-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
5379 lines (5198 loc) · 358 KB
/
index.html
File metadata and controls
5379 lines (5198 loc) · 358 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
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>RoboMME</title>
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro" rel="stylesheet">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/bulma/0.9.4/css/bulma.min.css">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.4.0/css/all.min.css">
<style>
body {
font-family: 'Noto Sans', sans-serif;
}
.hero-body {
padding: 2rem 1.5rem;
}
.publication-title {
font-family: 'Google Sans', sans-serif;
font-size: 2.5rem;
font-weight: bold;
}
.publication-title-line {
display: block;
white-space: nowrap;
}
.publication-spotlight {
margin-top: 0.5rem;
color: #d71920;
font-size: 1.5rem;
font-weight: 700;
}
.announcement-banner {
display: inline-flex;
align-items: center;
gap: 0.5rem;
margin: 1.25rem auto 0;
padding: 0.4rem 2.25rem;
border-radius: 999px;
border: 1.5px solid #f39c12;
background-color: #fff7f0;
font-size: 1.2rem;
color: #b95a00;
white-space: nowrap;
}
.announcement-banner a {
color: #b95a00;
text-decoration: underline;
text-decoration-thickness: 1px;
}
.announcement-badge {
font-weight: 600;
}
.publication-authors {
font-size: 1.2rem;
margin-top: 1rem;
}
.publication-venue {
font-size: 1.1rem;
color: #555;
margin-top: 0.5rem;
}
.task-suite {
margin-bottom: 2.5rem;
}
/* Widen hero container so title fits on two lines */
.hero .container.is-max-desktop {
max-width: 1200px;
}
.task-suite-title {
font-size: 1.5rem;
font-weight: bold;
color: #363636;
margin-bottom: 1rem;
padding-bottom: 0.4rem;
border-bottom: 2px solid #3273dc;
}
.task-grid {
display: grid;
grid-template-columns: repeat(2, 1fr);
gap: 1rem;
}
.task-card {
background: white;
border-radius: 8px;
padding: 0.8rem;
box-shadow: 0 2px 8px rgba(0,0,0,0.1);
transition: transform 0.2s, box-shadow 0.2s;
display: flex;
flex-direction: column;
}
.task-card:hover {
transform: translateY(-2px);
box-shadow: 0 4px 12px rgba(0,0,0,0.15);
}
.task-name {
font-size: 1.2rem;
font-weight: bold;
color: #3273dc;
margin-bottom: 0.4rem;
}
.task-instructions {
color: #4a4a4a;
line-height: 1.5;
font-size: 1.1rem;
margin-bottom: 0.8rem;
}
.task-card-content {
display: flex;
flex-direction: column;
height: 100%;
}
.video-wrapper {
flex-grow: 1;
display: flex;
align-items: flex-end;
justify-content: center;
}
@media screen and (max-width: 768px) {
.task-grid {
grid-template-columns: 1fr;
}
}
.video-container {
border-radius: 4px;
overflow: hidden;
width: 100%;
padding-bottom: 1.5rem; /* Reserve space for caption */
position: relative;
}
.publication-links.one-line {
display: flex;
flex-wrap: nowrap;
justify-content: center;
align-items: center;
gap: 0.4rem;
overflow-x: auto;
white-space: nowrap;
padding-bottom: 0.25rem;
}
.publication-links.one-line .link-block {
flex: 0 0 auto;
}
.publication-links.one-line .button {
margin: 0;
}
.video-container video {
display: block;
max-width: 100%;
height: auto;
border-radius: 4px;
}
.video-container video::-webkit-media-controls-panel {
background: linear-gradient(transparent, rgba(0,0,0,0.4));
}
.video-caption {
font-size: 0.85rem;
color: #f7220a;
font-style: italic;
position: absolute;
bottom: 0;
left: 0;
right: 0;
}
.video-caption.caption-black {
color: #333;
}
.container.is-widescreen {
max-width: 1344px;
}
.button.is-link {
background-color: #3273dc;
}
.section {
padding: 2rem 1.5rem;
}
/* Tab styles */
.tabs-container {
margin-bottom: 1.5rem;
}
.task-tabs {
display: flex;
justify-content: center;
gap: 0.5rem;
flex-wrap: wrap;
border-bottom: 2px solid #e0e0e0;
padding-bottom: 0;
}
.task-tab {
padding: 0.75rem 1.5rem;
font-size: 1.1rem;
font-weight: 600;
color: #666;
background: transparent;
border: none;
border-bottom: 3px solid transparent;
cursor: pointer;
transition: all 0.2s ease;
margin-bottom: -2px;
}
.task-tab:hover {
color: #3273dc;
