-
Notifications
You must be signed in to change notification settings - Fork 1
Expand file tree
/
Copy pathindex.html
More file actions
1274 lines (1215 loc) · 74.6 KB
/
index.html
File metadata and controls
1274 lines (1215 loc) · 74.6 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="description" content="RoboAgent">
<meta name="image" content="https://robopen.github.io/cover.png">
<meta name="keywords" content="large-scale robot learning robotic dataset">
<meta name="viewport" content="width=device-width, initial-scale=1">
<link rel="icon" type="image/x-icon" href="media/roboagent.png" />
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/bootstrap@4.0.0/dist/css/bootstrap.min.css"
integrity="sha384-Gn5384xqQ1aoWXA+058RXPxPg6fy4IWvTNh0E263XmFcJlSAwiGgFAW/dAiS6JXm" crossorigin="anonymous">
<title>RoboAgent: Towards Sample Efficient Robot Manipulation with Semantic Augmentations and Action Chunking
</title>
<meta property="og:type" content="website" />
<meta property="og:image" content="https://robopen.github.io/imgs/roboagent_webpreview.png" />
<meta property="og:image:type" content="image/jpg">
<meta property="og:url" content="https://robopen.github.io/index.html" />
<meta property="og:title" content="RoboAgent" />
<meta property="og:description"
content="Towards Sample Efficient Robot Manipulation with Semantic Augmentations and Action Chunking" />
<!--TWITTER-->
<meta name="twitter:card" content="summary_large_image" />
<meta name="twitter:title" content="RoboAgent" />
<meta name="twitter:description"
content="Towards Sample Efficient Robot Manipulation with Semantic Augmentations and Action Chunking" />
<link rel="stylesheet" href="index.css">
<link rel='stylesheet' href='https://cdn.jsdelivr.net/npm/bulma@0.9.3/css/bulma.min.css'>
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro" rel="stylesheet">
<link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/4.7.0/css/font-awesome.min.css">
<link href="https://fonts.googleapis.com/css2?family=Raleway" rel="stylesheet">
<link href="https://fonts.googleapis.com/css2?family=Poppins" rel="stylesheet">
<link rel="preconnect" href="https://fonts.googleapis.com">
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
<link href="https://fonts.googleapis.com/css2?family=Poppins:wght@100&display=swap" rel="stylesheet">
<link
href="https://fonts.googleapis.com/css2?family=Poppins:wght@100&family=Source+Sans+3:wght@300;400&display=swap"
rel="stylesheet">
<script src="https://code.jquery.com/jquery-3.2.1.slim.min.js"
integrity="sha384-KJ3o2DKtIkvYIK3UENzmM7KCkRr/rE9/Qpg6aAZGJwFDMVNA/GpGFF93hXpG5KkN"
crossorigin="anonymous"></script>
<script src="https://cdn.jsdelivr.net/npm/popper.js@1.12.9/dist/umd/popper.min.js"
integrity="sha384-ApNbgh9B+Y1QKtv3Rn7W3mgPxhU9K/ScQsAP7hUibX39j7fakFPskvXusvfa0b4Q"
crossorigin="anonymous"></script>
<script src="https://cdn.jsdelivr.net/npm/bootstrap@4.0.0/dist/js/bootstrap.min.js"
integrity="sha384-JZR6Spejh4U02d8jOt6vLEHfe/JQGiRRSQQxSfFWpi1MquVdAyjUar5+76PVCmYl"
crossorigin="anonymous"></script>
<!--ANALYTICS TAG-->
<script async src="https://www.googletagmanager.com/gtag/js?id=UA-44241506-1"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag() { window.dataLayer.push(arguments); }
gtag('js', new Date());
gtag('config', 'UA-44241506-1');
</script>
<link rel="stylesheet" href="./static/css/bulma.min.css">
<link rel="stylesheet" href="./static/css/bulma-carousel.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 src="./static/js/bulma-carousel.min.js"></script>
<script src="./static/js/bulma-carousel.js"></script>
<script src="./static/js/index.js"></script>
</head>
<script>
document.addEventListener('DOMContentLoaded', function () {
const videoElement = document.getElementById('bg-video');
// When the video metadata is loaded
videoElement.addEventListener('loadedmetadata', function () {
videoElement.style.opacity = 1.0;
});
});
window.addEventListener("scroll", function () {
let imageDiv = document.querySelector('.full-page-image');
const videoElement = document.getElementById('bg-video');
if (window.scrollY > 75) {
videoElement.style.opacity = 0;
}
// Check if page is scrolled
if (window.scrollY > 50) { // You can adjust this value based on your requirement
imageDiv.classList.add('banner-state');
imageDiv.classList.add('ncont');
imageDiv.style.backgroundImage = `url(media/small_banner.png)`;
};
});
function copyToClipboard(element) {
var $temp = $("<input>");
