-
Notifications
You must be signed in to change notification settings - Fork 6
Expand file tree
/
Copy pathindex.html
More file actions
703 lines (651 loc) · 37.3 KB
/
index.html
File metadata and controls
703 lines (651 loc) · 37.3 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
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>VIMA | General Robot Manipulation with Multimodal Prompts</title>
<script>
var task_map = {
"simple-object-manipulation": "simple_object_manipulation",
"visual-goal-reaching": "visual_goal_reaching",
"novel-concept-grounding": "novel_concept_grounding",
"one-shot-video-imitation": "one_shot_video_imitation",
"visual-constraint-satisfaction": "visual_constraint_satisfaction",
"visual-reasoning": "visual_reasoning"
};
function updateDemoVideo(category) {
// var demo = document.getElementById("single-menu-demos").value;
var task = document.getElementById(category + "-menu-tasks").value;
var inst = document.getElementById(category + "-menu-instances").value;
console.log(task_map[category], task, inst)
var video = document.getElementById(category + "-single-task-video");
video.src = "assets/videos/demos/" +
task_map[category] +
"/" +
task +
"/" +
inst +
".mp4";
video.playbackRate = 2.0;
video.play();
}
</script>
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro"
rel="stylesheet">
<link rel="stylesheet" href="./static/css/bulma.min.css">
<link rel="stylesheet" href="./static/css/bulma-carousel.min.css">
<link rel="stylesheet" href="./static/css/bulma-slider.min.css">
<link rel="stylesheet" href="./static/css/fontawesome.all.min.css">
<link rel="stylesheet" href="./static/css/academicons.min.css">
<link rel="stylesheet"
href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
<link rel="stylesheet" href="./static/css/index.css">
<script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
<script defer src="./static/js/fontawesome.all.min.js"></script>
<script src="./static/js/bulma-carousel.min.js"></script>
<script src="./static/js/bulma-slider.min.js"></script>
<script src="./static/js/index.js"></script>
</head>
<body>
<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">VIMA: General Robot Manipulation with Multimodal
Prompts</h1>
<h3 class="title is-4 conference-authors"><a target="_blank" href="https://icml.cc/">ICML 2023</a>
</h3>
<div class="is-size-5 publication-authors">
<span class="author-block">
<a target="_blank" href="https://yunfanj.com/">Yunfan Jiang</a><sup>1</sup>,
<a target="_blank" href="http://web.stanford.edu/~agrim/">Agrim Gupta</a><sup>1†</sup>,
<a target="_blank"
href="https://zcczhang.github.io/">Zichen "Charles" Zhang</a><sup>2†</sup>,
<a target="_blank" href="https://guanzhi.me/">Guanzhi Wang</a><sup>3 4†</sup>,
<a target="_blank" href="http://group.iiis.tsinghua.edu.cn/~milab/person-douyongqiang.html">Yongqiang Dou</a><sup>5</sup>,
<a target="_blank"
href="https://www.linkedin.com/in/yanjun-anthony-chen/">Yanjun Chen</a><sup>1</sup>,
<br>
<a target="_blank" href="https://profiles.stanford.edu/fei-fei-li">Li Fei-Fei</a><sup>1</sup>,
<a target="_blank" href="http://tensorlab.cms.caltech.edu/users/anima/">Anima Anandkumar</a><sup>3 4</sup>,
<a target="_blank" href="https://www.cs.utexas.edu/~yukez/">Yuke Zhu</a><sup>3 6‡</sup>,
<a target="_blank" href="https://jimfan.me/">Linxi "Jim" Fan</a><sup>3‡</sup>
</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block"><sup>1</sup>Stanford; </span>
<span class="author-block"><sup>2</sup>Macalester College, now at Allen Institute for AI; </span>
<span class="author-block"><sup>3</sup>NVIDIA; </span>
<span class="author-block"><sup>4</sup>Caltech; </span>
<span class="author-block"><sup>5</sup>Tsinghua; </span>
<span class="author-block"><sup>6</sup>UT Austin</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block">Work done during the first author's internship at NVIDIA</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block"><sup>†</sup>Equal Contribution</span>
<span class="author-block"><sup>‡</sup>Equal Advising </span>
</div>
<div class="column has-text-centered">
<div class="publication-links">
<!-- TODO PDF Link. -->
<span class="link-block">
<a target="_blank" href="https://arxiv.org/abs/2210.03094"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="ai ai-arxiv"></i>
</span>
<span>arXiv</span>
</a>
</span>
<span class="link-block">
