d00;
--shadow: 0 16px 44px rgba(23, 23, 23, 0.12);
font-family: Inter, ui-sans-serif, system-ui, -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;
color: var(--ink);
background: var(--paper);
}
* {
box-sizing: border-box;
}
body {
margin: 0;
background: var(--paper);
line-height: 1.55;
}
a {
color: inherit;
text-decoration: none;
}
.topbar {
position: sticky;
top: 0;
z-index: 20;
display: flex;
align-items: center;
justify-content: space-between;
gap: 20px;
padding: 14px clamp(18px, 4vw, 56px);
border-bottom: 1px solid rgba(45, 106, 79, 0.14);
background: rgba(255, 255, 255, 0.82);
color: var(--ink);
backdrop-filter: blur(16px);
box-shadow: 0 8px 30px rgba(23, 23, 23, 0.06);
}
.brand {
display: inline-flex;
align-items: center;
gap: 10px;
font-weight: 760;
letter-spacing: 0;
white-space: nowrap;
}
.brand-mark {
width: 26px;
height: 26px;
border: 2px solid rgba(45, 106, 79, 0.72);
border-radius: 6px;
display: inline-grid;
place-items: center;
font-size: 12px;
color: var(--accent);
}
.nav {
display: flex;
align-items: center;
gap: clamp(12px, 2vw, 24px);
color: rgba(23, 23, 23, 0.7);
font-size: 14px;
}
.nav a:hover {
color: var(--accent);
}
.hero {
position: relative;
min-height: 76vh;
display: flex;
align-items: center;
color: var(--ink);
background:
linear-gradient(135deg, #f8fbf7 0%, #edf6f0 38%, #eef4fb 72%, #fff6ea 100%);
overflow: hidden;
}
.hero::before {
content: "";
position: absolute;
inset: 0;
background-image:
linear-gradient(rgba(45, 106, 79, 0.07) 1px, transparent 1px),
linear-gradient(90deg, rgba(45, 106, 79, 0.07) 1px, transparent 1px);
background-size: 34px 34px;
mask-image: linear-gradient(180deg, rgba(0, 0, 0, 0.56), rgba(0, 0, 0, 0.08));
pointer-events: none;
}
.hero-inner {
position: relative;
z-index: 1;
width: min(1180px, calc(100% - 36px));
margin: 0 auto;
padding: 118px 0 76px;
}
.venue {
display: flex;
flex-wrap: wrap;
align-items: center;
gap: 10px;
margin-bottom: 18px;
}
.pill {
display: inline-flex;
align-items: center;
min-height: 30px;
padding: 4px 10px;
border: 1px solid rgba(45, 106, 79, 0.22);
border-radius: 6px;
color: #24543f;
font-size: 13px;
font-weight: 650;
background: rgba(255, 255, 255, 0.56);
}
h1 {
max-width: 980px;
margin: 0;
font-size: clamp(42px, 7vw, 82px);
line-height: 0.98;
letter-spacing: 0;
text-wrap: balance;
}
.subtitle {
max-width: 850px;
margin: 24px 0 0;
color: #424a45;
font-size: clamp(17px, 2.2vw, 23px);
}
.authors {
max-width: 900px;
margin-top: 24px;
color: #5a625d;
font-size: 15px;
}
.actions {
display: flex;
flex-wrap: wrap;
gap: 12px;
margin-top: 28px;
}
.button {
display: inline-flex;
align-items: center;
justify-content: center;
min-height: 42px;
padding: 9px 14px;
border-radius: 7px;
border: 1px solid rgba(45, 106, 79, 0.2);
background: var(--accent);
color: #fff;
font-weight: 720;
font-size: 14px;
box-shadow: 0 12px 26px rgba(45, 106, 79, 0.18);
}
.button.secondary {
background: rgba(255, 255, 255, 0.72);
color: var(--ink);
}
main {
background: var(--paper);
}
section {
padding: 82px clamp(18px, 4vw, 56px);
}
.section-inner {
width: min(1180px, 100%);
margin: 0 auto;
}
.section-kicker {
margin: 0 0 10px;
color: var(--accent);
font-size: 13px;
font-weight: 780;
letter-spacing: 0.08em;
text-transform: uppercase;
}
h2 {
max-width: 900px;
margin: 0;
font-size: clamp(30px, 4vw, 50px);