background-color: rgba(50, 115, 220, 0.05);
}
.task-tab.active {
color: #3273dc;
border-bottom-color: #3273dc;
background-color: rgba(50, 115, 220, 0.08);
}
.tab-content {
display: none;
}
.tab-content.active {
display: block;
animation: fadeIn 0.3s ease;
}
@keyframes fadeIn {
from { opacity: 0; transform: translateY(10px); }
to { opacity: 1; transform: translateY(0); }
}
@media screen and (max-width: 768px) {
.task-tab {
padding: 0.5rem 1rem;
font-size: 0.95rem;
}
}
/* Collapsible details styling */
details summary::-webkit-details-marker {
display: none;
}
details[open] summary span {
transform: rotate(90deg);
display: inline-block;
}
details summary:hover {
background-color: #f5f5f5;
}
/* Result labels for videos */
.result-label {
text-align: center;
font-weight: bold;
font-size: 0.95rem;
padding: 0.25rem 0.75rem;
border-radius: 4px;
margin-top: 0.5rem;
display: inline-block;
}
.result-label.success {
color: #23d160;
background-color: rgba(35, 209, 96, 0.1);
border: 1px solid #23d160;
}
.result-label.failure {
color: #ff3860;
background-color: rgba(255, 56, 96, 0.1);
border: 1px solid #ff3860;
}
.result-label-container {
text-align: center;
margin-top: 0.5rem;
}
/* Simulation Rollout Demos: larger task name */
.sim-demo-content .demo-task-header > span:first-child {
font-size: 1.2rem;
}
/* Simulation Rollout Demos: whole tab section 70% width, centered */
.sim-rollout-container {
max-width: 70%;
margin-left: auto;
margin-right: auto;
}
/* Simulation Rollout Demos: tab bar same width as task rows */
.sim-rollout-container .tabs-container {
max-width: 70%;
margin-left: auto;
margin-right: auto;
}
/* Simulation demo: task block same width as tab bar */
.sim-demo-content .sim-task-row {
max-width: 70%;
margin-left: auto;
margin-right: auto;
}
/* Center column content (selector + video) within each column */
.sim-demo-content .demo-video-col {
display: flex;
flex-direction: column;
align-items: center;
}
.sim-demo-content .demo-video-col .demo-video,
.sim-demo-content .demo-video-col .sim-model-selector {
width: 100%;
max-width: 100%;
}
/* Real Robot: same-height columns, model names at top, videos bottom-aligned */
.real-video-row {
align-items: stretch;
}
.real-video-row .video-col {
display: flex;
flex-direction: column;
}
.real-video-row .video-col .real-video {
margin-top: auto;
}
/* Simulation demos: larger task goal text */
.sim-task-row .demo-task-header > span:last-child {
font-size: 1.15rem;
}
/* Real Robot: larger task goal text */
.real-task-row .task-goal {
font-size: 1.2rem !important;
}
/* Sticky section nav */
.section-nav {
position: fixed;
top: 0;
left: 0;
right: 0;
z-index: 100;
background: rgba(255, 255, 255, 0.97);
box-shadow: 0 2px 10px rgba(0, 0, 0, 0.1);
transform: translateY(-100%);
transition: transform 0.3s ease;
backdrop-filter: blur(8px);
-webkit-backdrop-filter: blur(8px);
}
.section-nav.visible {
transform: translateY(0);
}
.section-nav-inner {
display: flex;
justify-content: center;
align-items: center;
gap: 0.25rem;
max-width: 1344px;
margin: 0 auto;
padding: 0.55rem 1rem;
}
.section-nav-link {
padding: 0.4rem 1rem;
font-size: 0.88rem;
font-weight: 600;
color: #555;
text-decoration: none;
border-radius: 20px;
transition: all 0.2s ease;
white-space: nowrap;
border: 1.5px solid transparent;
}
.section-nav-link:hover {
color: #3273dc;
background: rgba(50, 115, 220, 0.06);
}
.section-nav-link.active {
color: #3273dc;
background: rgba(50, 115, 220, 0.1);
border-color: rgba(50, 115, 220, 0.25);
}
@media screen and (max-width: 768px) {
.section-nav-inner {
gap: 0.1rem;
padding: 0.45rem 0.5rem;
overflow-x: auto;
}
.section-nav-link {
font-size: 0.78rem;
padding: 0.35rem 0.65rem;
}
}
</style>
</head>
<body>
<!-- Sticky section nav (hidden until scroll) -->
<div class="section-nav" id="sectionNav">
<div class="section-nav-inner">
<a class="section-nav-link" href="#sec-task-examples" data-target="sec-task-examples">RoboMME Task Examples</a>