<a target="_blank" href="assets/vima_paper.pdf"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-file-pdf"></i>
</span>
<span>PDF</span>
</a>
</span>
<!-- Code Link. -->
<span class="link-block">
<a target="_blank" href="https://github.com/vimalabs/VIMA"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>Code</span>
</a>
<a target="_blank" href="https://github.com/vimalabs/VIMA#pretrained-models"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fa fa-network-wired"></i>
</span>
<span>Models</span>
</a>
<a target="_blank" href="https://github.com/vimalabs/VimaBench"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-robot"></i>
</span>
<span>Benchmark</span>
</a>
<a target="_blank" href="https://huggingface.co/datasets/VIMA/VIMA-Data"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-database"></i>
</span>
<span>Dataset</span>
</a>
</span>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<section class="hero is-light is-small">
<div class="hero-body">
<div class="container">
<div id="results-carousel" class="carousel results-carousel">
<div class="item item-sweep_without_exceeding">
<video poster="" id="sweep_without_exceeding" autoplay controls muted loop height="100%"
playbackRate=2.0>
<source src="assets/videos/demos/visual_constraint_satisfaction/sweep_without_exceeding/1.mp4"
type="video/mp4">
</video>
</div>
<div class="item item-sweep_without_touching">
<video poster="" id="sweep_without_touching" autoplay controls muted loop height="100%"
playbackRate=2.0>
<source src="assets/videos/demos/visual_goal_reaching/rearrange_then_restore/1.mp4"
type="video/mp4">
</video>
</div>
<div class="item item-rotate">
<video poster="" id="rotate" autoplay controls muted loop height="100%">
<source src="assets/videos/demos/one_shot_video_imitation/follow_motion/1.mp4"
type="video/mp4">
</video>
</div>
<div class="item item-scene_understanding">
<video poster="" id="scene_understanding" autoplay controls muted loop height="100%">
<source src="assets/videos/demos/novel_concept_grounding/twist/1.mp4"
type="video/mp4">
</video>
</div>
<div class="item item-simple_manipulation">
<video poster="" id="simple_manipulation" autoplay controls muted loop height="100%">
<source src="assets/videos/demos/visual_reasoning/manipulate_old_neighbor/1.mp4"
type="video/mp4">
</video>
</div>
<div class="item item-novel_adj">
<video poster="" id="novel_adj" autoplay controls muted loop height="100%">
<source src="assets/videos/demos/one_shot_video_imitation/follow_order/1.mp4"
type="video/mp4">
</video>
</div>
<div class="item item-novel_adj_and_noun">
<video poster="" id="novel_adj_and_noun" autoplay controls muted loop height="100%">
<source src="assets/videos/demos/simple_object_manipulation/scene_understanding/1.mp4"
type="video/mp4">
</video>
</div>
<div class="item item-novel_noun">
<video poster="" id="novel_noun" autoplay controls muted loop height="100%">
<source src="assets/videos/demos/visual_constraint_satisfaction/sweep_without_touching/1.mp4"
type="video/mp4">
</video>
</div>
<div class="item item-twist">
<video poster="" id="twist" autoplay controls muted loop height="100%">
<source src="assets/videos/demos/novel_concept_grounding/novel_adj/1.mp4"
type="video/mp4">
</video>
</div>
<div class="item item-follow_motion">
<video poster="" id="follow_motion" autoplay controls muted loop height="100%">
<source src="assets/videos/demos/simple_object_manipulation/simple_manipulation/1.mp4"
type="video/mp4">
</video>
</div>
<div class="item item-follow_order">
<video poster="" id="follow_order" autoplay controls muted loop height="100%">
<source src="assets/videos/demos/visual_goal_reaching/rearrange/1.mp4"
type="video/mp4">
</video>
</div>
<div class="item item-rearrange">
<video poster="" id="rearrange" autoplay controls muted loop height="100%">
<source src="assets/videos/demos/simple_object_manipulation/rotate/1.mp4"
type="video/mp4">
</video>
</div>
<div class="item item-manipulate_old_neighbor">
<video poster="" id="manipulate_old_neighbor" autoplay controls muted loop height="100%">
<source src="assets/videos/demos/visual_reasoning/same_color/1.mp4"
type="video/mp4">
</video>
</div>
<div class="item item-pick_in_order_then_restore">
<video poster="" id="pick_in_order_then_restore" autoplay controls muted loop height="100%">
<source src="assets/videos/demos/visual_reasoning/pick_in_order_then_restore/1.mp4"