line-height: 1.08;
letter-spacing: 0;
}
.lead {
max-width: 920px;
margin: 18px 0 0;
color: var(--muted);
font-size: 18px;
}
.abstract {
display: grid;
grid-template-columns: 1.08fr 0.92fr;
gap: 34px;
align-items: start;
}
.statement {
padding-left: 22px;
border-left: 4px solid var(--accent);
font-size: 20px;
line-height: 1.45;
}
.facts {
display: grid;
gap: 12px;
}
.fact {
border: 1px solid var(--line);
border-radius: 8px;
padding: 18px;
background: var(--soft);
}
.fact b {
display: block;
margin-bottom: 4px;
font-size: 15px;
}
.fact span {
color: var(--muted);
font-size: 14px;
}
.band {
background: #f4f0ea;
border-top: 1px solid #e5dfd5;
border-bottom: 1px solid #e5dfd5;
}
.figure-wrap {
margin-top: 32px;
overflow: hidden;
border: 1px solid var(--line);
border-radius: 8px;
background: #fff;
box-shadow: var(--shadow);
}
.figure-wrap img {
display: block;
width: 100%;
height: auto;
}
.overview-figure {
grid-column: 1 / -1;
margin-top: 6px;
}
.pipeline {
display: grid;
grid-template-columns: repeat(3, 1fr);
gap: 16px;
margin-top: 32px;
}
.step {
border: 1px solid var(--line);
border-radius: 8px;
padding: 20px;
background: #fff;
}
.step .num {
color: var(--accent-2);
font-weight: 800;
font-size: 13px;
letter-spacing: 0.08em;
text-transform: uppercase;
}
.step h3,
.video-card h3 {
margin: 8px 0 8px;
font-size: 20px;
line-height: 1.2;
letter-spacing: 0;
}
.step p,
.video-card p {
margin: 0;
color: var(--muted);
font-size: 14px;
}
.video-grid {
display: grid;
grid-template-columns: repeat(5, minmax(0, 1fr));
gap: 16px;
margin-top: 32px;
}
.video-grid.real {
grid-template-columns: repeat(2, minmax(0, 1fr));
}
.video-card {
overflow: hidden;
border: 1px solid var(--line);
border-radius: 8px;
background: #fff;
box-shadow: 0 10px 28px rgba(23, 23, 23, 0.08);
}
.video-frame {
position: relative;
background: #0d0d0d;
}
.video-frame.portrait {
aspect-ratio: 9 / 16;
}
.video-frame.landscape {
aspect-ratio: 16 / 9;
}
video {
display: block;
width: 100%;
height: 100%;
object-fit: contain;
background: #0d0d0d;
}
.video-body {
padding: 15px;
}
.tag-row {
display: flex;
flex-wrap: wrap;
gap: 8px;
margin-top: 14px;
}
.chip {
display: inline-flex;
align-items: center;
min-height: 26px;
padding: 3px 9px;
border-radius: 6px;
background: #eef4ef;
color: #24543f;
font-size: 12px;
font-weight: 700;
}
.results {
display: grid;
grid-template-columns: repeat(3, 1fr);
gap: 16px;
margin-top: 32px;
}
.metric {
border-top: 4px solid var(--accent);
border-radius: 8px;
padding: 22px;
background: #fff;
box-shadow: 0 10px 28px rgba(23, 23, 23, 0.08);
}
.metric strong {
display: block;
font-size: 32px;
line-height: 1;
color: var(--accent-2);
}
.metric span {
display: block;
margin-top: 9px;
color: var(--muted);
font-size: 14px;
}
.bibtex {
margin-top: 26px;
overflow-x: auto;
border-radius: 8px;
background: #151515;
color: #f6f6f6;
}
pre {
margin: 0;
padding: 22px;
font-size: 13px;
line-height: 1.5;
}
footer {
padding: 34px clamp(18px, 4vw, 56px);
border-top: 1px solid var(--line);
color: var(--muted);
font-size: 14px;
background: #f7f7f5;
}
footer .section-inner {
display: flex;
justify-content: space-between;
gap: 20px;
flex-wrap: wrap;
}
@media (max-width: 980px) {
.nav {
display: none;
}
.abstract,
.pipeline,
.results {
grid-template-columns: 1fr;
}
.video-grid,