<a class="section-nav-link" href="#sec-mme-vla" data-target="sec-mme-vla">MME-VLA Suite</a>
<a class="section-nav-link" href="#sec-experiments" data-target="sec-experiments">RoboMME Experiments</a>
<a class="section-nav-link" href="#sec-sim-rollouts" data-target="sec-sim-rollouts">Simulation Rollout Demos</a>
<a class="section-nav-link" href="#sec-real-rollouts" data-target="sec-real-rollouts">Real Robot Rollout Demos</a>
</div>
</div>
<nav class="navbar" role="navigation" aria-label="main navigation">
<div class="navbar-brand">
<a role="button" class="navbar-burger" aria-label="menu" aria-expanded="false" data-target="navMenu">
<span aria-hidden="true"></span>
<span aria-hidden="true"></span>
<span aria-hidden="true"></span>
</a>
</div>
<div id="navMenu" class="navbar-menu">
<div class="navbar-start" style="flex-grow: 1; justify-content: center;">
<a class="navbar-item" href="leaderboard.html">
<span style="margin-left: 0.5rem; font-size: 1.2rem;">🏆 Leaderboard</span>
</a>
<a class="navbar-item" href="challenge.html">
<span style="margin-left: 0.5rem; font-size: 1.2rem;">🚀 CVPR 2026 Challenge</span>
</a>
<div class="navbar-item has-dropdown is-hoverable">
<a class="navbar-link">
<span style="font-size: 1.2rem;">More Research</span>
</a>
<div class="navbar-dropdown">
<a class="navbar-item" href="https://aimbot-reticle.github.io/" target="_blank">
AimBot (CoRL'25)
</a>
<a class="navbar-item" href="https://rich-language-failure-recovery.github.io" target="_blank">
RACER (ICRA'25)
</a>
<a class="navbar-item" href="https://github.com/sled-group/navchat" target="_blank">
NavChat (ICRA'24)
</a>
</div>
</div>
</div>
</div>
</nav>
<!-- Title & Authors -->
<section class="hero">
<div class="hero-body">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column has-text-centered">
<h1 class="title is-1 publication-title">
<span class="publication-title-line">RoboMME: Benchmarking and Understanding</span>
<span class="publication-title-line">Memory for Robotic Generalist Policies</span>
</h1>
<div class="publication-spotlight">ICML 2026 Spotlight</div>
<div class="is-size-5 publication-authors">
<span class="author-block">
<a href="https://yinpeidai.github.io/" target="_blank">Yinpei Dai</a><sup>1</sup>,
</span>
<span class="author-block">
<a href="https://hongzefu.github.io/" target="_blank">Hongze Fu</a><sup>1</sup>,
</span>
<span class="author-block">
<a href="https://jayjunlee.github.io" target="_blank">Jayjun Lee</a><sup>1</sup>,
</span>
<span class="author-block">
<a href="https://sites.google.com/view/yuejiangliu/home" target="_blank">Yuejiang Liu</a><sup>2</sup>,
</span>
<span class="author-block">
<a href="https://haoranzhangumich.github.io/" target="_blank">Haoran Zhang</a><sup>1</sup>,
</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block">
<a href="https://jedyang.com/" target="_blank">Jianing Yang</a><sup>3</sup>,
</span>
<span class="author-block">
<a href="https://ai.stanford.edu/~cbfinn/" target="_blank">Chelsea Finn</a><sup>2</sup>,
</span>
<span class="author-block">
<a href="https://www.mmintlab.com/people/nima-fazeli/" target="_blank">Nima Fazeli</a><sup>1</sup><sup><span style="font-size:50%;">★</span></sup>,
</span>
<span class="author-block">
<a href="https://web.eecs.umich.edu/~chaijy/" target="_blank">Joyce Chai</a><sup>1</sup><sup><span style="font-size:50%;">★</span></sup>
</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block"><sup><span style="font-size:50%;">★</span></sup>Equal Advising</span> <br>
<span class="author-block"><sup>1</sup>University of Michigan,</span>
<span class="author-block"><sup>2</sup>Stanford University,</span>
<span class="author-block"><sup>3</sup>Figure AI</span>
</div>
<div class="announcement-banner">
<span>🔥 <a href="challenge.html" target="_blank">RoboMME Challenge</a> is now open! It will be part of the challenges in the <a href="https://foundation-models-meet-embodied-agents.github.io/cvpr2026/" target="_blank">FMEA Workshop @ CVPR 2026</a> !