type="video/mp4">
</video>
</div>
<div class="item item-same_color">
<video poster="" id="same_color" autoplay controls muted loop height="100%">
<source src="assets/videos/demos/novel_concept_grounding/novel_noun/1.mp4"
type="video/mp4">
</video>
</div>
<div class="item item-rearrange_then_restore">
<video poster="" id="rearrange_then_restore" autoplay controls muted loop height="100%">
<source src="assets/videos/demos/visual_reasoning/same_profile/1.mp4"
type="video/mp4">
</video>
</div>
<div class="item item-same_profile">
<video poster="" id="same_profile" autoplay controls muted loop height="100%">
<source src="assets/videos/demos/novel_concept_grounding/novel_adj_and_noun/1.mp4"
type="video/mp4">
</video>
</div>
</div>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<!-- Abstract. -->
<div class="columns is-centered has-text-centered">
<div class="column">
<h2 class="title is-3">Abstract</h2>
<div class="content has-text-justified">
<p style="font-size: 125%">
Prompt-based learning has emerged as a successful paradigm in natural language processing, where
a single general-purpose language model can be instructed to perform any task specified by input
prompts. Yet task specification in robotics comes in various forms, such as imitating one-shot
demonstrations, following language instructions, and reaching visual goals. They are often
considered different tasks and tackled by specialized models. We show that a wide spectrum of
robot manipulation tasks can be expressed with <i>multimodal prompts</i>, interleaving textual
and visual tokens. Accordingly, we develop a new simulation benchmark that consists of thousands
of procedurally-generated tabletop tasks with multimodal prompts, 600K+ expert trajectories for
imitation learning, and a four-level evaluation protocol for systematic generalization. We
design a transformer-based robot agent, VIMA, that processes these prompts and outputs motor
actions autoregressively. VIMA features a recipe that achieves strong model scalability and data
efficiency. It outperforms alternative designs in the hardest zero-shot generalization setting
by up to 2.9x task success rate given the same training data. With 10x less training data, VIMA
still performs 2.7x better than the best competing variant.
</p>
</div>
</div>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-widescreen">
<div class="rows">
<div class="rows is-centered ">
<div class="row is-full-width">
<img src="assets/images/pull.png" class="interpolation-image"
alt="" style="display: block; margin-left: auto; margin-right: auto"/>
<br>
<span style="font-size: 110%"><b>Multimodal prompts for task specification.</b> We observe that many robot manipulation tasks can be expressed as <i>multimodal prompts</i> that interleave language and image/video frames. We propose VIMA, an embodied agent capable of processing mulitimodal prompts (left) and controlling a robot arm to solve the task (right).</span>
</div>
</div>
</div>
</div>
</section>
<!--Model-->
<section class="section">
<div class="container is-max-widescreen">
<div class="rows">
<div class="rows is-centered ">
<div class="row is-full-width">
<h2 class="title is-3"><span class="dvima">VIMA: Visuomotor Attention Agent</span></h2>
<video poster="" autoplay controls muted loop height="100%">
<source src="assets/videos/vima_arch_animation.mp4"
type="video/mp4">
</video>
<br>
<span style="font-size: 110%">
<span style="font-weight: bold">VIMA architecture.</span> We encode the multimodal prompts with a pre-trained T5 model, and condition the robot controller on the prompt through cross-attention layers. The controller is a causal transformer decoder consisting of alternating self and cross attention layers that predicts motor commands conditioned on prompts and interaction history.</span>
</div>
</div>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-widescreen">
<div class="rows">
<div class="rows is-centered ">
<div class="row is-full-width">
<h2 class="title is-3"><span
class="dvima">VIMA-Bench: Benchmark for Multimodal Robot Learning</span></h2>
<span style="font-size: 125%">
We provide 17 representative tasks with multimodal prompt templates, which can be procedurally instantiated into thousands of individual instances by various combinations of textures and tabletop objects.