.video-grid.real {
grid-template-columns: repeat(2, minmax(0, 1fr));
}
}
@media (max-width: 620px) {
.hero {
min-height: 78vh;
}
.hero-inner {
padding-top: 72px;
}
section {
padding: 58px 18px;
}
.video-grid,
.video-grid.real {
grid-template-columns: 1fr;
}
.button {
width: 100%;
}
}
</style>
</head>
<body>
<header class="topbar">
<a class="brand" href="#top" aria-label="A2World home">
<span class="brand-mark">A2</span>
<span>A2World</span>
</a>
<nav class="nav" aria-label="Primary navigation">
<a href="#abstract">Abstract</a>
<a href="#method">Method</a>
<a href="#rollouts">World Rollouts</a>
<a href="#real-robot">Real Robot</a>
<a href="#citation">Citation</a>
</nav>
</header>
<section id="top" class="hero">
<div class="hero-inner">
<div class="venue">
<span class="pill">ECCV 2026</span>
<span class="pill">World Model</span>
<span class="pill">Robot Learning</span>
</div>
<h1>Learning Transferable Dynamics Priors from Action to World Modeling</h1>
<p class="subtitle">
A2World pretrains action-conditioned multi-view world models on large-scale robot manipulation trajectories,
then transfers the learned dynamics prior to simulation rollouts and instruction-conditioned control.
</p>
<p class="authors">
Ze Huang*, Jiahui Zhang*, Hairuo Liu*, Chenxi Zhang, Ran Cheng, Li Zhang<br />
Fudan University · Shanghai Innovation Institute · Shanghai Jiao Tong University · McGill University
</p>
<div class="actions">
<a class="button" href="#citation">ECCV 2026</a>
<a class="button secondary" href="https://github.com/LogosRoboticsGroup/A2World" target="_blank" rel="noopener">Code</a>
<a class="button secondary" href="#rollouts">Watch Rollouts</a>
</div>
</div>
</section>
<main>
<section id="abstract">
<div class="section-inner abstract">
<div>
<p class="section-kicker">Overview</p>
<h2>Action-conditioned world modeling as a reusable dynamics prior.</h2>
<p class="lead">
A2World learns how robot actions drive visual scene evolution. Instead of treating video generation as an
isolated rendering task, it uses action-conditioned future prediction to build a dynamics prior that can be
reused by both simulator-centric and policy-centric robot learning.
</p>
</div>
<div class="facts">
<div class="fact">
<b>Large-scale pretraining</b>
<span>2.1M+ robot manipulation trajectories spanning 20+ robot embodiments.</span>
</div>
<div class="fact">
<b>Two downstream transfers</b>
<span>A2World-sim rolls out long-horizon observations; A2World-policy jointly predicts video and actions.</span>
</div>
<div class="fact">
<b>Real robot validation</b>
<span>Flexiv dual-arm tasks covering insertion, reorientation, switch interaction, lifting, and deformable-object handling.</span>
</div>
</div>
<div class="figure-wrap overview-figure">
<img src="docs/resources/teaser.png" alt="A2World overview: world model pretraining and prior transfer" loading="lazy" />
</div>
</div>
</section>
<section class="band" id="method">
<div class="section-inner">
<p class="section-kicker">Method</p>
<h2>One pretrained dynamics prior, two robot-learning interfaces.</h2>
<p class="lead">
A2World starts from action-to-video world model pretraining, then adapts the same prior into a simulator and
a policy. The shared representation connects visual rollout quality with downstream robot behavior.