$("body").append($temp);
$temp.val($(element).text()).select();
document.execCommand("copy");
$temp.remove();
}
</script>
<body class="vsc-initialized">
<div class="full-page-image">
<video id="bg-video" autoplay loop muted playsinline>
<source src="media/big_banner.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
<!-- You can add an overlay and content inside the container -->
<div class="orange"></div>
<div class="jay" style="padding: 0 20px">
<h1 style="font-family: Poppins; font-size: 250%;">RoboAgent: A universal agent with 12 Skills</h1>
<p style="font-family: Poppins; font-size: 100%;">Universal RoboAgent exhibiting its skills across diverse
tasks in unseen scenarios</p>
</div>
</div>
<!-- <div class="content is-centered has-text-centered" style="margin-right: 20%; margin-left: 20%; "> -->
<div class="content is-centered has-text-centered">
<section class="hero">
<div class="hero-body">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column has-text-centered">
<h2 class="title is-1 publication-title">
<strong>RoboAgent</strong>
</h2>
<h2 class="subtitle" style="font-family: 'Source Sans 3', sans-serif; font-weight: 300;">
<!-- <h2 style="font-family: 'Poppins', sans-serif; font-weight: 100;"> -->
<strong>
Towards Sample Efficient Robot Manipulation with Semantic Augmentations and Action
Chunking
</strong>
</h2>
<h6 class="subtitle" style="font-family: 'Source Sans 3', sans-serif; font-weight: 50;">
<!-- <h2 style="font-family: 'Poppins', sans-serif; font-weight: 100;"> -->
<strong style="color: red;">
Outstanding Presentation Award @ <a href="https://www.robot-learning.ml/2023/"
style="color: red;">6th Robot Learning Workshop</a>, NeurIPS 2023
</strong><br>
<strong style="color: red;">
Accepted at <a href="https://2024.ieee-icra.org/" style="color: red;">2024 IEEE International Conference on
Robotics and Automation</a>
</strong>
</h6>
<div class="is-size-5 publication-authors">
<span class="author-block">
<a href="https://homangab.github.io/" target="_blank">Homanga Bharadhwaj*</a>,
</span>
<span class="author-block">
<a href="https://jdvakil.github.io" target="_blank">Jay Vakil*</a>,
</span>
<span class="author-block">
<a href="https://mohitsharma0690.github.io/" target="_blank">Mohit Sharma*</a>,
</span>
<span class="author-block">
<a href="http://www.cs.cmu.edu/~abhinavg/" target="_blank">Abhinav Gupta</a>,
</span>
<span class="author-block">
<a href="http://shubhtuls.github.io/" target="_blank">Shubham Tulsiani</a>,
</span>
<span class="author-block">
<a href="https://vikashplus.github.io" target="_blank">Vikash Kumar</a>
</span>
</div>
<div class="is-size-8 publication-authors">
<span class="author-block">
*equal contribution, Carnegie Mellon University and Meta AI
</span>
</div>
<div class="column has-text-centered">
<div class="publication-links">
<span class="link-block">
<a href="media/roboagent.pdf"
class="external-link button is-normal is-rounded is-dark" target="_blank">
<span class="icon">
<i class="fa fa-file-pdf-o" style="font-size:24px"></i>
</span>
<span>Paper</span>
</a>
</span>
<span class="link-block">
<a href="https://arxiv.org/pdf/2309.01918.pdf"
class="external-link button is-normal is-rounded is-dark" target="_blank">
<span class="icon">
<i class="ai ai-arxiv" style="font-size:24px"></i>
</span>
<span>arXiv</span>
</a>
</span>
<span class="link-block">
<a href="https://github.com/robopen/roboagent/"
class="external-link button is-normal is-rounded is-dark" target="_blank">
<span class="icon">
<i class="fa fa-github" style="font-size:24px"></i>
</span>
<span>Code</span>
</a>
</span>
<span class="link-block">
<a href="https://robopen.github.io/roboset" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fa fa-database" style="font-size:24px"></i>
</span>
<span>RoboSet</span>
</a>
</span>
<span class="link-block">
<a href="https://robopen.github.io/supplementary.html" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fa fa-video-camera" style="font-size: 24px;"></i>
</span>
<span>Supplementary Videos</span>
</a>
</span>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
</div>
<div class="container is-max-desktop">
<div class="content is-centered has-text-centered">
<div class="column has-text-justified">
<!-- Make central claim -->
<p style="text-align: center;">
RoboAgent can <strong>efficiently</strong> acquire a <strong>wide diversity of non-trivial
skills</strong> and can <strong>generalize them to diverse unseen scenarios</strong>.