</span>
<br>
<br>
<br>
<div class="columns">
<!-- Simple Object Manipulation -->
<div class="column has-text-left">
<h3 class="title is-5">Simple Object Manipulation</h3>
<div class="select is-medium">
<select id="simple-object-manipulation-menu-tasks"
onchange="updateDemoVideo('simple-object-manipulation')">
<option value="simple_manipulation" selected="selected">Visual Manipulation</option>
<option value="rotate">Rotate</option>
<option value="scene_understanding">Scene Understanding</option>
</select>
</div>
<div class="select is-medium">
<select id="simple-object-manipulation-menu-instances"
onchange="updateDemoVideo('simple-object-manipulation')">
<option value="1" selected="selected">1</option>
<option value="2">2</option>
<option value="3">3</option>
<option value="4">4</option>
<option value="5">5</option>
</select>
</div>
<video id="simple-object-manipulation-single-task-video"
controls
muted
autoplay
loop
width="100%">
<source src="assets/videos/demos/simple_object_manipulation/simple_manipulation/1.mp4"
type="video/mp4">
</video>
</div>
<!-- Visual Goal Reaching -->
<div class="column has-text-left">
<h3 class="title is-5">Visual Goal Reaching</h3>
<div class="select is-medium">
<select id="visual-goal-reaching-menu-tasks"
onchange="updateDemoVideo('visual-goal-reaching')">
<option value="rearrange" selected="selected">Rearrange</option>
<option value="rearrange_then_restore">Rearrange Then Restore</option>
</select>
</div>
<div class="select is-medium">
<select id="visual-goal-reaching-menu-instances"
onchange="updateDemoVideo('visual-goal-reaching')">
<option value="1" selected="selected">1</option>
<option value="2">2</option>
<option value="3">3</option>
<option value="4">4</option>
<option value="5">5</option>
</select>
</div>
<video id="visual-goal-reaching-single-task-video"
controls
muted
autoplay
loop
width="100%">
<source src="assets/videos/demos/visual_goal_reaching/rearrange/1.mp4"
type="video/mp4">
</video>
</div>
</div>
<br>
<div class="columns">
<!-- Novel Concept Grounding -->
<div class="column has-text-left">
<h3 class="title is-5">Novel Concept Grounding</h3>
<div class="select is-medium">
<select id="novel-concept-grounding-menu-tasks"
onchange="updateDemoVideo('novel-concept-grounding')">
<option value="novel_adj_and_noun" selected="selected">Novel Adjective and Noun
</option>
<option value="novel_adj">Novel Adjective</option>
<option value="novel_noun">Novel Noun</option>
<option value="twist">Twist</option>
</select>
</div>
<div class="select is-medium">
<select id="novel-concept-grounding-menu-instances"
onchange="updateDemoVideo('novel-concept-grounding')">
<option value="1" selected="selected">1</option>
<option value="2">2</option>
<option value="3">3</option>
<option value="4">4</option>
<option value="5">5</option>
</select>
</div>
<video id="novel-concept-grounding-single-task-video"
controls
muted
autoplay
loop
width="100%">
<source src="assets/videos/demos/novel_concept_grounding/novel_adj_and_noun/1.mp4"
type="video/mp4">
</video>
</div>
<!-- One Shot Video Imitation -->
<div class="column has-text-left">
<h3 class="title is-5">One-shot Video Imitation</h3>
<div class="select is-medium">
<select id="one-shot-video-imitation-menu-tasks"
onchange="updateDemoVideo('one-shot-video-imitation')">
<option value="follow_motion" selected="selected">Follow Motion</option>
<option value="follow_order">Follow Order</option>
</select>
</div>