</p>
<div class="figure-wrap">
<img src="docs/resources/extensions.png" alt="A2World model extensions for simulator and policy transfer" loading="lazy" />
</div>
<div class="pipeline">
<div class="step">
<span class="num">01 Pretrain</span>
<h3>A2World base model</h3>
<p>Predicts future multi-view manipulation videos from initial observations and future action chunks.</p>
</div>
<div class="step">
<span class="num">02 Simulate</span>
<h3>A2World-sim</h3>
<p>Uses pose-guided history and autoregressive rollout to replace expensive real-robot evaluation.</p>
</div>
<div class="step">
<span class="num">03 Control</span>
<h3>A2World-policy</h3>
<p>Performs joint video-action diffusion for instruction-conditioned real-robot execution.</p>
</div>
</div>
</div>
</section>
<section id="rollouts">
<div class="section-inner">
<p class="section-kicker">World Model Rollouts</p>
<h2>Generated rollouts visualize the learned action-conditioned dynamics.</h2>
<p class="lead">
These videos show A2World world model rollouts on real-robot manipulation scenarios. They are predictions of
future interaction dynamics, not direct camera recordings of policy execution.
</p>
<div class="video-grid">
<article class="video-card">
<div class="video-frame portrait">
<video controls muted playsinline preload="metadata" poster="docs/resources/posters/flip_small_box.jpg">
<source src="docs/resources/world-model-rollouts/flip_small_box.mp4" type="video/mp4" />
</video>
</div>
<div class="video-body">
<h3>Flip small box</h3>
<p>Reorientation under contact-rich manipulation.</p>
<div class="tag-row"><span class="chip">world rollout</span><span class="chip">OOD dynamics</span></div>
</div>
</article>
<article class="video-card">
<div class="video-frame portrait">
<video controls muted playsinline preload="metadata" poster="docs/resources/posters/insert_memory_module.jpg">
<source src="docs/resources/world-model-rollouts/insert_memory_module.mp4" type="video/mp4" />
</video>
</div>
<div class="video-body">
<h3>Insert RAM module</h3>
<p>Precision alignment and insertion dynamics.</p>
<div class="tag-row"><span class="chip">world rollout</span><span class="chip">precision</span></div>
</div>
</article>
<article class="video-card">
<div class="video-frame portrait">
<video controls muted playsinline preload="metadata" poster="docs/resources/posters/lift_box_high.jpg">
<source src="docs/resources/world-model-rollouts/lift_box_high.mp4" type="video/mp4" />
</video>
</div>
<div class="video-body">
<h3>Lift box high</h3>
<p>Longer-horizon object lifting and transport.</p>
<div class="tag-row"><span class="chip">world rollout</span><span class="chip">lifting</span></div>
</div>
</article>
<article class="video-card">
<div class="video-frame portrait">
<video controls muted playsinline preload="metadata" poster="docs/resources/posters/put_chain_in_the_box.jpg">
<source src="docs/resources/world-model-rollouts/put_chain_in_the_box.mp4" type="video/mp4" />
</video>
</div>
<div class="video-body">
<h3>Put chain in the box</h3>
<p>Deformable-object handling with container interaction.</p>
<div class="tag-row"><span class="chip">world rollout</span><span class="chip">deformable</span></div>
</div>
</article>
<article class="video-card">
<div class="video-frame portrait">
<video controls muted playsinline preload="metadata" poster="docs/resources/posters/electric_valve_switch.jpg">
<source src="docs/resources/world-model-rollouts/electric_valve_switch.mp4" type="video/mp4" />
</video>
</div>
<div class="video-body">
<h3>Toggle power switch</h3>
<p>Interactive switch manipulation with small contact changes.</p>
<div class="tag-row"><span class="chip">world rollout</span><span class="chip">switching</span></div>
</div>
</article>
</div>
</div>
</section>
<section class="band" id="real-robot">
<div class="section-inner">
<p class="section-kicker">A2World-policy</p>
<h2>Real-robot executions test whether the learned prior transfers to control.</h2>
<p class="lead">
A2World-policy is evaluated on a Flexiv dual-arm suite. These clips show real executions on the same task
families used to stress precision, contact, object lifting, switch interaction, and deformable-object handling.