</p>
<!-- Add videos highlighting results-->
<div class="column has-text-centered">
<div class="publication-video">
<video poster="" autoplay muted loop playsinline height="100%">
<source src="./media/roboagent_concept.mp4" type="video/mp4">
caption
</video>
</div>
</div>
<!-- Video caption -->
<p>
Trained merely on <strong>7500 trajectories</strong>, we are demonstrating a universal
RoboAgent that can exhibit a diverse set of <strong>12 non-trivial manipulation skills</strong>
(beyond picking/pushing, including articulated object manipulation and object
re-orientation)<strong> across
38 tasks</strong>
and can generalize them to <strong> 100s</strong> of diverse unseen scenarios</strong> (involving
unseen objects, unseen
tasks, and to completely unseen kitchens). RoboAgent can also <strong>evolve its
capabilities</strong> with new experiences.
<!-- <strong>RoboAgent</strong> is capable of 12 unique skills demonstrated across <strong>38 tasks</strong> categorized into <strong>6 kitchen activities</strong>. RoboAgent is capable of a <strong>wide diversity of non-trivial skills</strong> beyond picking/pushing, including articulated object manipulation and object re-orientation. RoboAgent also <strong>generalizes to diverse unseen scenarios</strong> involving unseen objects, unseen tasks, and to completely unseen kitchens. -->
</p>
<h2 class="title is-3">
RoboAgent can exhibit 12 skills across 38 tasks
</h2>
</div>
</div>
</div>
<div class="container is-max-desktop">
<div class="content is-centered has-text-centered">
<div class="column has-text-justified">
<section class="hero is-light is-small">
<div class="hero-body">
<div class="container">
<h5 id="vidtit">
<!-- <b>
Skills 1-6
</b> -->
</h5>
<div id="results-carousel" class="carousel results-carousel">
<div class="card" id="slide-open">
<div class="card-image">
<video poster="" autoplay muted loop playsinline height="100%">
<source src="media/skills/crop_720w/slide_open_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Slide-Open
</div>
</div>
</div>
<div class="card" id="slide-close">
<div class="card-image">
<video poster="" autoplay muted loop playsinline height="100%">
<source src="media/skills/crop_720w/slide_close_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Slide-Close
</div>
</div>
</div>
<div class="card" id="flap-open">
<div class="card-image">
<video poster="" autoplay muted loop playsinline height="100%">
<source src="media/skills/crop_720w/flap_open_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Flap-Open
</div>
</div>
</div>
<div class="card" id="flap-close">
<div class="card-image">
<video poster="" autoplay muted loop playsinline height="100%">
<source src="media/skills/crop_720w/flap_close_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Flap-Close
</div>
</div>
</div>
<div class="card" id="cap">
<div class="card-image">
<video poster="" autoplay muted loop playsinline height="100%">
<source src="media/skills/crop_720w/close_lid_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Cap
</div>
</div>
</div>
<div class="card" id="uncap">
<div class="card-image">
<video poster="" autoplay muted loop playsinline height="100%">
<source src="media/skills/crop_720w/uncap_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Uncap
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<section class="hero is-light is-small">
<div class="hero-body">
<div class="container">
<h5 id="vidtit">
<!-- <b>
Skills 7-12
</b> -->
</h5>
<div id="results-carousel" class="carousel results-carousel">
<div class="card" id="pick">
<div class="card-image">
<video poster="" autoplay muted loop playsinline height="100%">
<source src="media/skills/crop_720w/pick_butter_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Pick
</div>
</div>
</div>
<div class="card" id="place">
<div class="card-image">
<video poster="" autoplay muted loop playsinline height="100%">
<source src="media/skills/crop_720w/place_tea_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Place
</div>
</div>
</div>
<div class="card" id="slide-in">
<div class="card-image">