</span>
</div>
<div class="column has-text-centered" style="margin-top: 1.5rem;">
<div class="publication-links one-line">
<span class="link-block">
<a href="https://arxiv.org/abs/2603.04639" class="button is-link is-rounded" target="_blank">
<span class="icon"><i class="fas fa-file-pdf"></i></span>
<span>Paper</span>
</a>
</span>
<!-- <span class="link-block">
<a href="leaderboard.html" class="button is-link is-rounded">
<span class="icon"><i class="fas fa-trophy"></i></span>
<span>Leaderboard</span>
</a>
</span> -->
<span class="link-block">
<a href="https://github.com/RoboMME/robomme_benchmark" class="button is-link is-rounded" target="_blank">
<span class="icon"><i class="fab fa-github"></i></span>
<span>Benchmark</span>
</a>
</span>
<span class="link-block">
<a href="https://github.com/RoboMME/robomme_policy_learning" class="button is-link is-rounded" target="_blank">
<span class="icon"><i class="fab fa-github"></i></span>
<span>Policy Learning</span>
</a>
</span>
<span class="link-block">
<a href="https://huggingface.co/datasets/Yinpei/robomme_data_h5" class="button is-link is-rounded" target="_blank">
<span class="icon"><i class="fas fa-database"></i></span>
<span>Dataset</span>
</a>
</span>
<span class="link-block">
<a href="https://huggingface.co/Yinpei/mme_vla_suite" class="button is-link is-rounded" target="_blank">
<span class="icon"><i class="fas fa-database"></i></span>
<span>Models</span>
</a>
</span>
<span class="link-block">
<a href="https://huggingface.co/collections/Yinpei/robomme" class="button is-link is-rounded" target="_blank">
<span class="icon">
<img src="image/huggingface_logo.svg" alt="Hugging Face" style="height: 1em; vertical-align: middle;">
</span>
<span>Collections</span>
</a>
</span>
<span class="link-block">
<a href="https://huggingface.co/spaces/HongzeFu/RoboMME" class="button is-link is-rounded" target="_blank">
<span class="icon">
<img src="image/huggingface_logo.svg" alt="Hugging Face" style="height: 1em; vertical-align: middle;">
</span>
<span>Demo</span>
</a>
</span>
<span class="link-block">
<a href="https://github.com/RoboMME/robomme_benchmark/blob/main/doc/Wechat.jpg" class="button is-link is-rounded" target="_blank">
<span class="icon"><i class="fab fa-weixin"></i></span>
<span>WeChat</span>
</a>
</span>
<span class="link-block">
<a href="https://discord.gg/mGYRqFGz" class="button is-link is-rounded" target="_blank">
<span class="icon"><i class="fab fa-discord"></i></span>
<span>Discord</span>
</a>
</span>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<section class="section" style="padding-top: 0; padding-bottom: 1rem;">
<div class="container is-widescreen">
<p class="has-text-centered" style="font-size: 1.3rem; color: #4a4a4a; line-height: 1.5;">
TL;DR: RoboMME is a large-scale benchmark for memory-augmented robotic manipulation, evaluating how well models remember, reason, and act across temporal, spatial, object, and procedural memory.
</p>
</div>
</section>
<!-- Benchmark Figure -->
<section class="section" style="padding-bottom: 2rem;">
<div class="container is-widescreen">
<figure class="image" style="max-width: 1200px; margin: 0 auto;">
<video src="video/intro.mp4" autoplay muted loop playsinline controls style="width: 100%; display: block; border-radius: 6px;">
Your browser does not support the video tag.
</video>
<figcaption style="margin-top: 1rem; text-align: justify; color: #4a4a4a; font-size: 1.05rem; line-height: 1.35; max-width: 1200px; margin-left: auto; margin-right: auto;">
<b>RoboMME</b> is a large-scale, cognitively motivated robotic benchmark for memory-augmented manipulation, comprising four task suites that target distinct memory types:
(1) The <b>Counting</b> task suite emphasizes <b>temporal memory</b>, requiring robots to accumulate and reason over past events;
(2) The <b>Permanence</b> task suite focuses on <b>spatial memory</b>, requiring tracking of object locations under occlusion and environmental changes;
(3) The <b>Reference</b> task suite evaluates <b>object memory</b>, requiring identification under varied referential cues;
(4) The <b>Imitation</b> task suite assesses <b>procedural memory</b>, measuring the ability to reproduce previously demonstrated behaviors.
<!-- Overall, RoboMME includes 16 tasks with 770k high-quality training timesteps for systematic evaluation of memory-augmented robotic policies -->
</figcaption>
</figure>
</div>
</section>
<!-- <b>RoboMME</b> is a cognitively motivated, large-scale <b>Robo</b>tic benchmark for systematic <b>M</b>emory-augmented <b>M</b>anipulation <b>E</b>valuation. It comprises four task suites, <i>Counting</i>, <i>Permanence</i>, <i>Reference</i>, and <i>Imitation</i>, that target temporal, spatial, object, and procedural memory, for a total of <b>16</b> tasks and <b>773k</b> training timesteps. -->
<!-- Abstract -->
<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Abstract</h2>
<div class="content has-text-justified" style="font-size: 1.1rem;">
<p>
Memory is critical for long-horizon and history-dependent robotic manipulation.