</p>
<div class="video-grid real">
<article class="video-card">
<div class="video-frame landscape">
<video controls muted playsinline preload="metadata" poster="docs/resources/posters/put-chain-in-box.jpg">
<source src="docs/resources/real-world-execution/put-chain-in-box.mp4" type="video/mp4" />
</video>
</div>
<div class="video-body">
<h3>Put chain in the box</h3>
<p>Deformable-object handling in a real execution.</p>
</div>
</article>
<article class="video-card">
<div class="video-frame landscape">
<video controls muted playsinline preload="metadata" poster="docs/resources/posters/toggle-power-switch.jpg">
<source src="docs/resources/real-world-execution/toggle-power-switch.mp4" type="video/mp4" />
</video>
</div>
<div class="video-body">
<h3>Toggle power switch</h3>
<p>Small-contact switching with a dual-arm platform.</p>
</div>
</article>
<article class="video-card">
<div class="video-frame landscape">
<video controls muted playsinline preload="metadata" poster="docs/resources/posters/flip-small-box.jpg">
<source src="docs/resources/real-world-execution/flip-small-box.mp4" type="video/mp4" />
</video>
</div>
<div class="video-body">
<h3>Flip small box</h3>
<p>Object reorientation with contact-rich dynamics.</p>
</div>
</article>
<article class="video-card">
<div class="video-frame landscape">
<video controls muted playsinline preload="metadata" poster="docs/resources/posters/insert-ram-module.jpg">
<source src="docs/resources/real-world-execution/insert-ram-module.mp4" type="video/mp4" />
</video>
</div>
<div class="video-body">
<h3>Insert RAM module</h3>
<p>Precision insertion under visual and instruction conditioning.</p>
</div>
</article>
<article class="video-card">
<div class="video-frame landscape">
<video controls muted playsinline preload="metadata" poster="docs/resources/posters/lift-box-high.jpg">
<source src="docs/resources/real-world-execution/lift-box-high.mp4" type="video/mp4" />
</video>
</div>
<div class="video-body">
<h3>Lift box high</h3>
<p>Object lifting with visible progress toward the target state.</p>
</div>
</article>
</div>
</div>
</section>
<section id="results">
<div class="section-inner">
<p class="section-kicker">Highlights</p>
<h2>Experiments connect visual world modeling with robot-learning utility.</h2>
<p class="lead">
The paper evaluates A2World from complementary simulator-centric and policy-centric perspectives, showing that
action-conditioned pretraining produces dynamics priors that transfer beyond visual generation.
</p>
<div class="results">
<div class="metric">
<strong>2.1M+</strong>
<span>Robot manipulation trajectories used for action-conditioned world model pretraining.</span>
</div>
<div class="metric">
<strong>20+</strong>
<span>Robot embodiments covered by the pretraining data mixture.</span>
</div>
<div class="metric">
<strong>2</strong>
<span>Downstream transfers: long-horizon simulation and real-robot policy execution.</span>
</div>
</div>
</div>
</section>
<section class="band" id="citation">
<div class="section-inner">
<p class="section-kicker">Citation</p>
<h2>Reference</h2>
<p class="lead">If this project is useful for your research, please cite the paper, in Proceedings of ECCV 2026.</p>
<div class="bibtex">
<pre><code>@inproceedings{huang2026a2world,
title={Learning Transferable Dynamics Priors from Action to World Modeling},
author={Huang, Ze and Zhang, Jiahui and Liu, Hairuo and Zhang, Chenxi and Cheng, Ran and Zhang, Li},
booktitle={Proceedings of the European Conference on Computer Vision (ECCV)},
year={2026},
}</code></pre>
</div>
</div>
</section>
</main>
<footer>
<div class="section-inner">
<span>A2World · Learning Transferable Dynamics Priors from Action to World Modeling</span>
<span>ECCV 2026</span>
</div>
</footer>
</body>
</html>