<div class="select is-medium">
<select id="one-shot-video-imitation-menu-instances"
onchange="updateDemoVideo('one-shot-video-imitation')">
<option value="1" selected="selected">1</option>
<option value="2">2</option>
<option value="3">3</option>
<option value="4">4</option>
<option value="5">5</option>
</select>
</div>
<video id="one-shot-video-imitation-single-task-video"
controls
muted
autoplay
loop
width="100%">
<source src="assets/videos/demos/one_shot_video_imitation/follow_motion/1.mp4"
type="video/mp4">
</video>
</div>
</div>
<br>
<div class="columns">
<!-- Visual Constraint Satisfaction -->
<div class="column has-text-left">
<h3 class="title is-5">Visual Constraint Satisfaction</h3>
<div class="select is-medium">
<select id="visual-constraint-satisfaction-menu-tasks"
onchange="updateDemoVideo('visual-constraint-satisfaction')">
<option value="sweep_without_exceeding" selected="selected">Sweep without
Exceeding
</option>
<option value="sweep_without_touching">Sweep without Touching</option>
</select>
</div>
<div class="select is-medium">
<select id="visual-constraint-satisfaction-menu-instances"
onchange="updateDemoVideo('visual-constraint-satisfaction')">
<option value="1" selected="selected">1</option>
<option value="2">2</option>
<option value="3">3</option>
<option value="4">4</option>
<option value="5">5</option>
</select>
</div>
<video id="visual-constraint-satisfaction-single-task-video"
controls
muted
autoplay
loop
playbackRate="2.0"
width="100%">
<source src="assets/videos/demos/visual_constraint_satisfaction/sweep_without_exceeding/1.mp4"
type="video/mp4">
</video>
</div>
<!-- Visual Reasoning -->
<div class="column has-text-left">
<h3 class="title is-5">Visual Reasoning</h3>
<div class="select is-medium">
<select id="visual-reasoning-menu-tasks" onchange="updateDemoVideo('visual-reasoning')">
<option value="manipulate_old_neighbor" selected="selected">Manipulate Old
Neighbor
</option>
<option value="pick_in_order_then_restore">Pick in Order Then Restore</option>
<option value="same_color">Same Texture</option>
<option value="same_profile">Same Shape</option>
</select>
</div>
<div class="select is-medium">
<select id="visual-reasoning-menu-instances"
onchange="updateDemoVideo('visual-reasoning')">
<option value="1" selected="selected">1</option>
<option value="2">2</option>
<option value="3">3</option>
<option value="4">4</option>
<option value="5">5</option>
</select>
</div>
<video id="visual-reasoning-single-task-video"
controls
muted
autoplay
loop
width="100%">
<source src="assets/videos/demos/visual_reasoning/manipulate_old_neighbor/1.mp4"
type="video/mp4">
</video>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<!--Experiments-->
<section class="section">
<div class="container is-max-widescreen">
<div class="rows">
<div class="rows is-centered ">
<div class="row is-full-width">
<h2 class="title is-3"><span
class="dvima">Experiments</span></h2>
<p style="font-size: 125%">
We answer three main questions during experiments:
<ul style="font-size: 125%; padding-left: 5%">
<li>
1. What is the best recipe for building multi-task transformer-based robot agents with
multimodal prompts?
</li>
<li>
2. What are the <span style="font-weight: bold">scaling properties</span> of our approach in
model capacity and data size?
</li>
<li>
3. How do different components, such as visual tokenizers, prompt conditioning, and prompt
encoding, affect robot performance?