<video poster="" autoplay muted loop playsinline height="100%">
<source src="media/skills/crop_720w/slide_in_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Slide-in
</div>
</div>
</div>
<div class="card" id="slide-out">
<div class="card-image">
<video poster="" autoplay muted loop playsinline height="100%">
<source src="media/skills/crop_720w/slide_out_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Slide-Out
</div>
</div>
</div>
<div class="card" id="wipe">
<div class="card-image">
<video poster="" autoplay muted loop playsinline height="100%">
<source src="media/skills/crop_720w/wipe_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Wipe
</div>
</div>
</div>
</div>
</div>
</div>
</section>
</div> <!-- Wrap up show and tell -->
</div>
</div>
<!-- </section> -->
<section class="section">
<div class="container is-max-desktop">
<div class="content is-centered has-text-centered">
<!-- Paint the vision / Overview-->
<div class="column has-text-justified">
<h2>Towards a universal robotic agent </h2>
<!-- <h4 style="font-family: 'Source Sans 3', sans-serif; font-weight: 400;"> A causality dilemma</h4> -->
<strong>A causality dilemma</strong>: The grand aim of having a single robot that can
manipulate arbitrary objects in diverse settings has been a distant goal for several decades. This
is in-part because of the paucity of diverse robotics datasets to train such agents, at the same
time absence of generic agents than can generate such dataset.
<br>
<!-- <h4 style="font-family: 'Source Sans 3', sans-serif; font-weight: 400;"> Escaping the vicious circle</h4> -->
<strong>Escaping the vicious circle</strong>: To escape this vicious circle our focus is on
developing an efficient paradigm that can deliver a universal agent capable acquiring multiple
skills under a practical data budget and generalizing them to diverse unseen situations.
<div class="column has-text-centered">
<div class="publication-video">
<iframe src="https://www.youtube.com/embed/_nArEh7fUzk?autoplay=1?controls=0"
title="Overview of MT-ACT capabilities" frameborder="0" allowfullscreen>
</iframe>
</div>
</div>
RoboAgent is a culmination of effort spanning over two years. It builds on the following modular and
recompensable ingredients -
<ul style="text-align: left;">
<li> <a href=https://youtu.be/tUfWqIz5aR4 /><strong>RoboPen</strong></a> - a distributed
robotics infrastructure build with commodity hardware capable of long term uninterrupted
operations.
<li> <a href=https://sites.google.com/view/robohive /><strong>RoboHive</strong></a> - a unified
framework for robot learning across simulation and real-world operations.
<li> <a href=https://sites.google.com/view/robohive/roboset><strong>RoboSet</strong></a> - a
high quality dataset representing multiple skills with everyday objects in diverse
scenarios.
<li> <a href=media/roboagent.pdf><strong>MT-ACT</strong></a> - an efficient language conditioned
multi-task offline imitation learning framework that multiplies offline datasets by creating
a
diverse collection of semantic augmentations over the existing robot’s experiences and
employs a novel policy architecture with efficient action representation to recover
performant policies under a data budget.</li>
</ul>
</p>
</div>
<!-- Dive into Details -->
<div class="column">
<h2 class="title is-3"> RoboSet: Diverse multi-skill multi-task multi-modal dataset</h2>
<p>
Building a robotic agent that can generalize to many different scenarios requires a dataset with
broad coverage. With the recognition that scaling efforts will generally help (e.g. <a
href=https://robotics-transformer1.github.io />RT-1</a> presents results with ~130,000 robot
trajectories), our goal is to understand the principles of <strong> efficiency </strong> and
<strong> generalization</strong> in learning system under a <strong>data budget</strong>. Low
data regimes often results in over-fitting. Our main aim is to thus develop a powerful paradigms
that can learn a generalizable universal policy while avoiding overfitting in this low-data
regime.