Such tasks often involve counting repeated actions or manipulating objects that become temporarily occluded.
Recent vision-language-action (VLA) models have begun to incorporate memory mechanisms; however, their evaluations remain confined to narrow, non-standardized settings. This limits their systematic understanding, comparison, and progress measurement.
To address these challenges, we introduce <b>RoboMME</b>: a large-scale standardized benchmark for evaluating and advancing VLA models in long-horizon, history-dependent scenarios. Our benchmark comprises <b>16</b> manipulation tasks constructed under a carefully designed taxonomy that evaluates temporal, spatial, object, and procedural memory.
We further develop a suite of <b>14</b> memory-augmented VLA variants built on the π0.5 backbone to systematically explore different memory representations across multiple integration strategies.
Experimental results show that the effectiveness of memory representations is highly task-dependent, with each design offering distinct advantages and limitations across different tasks.
</p>
</div>
</div>
</div>
</div>
</section>
<!-- Simulation Tasks -->
<section id="sec-task-examples" class="section" style="background-color: #f5f5f5;">
<div class="container is-widescreen">
<h2 class="title is-3 has-text-centered" style="margin-bottom: 1.5rem;">RoboMME Task Examples</h2>
<!-- Tab Navigation -->
<div class="tabs-container">
<div class="task-tabs">
<button class="task-tab active" data-tab="counting">Task Suite 1: Counting</button>
<button class="task-tab" data-tab="permanence">Task Suite 2: Permanence</button>
<button class="task-tab" data-tab="reference">Task Suite 3: Reference</button>
<button class="task-tab" data-tab="imitation">Task Suite 4: Imitation</button>
</div>
</div>
<!-- Tab Content: Counting -->
<div class="tab-content active" id="counting">
<div class="task-suite">
<div class="task-grid">
<!-- Task 1.1 -->
<div class="task-card">
<div class="task-name">Task 1.1: BinFill</div>
<div class="task-instructions">
<strong>Instruction:</strong> Put two green cubes into the bin and press the button to stop.
</div>
<div class="video-wrapper">
<div class="video-container">
<video autoplay muted loop playsinline controls>
<source src="video/BinFill.mp4" type="video/mp4">
</video>
</div>
</div>
</div>
<!-- Task 1.2 -->
<div class="task-card">
<div class="task-name">Task 1.2: PickXtimes</div>
<div class="task-instructions">
<strong>Instruction:</strong> Pick up the blue cube and place it on the target, repeating this pick-and-place action three times, then press the button to stop.
</div>
<div class="video-wrapper">
<div class="video-container">
<video autoplay muted loop playsinline controls>
<source src="video/PickXtimes.mp4" type="video/mp4">
</video>
</div>
</div>
</div>
<!-- Task 1.3 -->
<div class="task-card">
<div class="task-name">Task 1.3: SwingXtimes</div>
<div class="task-instructions">
<strong>Instruction:</strong> Pick up the red cube, move it to the right-side target and then to the left-side target, repeating this right-to-left swing motion three times, then put down the cube and press the button to stop.
</div>
<div class="video-wrapper">
<div class="video-container">
<video autoplay muted loop playsinline controls>
<source src="video/SwingXtimes.mp4" type="video/mp4">
</video>
<div class="video-caption caption-black">Spatial directions (e.g., left/right) follow the robot base coordinate frame.</div>
</div>
</div>
</div>
<!-- Task 1.4 -->
<div class="task-card">
<div class="task-name">Task 1.4: StopCube</div>
<div class="task-instructions">
<strong>Instruction:</strong> Press the button to stop the cube exactly at the target on its third visit.
</div>
<div class="video-wrapper">
<div class="video-container">
<video autoplay muted loop playsinline controls>
<source src="video/StopCube.mp4" type="video/mp4">
</video>
</div>
</div>
</div>
</div>
</div>
</div>
<!-- Tab Content: Permanence -->
<div class="tab-content" id="permanence">
<div class="task-suite">
<div class="task-grid">
<div class="task-card">
<div class="task-name">Task 2.1: VideoUnmask</div>
<div class="task-instructions">
<strong>Instruction:</strong> Watch the video carefully, then pick up the container hiding the green cube.
</div>
<div class="video-wrapper">
<div class="video-container">
<video autoplay muted loop playsinline controls>
<source src="video/VideoUnmask.mp4" type="video/mp4">
</video>
<div class="video-caption">Red-bordered frames denote the video-based initial observation prior to execution.</div>
</div>
</div>
</div>
<div class="task-card">
<div class="task-name">Task 2.2: ButtonUnmask</div>
<div class="task-instructions">
<strong>Instruction:</strong> First press the button, then pick up the container hiding the green cube, finally pick up another container hiding the red cube.