</li>
</ul>
</p>
<br>
<br>
<h3 class="title is-4"><span
class="dvima">Evaluation Results</span></h3>
<img src="assets/images/scalability.png" class="interpolation-image"
alt="" style="display: block; margin-left: auto; margin-right: auto"/>
<br>
<span style="font-size: 110%">
<span style="font-weight: bold">Scaling model and data.</span> <i>Top</i>: We compare performance of different methods with model sizes ranging from 2M to 200M parameters. Across all model sizes and generalization levels, VIMA outperforms baseline variants. <i>Bottom</i>: For a fixed model size of 92M parameters we compare the effect of imitation learning dataset size with 0.1%, 1%, 10%, and full data. VIMA is extremely sample efficient and can achieve performance comparable to other methods with 10x less data.
</span>
<br>
<br>
<br>
<h3 class="title is-4"><span
class="dvima">Ablation Studies </span></h3>
<br>
<img src="assets/images/ablation_input_process.png" class="interpolation-image"
alt="" style="display: block; margin-left: auto; margin-right: auto"/>
<br>
<span style="font-size: 110%">
<span style="font-weight: bold">Ablation on visual tokenizers.</span> We compare the performance of VIMA-200M model across different visual tokenizers. Our proposed object tokens outperform all methods that learn directly from raw pixels, and <i>Object Perceiver</i> that downsamples the object sequence to a fixed number of tokens.
</span>
<br>
<br>
<br>
<br>
<img src="assets/images/global_seq_mod.png" class="interpolation-image"
alt="" style="display: block; margin-left: auto; margin-right: auto"/>
<br>
<span style="font-size: 110%">
<span style="font-weight: bold">Ablation on prompt conditioning.</span> We compare our method (<i>xattn</i>: cross-attention prompt conditioning) with a vanilla transformer decoder (<i>gpt-decoder</i>) across different model sizes. Cross-attention is especially helpful in low-parameter regime and for harder generalization tasks.
</span>
</div>
</div>
</div>
</div>
</section>
<!--Conclusion-->
<section class="section">
<div class="container is-max-widescreen">
<div class="rows">
<div class="rows is-centered ">
<div class="row is-full-width">
<h2 class="title is-3"><span
class="dvima">Conclusion</span></h2>
<p style="font-size: 125%">
In this work, we introduce a novel <i>multimodal</i> prompting formulation that converts diverse
robot manipulation tasks into a uniform sequence modeling problem. We instantiate this
formulation in VIMA-Bench, a diverse benchmark with multimodal tasks and systematic evaluation
protocols for generalization. We propose VIMA, a conceptually simple transformer-based agent
capable of solving tasks such as visual goal reaching, one-shot video imitation, and novel
concept grounding with a single model. Through comprehensive experiments, we show that VIMA
exhibits strong model scalability and zero-shot generalization. Therefore, we recommend our
agent design as a solid starting point for future work.
</p>
</div>
</div>
</div>
</div>
</section>
<section class="section" id="BibTeX">
<div class="container is-max-widescreen content">
<h2 class="title">BibTeX</h2>
<pre><code>@inproceedings{jiang2023vima,
title = {VIMA: General Robot Manipulation with Multimodal Prompts},
author = {Yunfan Jiang and Agrim Gupta and Zichen Zhang and Guanzhi Wang and Yongqiang Dou and Yanjun Chen and Li Fei-Fei and Anima Anandkumar and Yuke Zhu and Linxi Fan},
booktitle = {Fortieth International Conference on Machine Learning},
year = {2023}
}</code></pre>
</div>
</section>
<footer class="footer">
<div class="container">
<div class="columns is-centered">
<div class="column">
<div class="content has-text-centered">
<p>
Website template borrowed from <a
href="https://github.com/nerfies/nerfies.github.io">NeRFies</a> and <a
href="https://github.com/cliport/cliport.github.io">CLIPort</a>.
</p>
</div>
</div>
</div>
</div>
</footer>
</body>
</html>