</div>
<div class="video-wrapper">
<div class="video-container">
<video autoplay muted loop playsinline controls>
<source src="video/ButtonUnmask.mp4" type="video/mp4">
</video>
</div>
</div>
</div>
<div class="task-card">
<div class="task-name">Task 2.3: VideoUnmaskSwap</div>
<div class="task-instructions">
<strong>Instruction:</strong> Watch the video carefully, then pick up the container hiding the blue cube, finally pick up another container hiding the red cube.
</div>
<div class="video-wrapper">
<div class="video-container">
<video autoplay muted loop playsinline controls>
<source src="video/VideoUnmaskSwap.mp4" type="video/mp4">
</video>
<div class="video-caption">Red-bordered frames denote the video-based initial observation prior to execution.</div>
</div>
</div>
</div>
<div class="task-card">
<div class="task-name">Task 2.4: ButtonUnmaskSwap</div>
<div class="task-instructions">
<strong>Instruction:</strong> First press both buttons on the table, then pick up the container hiding the red cube.
</div>
<div class="video-wrapper">
<div class="video-container">
<video autoplay muted loop playsinline controls>
<source src="video/ButtonUnmaskSwap.mp4" type="video/mp4">
</video>
</div>
</div>
</div>
</div>
</div>
</div>
<!-- Tab Content: Reference -->
<div class="tab-content" id="reference">
<div class="task-suite">
<div class="task-grid">
<div class="task-card">
<div class="task-name">Task 3.1: PickHighlight</div>
<div class="task-instructions">
<strong>Instruction:</strong> First press the button, then pick up all highlighted cubes, finally press the button again to stop.
</div>
<div class="video-wrapper">
<div class="video-container">
<video autoplay muted loop playsinline controls>
<source src="video/PickHighlight.mp4" type="video/mp4">
</video>
</div>
</div>
</div>
<div class="task-card">
<div class="task-name">Task 3.2: VideoRepick</div>
<div class="task-instructions">
<strong>Instruction:</strong> Watch the video carefully, identify the cube that was picked up, then pick up and place down the same cube twice, finally press the button to stop.
</div>
<div class="video-wrapper">
<div class="video-container">
<video autoplay muted loop playsinline controls>
<source src="video/VideoRepick.mp4" type="video/mp4">
</video>
<div class="video-caption">Red-bordered frames denote the video-based initial observation prior to execution.</div>
</div>
</div>
</div>
<div class="task-card">
<div class="task-name">Task 3.3: VideoPlaceButton</div>
<div class="task-instructions">
<strong>Instruction:</strong> Watch the video carefully, then place the blue cube on the target where it was last placed <i>before</i> the button was pressed.
</div>
<div class="video-wrapper">
<div class="video-container">
<video autoplay muted loop playsinline controls>
<source src="video/VideoPlaceButton.mp4" type="video/mp4">
</video>
<div class="video-caption">Red-bordered frames denote the video-based initial observation prior to execution.</div>
</div>
</div>
</div>
<div class="task-card">
<div class="task-name">Task 3.4: VideoPlaceOrder</div>
<div class="task-instructions">
<strong>Instruction:</strong> Watch the video carefully and place the green cube on the <i>third</i> target where it was placed.
</div>
<div class="video-wrapper">
<div class="video-container">
<video autoplay muted loop playsinline controls>
<source src="video/VideoPlaceOrder.mp4" type="video/mp4">
</video>
<div class="video-caption">Red-bordered frames denote the video-based initial observation prior to execution.</div>
</div>
</div>
</div>
</div>
</div>
</div>
<!-- Tab Content: Imitation -->
<div class="tab-content" id="imitation">
<div class="task-suite">
<div class="task-grid">
<div class="task-card">
<div class="task-name">Task 4.1: MoveCube</div>
<div class="task-instructions">
<strong>Instruction:</strong> Watch the video carefully, then move the cube to the target in the same manner shown in the video.
</div>
<div class="video-wrapper">
<div class="video-container">
<video autoplay muted loop playsinline controls>
<source src="video/MoveCube.mp4" type="video/mp4">
</video>
<div class="video-caption">Red-bordered frames denote the video-based initial observation prior to execution.</div>
</div>
</div>
</div>
<div class="task-card">
<div class="task-name">Task 4.2: InsertPeg</div>
<div class="task-instructions">
<strong>Instruction:</strong> Watch the video carefully, then grasp the same peg at the same end and insert it into the same side of the box as in the video.
</div>
<div class="video-wrapper">
<div class="video-container">
<video autoplay muted loop playsinline controls>
<source src="video/InsertPeg.mp4" type="video/mp4">
</video>
<div class="video-caption">Red-bordered frames denote the video-based initial observation prior to execution.</div>
</div>
</div>
</div>
<div class="task-card">
<div class="task-name">Task 4.3: PatternLock</div>
<div class="task-instructions">
<strong>Instruction:</strong> Watch the video carefully, then use the stick attached to the robot to retrace the same pattern shown in the video.
</div>
<div class="video-wrapper">
<div class="video-container">
<video autoplay muted loop playsinline controls>
<source src="video/DrawPattern.mp4" type="video/mp4">
</video>
<div class="video-caption">Red-bordered frames denote the video-based initial observation prior to execution.</div>
</div>
</div>
</div>
<div class="task-card">
<div class="task-name">Task 4.4: RouteStick</div>
<div class="task-instructions">
<strong>Instruction:</strong> Watch the video carefully, then use the stick attached to the robot to navigate around the sticks on the table, following the same path shown in the video.
</div>
<div class="video-wrapper">
<div class="video-container">
<video autoplay muted loop playsinline controls>
<source src="video/RouteStick.mp4" type="video/mp4">
</video>
<div class="video-caption">Red-bordered frames denote the video-based initial observation prior to execution.</div>
</div>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- MME-VLA Architecture -->
<section id="sec-mme-vla" class="section" style="background-color: #fff;">
<div class="container is-widescreen">
<h2 class="title is-3 has-text-centered" style="margin-bottom: 1.5rem;">MME-VLA Suite</h2>
<!-- Intro text -->
<div style="margin-bottom: 1.5rem; max-width: 1200px; margin-left: auto; margin-right: auto;">
<p style="text-align: justify; color: #4a4a4a; font-size: 1.1rem; line-height: 1.6;">
Building on RoboMME, we construct a family of memory-augmented vision-language-action (VLA) models based on the π0.5 backbone, collectively termed the <b>MME-VLA suite</b>. We systematically compare different memory representations and their integration mechanisms.
</p>
</div>
<div class="columns is-centered">
<div class="column is-full">
<figure class="image" style="max-width: 1200px; margin: 0 auto;">
<img src="image/model_design.jpg" alt="MME-VLA Architecture" style="width: 100%; display: block;">
<figcaption style="margin-top: 1rem; text-align: justify; color: #4a4a4a; font-size: 1.05rem; line-height: 1.35; max-width: 1200px; margin-left: auto; margin-right: auto;">
<b>Framework of MME-VLA Suite.</b> The top part illustrates three memory representations, each with two instantiations: (1) <b><i>Symbolic Memory</i></b> summarizes past interactions as high-level abstractions via language-based subgoals, optionally grounded to image pixels; (2) <b><i>Perceptual Memory</i></b> encodes history as raw visual tokens, using either token dropping to remove redundancy or uniform frame sampling to preserve essential context; (3) <b><i>Recurrent Memory</i></b> compresses history into fixed-size latent states through test-time training or recurrent memory transformers. The bottom part shows three end-to-end integration mechanisms for differentiable memory (perceptual and recurrent): (1) <b><i>Memory-as-Context</i></b> directly concatenates memory tokens with observation tokens; (2) <b><i>Memory-as-Modulator</i></b> applies adaptive LayerNorm to modulate the action expert; (3) <b><i>Memory-as-Expert</i></b> adds a separate lightweight memory expert that interacts with other experts through block-wise causal attention.
<!-- Symbolic memory is integrated by simple concatenation with language tokens, without π<sub>0.5</sub> architectural modification (not shown in the figure). -->
</figcaption>
</figure>
</div>
</div>
</div>
</section>
<!-- RoboMME Experiments -->
<section id="sec-experiments" class="section" style="background-color: #f5f5f5;">
<div class="container is-widescreen">
<h2 class="title is-3 has-text-centered" style="margin-bottom: 1.5rem;">RoboMME Experiments</h2>
<!-- MME-VLA Variants Description -->
<div style="color: #4a4a4a; font-size: 1.1rem; line-height: 1.6; margin-bottom: 1.5rem;">
<p style="margin-bottom: 1rem; margin-left: 4.5rem;">We fine-tune a total of <b>14 memory-augmented VLA</b> variants based on π<sub>0.5</sub>:</p>
<!-- <span style="font-variant: small-caps;">GroundSG+Oracle</span>. -->
<ul style="margin-left: 6rem; max-width: 1150px; font-size: 1.05rem; line-height: 1.6; list-style-type: disc;">
<li style="margin-bottom: 0.5rem;"><b>Symbolic Memory Variants:</b> <span style="font-variant: small-caps; color: #060606;">SimpleSG</span> (Simple Subgoal) or <span style="font-variant: small-caps; color: #060606;">GroundSG</span> (Grounded Subgoal), and leverage <span style="font-variant: small-caps; color: #e83636;">Gemini</span> (prompt-based Gemini-2.5-Pro), <span style="font-variant: small-caps; color: #e83636;">QwenVL</span> (fine-tuned Qwen3-VL-4B), or <span style="font-variant: small-caps; color: #e83636;">Oracle</span> (simulator ground-truth) as subgoal predictors → <b>2 VLA variants & 3 subgoal predictors</b></li>
<li style="margin-bottom: 0.5rem;"><b>Perceptual Memory Variants:</b> <span style="font-variant: small-caps; color: #060606;">TokenDrop</span> (Token dropping) or <span style="font-variant: small-caps; color: #060606;">FrameSamp</span> (Frame sampling), combined with <span style="font-variant: small-caps; color: #e83636;">Context</span> (memory-as-context), <span style="font-variant: small-caps; color: #e83636;">Modul</span> (memory-as-modulator), or <span style="font-variant: small-caps; color: #e83636;">Expert</span> (memory-as-expert) integration mechanisms → <b>2 × 3 = 6 VLA variants</b></li>
<li style="margin-bottom: 0.5rem;"><b>Recurrent Memory Variants:</b> <span style="font-variant: small-caps; color: #060606;">TTT</span> (Test-Time Training) or <span style="font-variant: small-caps; color: #060606;">RMT</span> (Recurrent Memory Transformer), combined with <span style="font-variant: small-caps; color: #e83636;">Context</span> (memory-as-context), <span style="font-variant: small-caps; color: #e83636;">Modul</span> (memory-as-modulator), or <span style="font-variant: small-caps; color: #e83636;">Expert</span> (memory-as-expert) integration mechanisms → <b>2 × 3 = 6 VLA variants</b></li>
</ul>
<table style="width: auto; max-width: 900px; margin: 1rem auto; border-collapse: collapse; font-size: 1.05rem;">
<tr style="background: #f5f5f5;">
<th style="padding: 0.75rem; text-align: middle; border: 1px solid #ddd;">Memory Representation</th>
<th style="padding: 0.75rem; text-align: middle; border: 1px solid #ddd;">Method</th>
<th style="padding: 0.75rem; text-align: middle; border: 1px solid #ddd;">Subgoal Predictor</th>
<th style="padding: 0.75rem; text-align: middle; border: 1px solid #ddd;">Integration Mechanism</th>
</tr>
<tr>
<td style="padding: 0.75rem; border: 1px solid #ddd;">Symbolic</td>
<td style="padding: 0.75rem; border: 1px solid #ddd;"><span style="font-variant: small-caps; color: #060606;">SimpleSG</span>, <span style="font-variant: small-caps; color: #060606;">GroundSG</span></td>
<td style="padding: 0.75rem; border: 1px solid #ddd;"><span style="font-variant: small-caps; color: #e83636;">Gemini</span>, <span style="font-variant: small-caps; color: #e83636;">QwenVL</span>, <span style="font-variant: small-caps; color: #e83636;">Oracle</span></td>
<td style="padding: 0.75rem; border: 1px solid #ddd;">--</td>
</tr>
<tr>
<td style="padding: 0.75rem; border: 1px solid #ddd;">Perceptual</td>
<td style="padding: 0.75rem; border: 1px solid #ddd;"><span style="font-variant: small-caps; color: #060606;">TokenDrop</span>, <span style="font-variant: small-caps; color: #060606;">FrameSamp</span></td>
<td style="padding: 0.75rem; border: 1px solid #ddd;">--</td>
<td style="padding: 0.75rem; border: 1px solid #ddd;"><span style="font-variant: small-caps; color: #e83636;">Context</span>, <span style="font-variant: small-caps; color: #e83636;">Modul</span>, <span style="font-variant: small-caps; color: #e83636;">Expert</span></td>
</tr>
<tr>
<td style="padding: 0.75rem; border: 1px solid #ddd;">Recurrent</td>
<td style="padding: 0.75rem; border: 1px solid #ddd;"><span style="font-variant: small-caps; color: #060606;">TTT</span>, <span style="font-variant: small-caps; color: #060606;">RMT</span></td>
<td style="padding: 0.75rem; border: 1px solid #ddd;">--</td>
<td style="padding: 0.75rem; border: 1px solid #ddd;"><span style="font-variant: small-caps; color: #e83636;">Context</span>, <span style="font-variant: small-caps; color: #e83636;">Modul</span>, <span style="font-variant: small-caps; color: #e83636;">Expert</span></td>
</tr>
</table>