</p>
<div class="column overlay-image has-text-centered">
<img src="imgs/data_skill_landscape.png">
Skill vs DataSet landscape in Robot Learning.
</div>
<!-- <h3 style="font-family: 'Source Sans 3', sans-serif; font-weight: 400;"> Dataset </h3> -->
<p>
The dataset <strong>RoboSet(MT-ACT)</strong> used for training RoboAgent consists of merely
7,500 trajectories (18x less data than RT1). The dataset was collected ahead of time, and was
kept frozen. It consists of high quality (mostly successful) trajectories collected using human
teleoperation on commodity robotics hardware (<a href=https://www.franka.de />Franka-Emika
robots</a> with <a href="https://robotiq.com/products/2f85-140-adaptive-robot-gripper">Robotiq
gripper</a>) across multiple tasks and scenes.
RoboSet(MT-ACT) <strong>sparsely</strong> covers 12 unique skills in a few different contexts.
It was collected by dividing everyday kitchen activities (e.g. making tea, baking) into
different sub-tasks, each representing a unique skill. The dataset includes common pick-place
skills but also includes contact-rich skills such as <em>wipe, cap</em> as well as skills
involving articulated objects.
<div class="column overlay-image has-text-centered">
<img src="media/data_collection_setup_01.png">
A snapshot of our robot system and the objects used during data collection.
</div>
</p>
<p>
In addition to the RoboSet(MT-ACT) we use for training RoboAgent, we are also releasing <a
href=robopen.github.io/roboset />RoboSet</a> a much larger dataset collected over the course
of a few related project containing a total of 100,050 trajectories, including non-kitchen
scenes. <strong>We are open-sourcing our entire <a
href=robopen.github.io/roboset />RoboSet</a></strong> to facilitate and accelerate
open-source research in robot-learning.
</p>
<h2 class="title is-3"> MT-ACT: Multi-Task Action Chunking Transformer</h2>
<div class="column has-text-left">
RoboAgent builds on two critical insights to learn generalizable policies in low-data regimes.
It leverages world priors from foundation models to avoid mode collapse and a novel efficient
policy representations capable of ingesting highly multi-modal data.
<ul>
<li>
<strong>Semantic Augmentations:</strong> RoboAgent injects world priors from existing
foundation models by creating semantic augmentations of the RoboSet(MT-ACT). The
resulting dataset multiplies robots experiences with world priors at no extra
human/robot cost. We use <a href="https://segment-anything.com/">SAM</a> to segment
target objects and semantically augment them to different objects with shape, color,
texture variations.
</li>
<li>
<strong>Efficient Policy Representation:</strong> The resulting dataset is heavily
multi-modal and contains a rich diversity of skills, tasks, and scenarios. We adapt <a
href="https://tonyzhaozh.github.io/aloha/">action-chunking</a> to multi-task
settings to develop <strong>MT-ACT</strong> -- a novel efficient policy representation
that can ingest highly multi-modal dataset while avoiding over-fitting in low data
budget settings.
</li>
</ul>
</div>
<div class="column is-centered is-vcentered has-text-centered">
<video autoplay="" muted="" loop="" playsinline="" width="80%">
<source src="./media/roboagent_architecture.mp4" type="video/mp4">
</video>
</div>
<br>
<h3 style="font-family: 'Source Sans 3', sans-serif; font-weight: 400;">
RoboAgent is more sample-efficient than existing methods.
</h3>
<!-- <h2> RoboAgent is more sample-efficient than existing methods. </h2> -->
<br>
<div class="columns is-vcentered">
<div class="column is-half">
<p class="has-text-justified">
Figure on the right compares our proposed MT-ACT policy representation against several
imitation learning architectures. For this result we use environment variations that
include object pose changes and some lighting changes only. Somewhat similar to previous
works, we refer to this as L1-generalization. From our results we can clearly see that
using action-chunking to model sub-trajectories significantly outperforms all baselines,
thereby reinforcing the effectiveness of our proposed policy representation for sample
efficient learning.
</p>
</div>
<div class="column overlay-image has-text-centered">
<img src="media/l1_website.png">
</div>
</div>
<br>
<h3 style="font-family: 'Source Sans 3', sans-serif; font-weight: 400;">
RoboAgent performs well across multiple levels of generalization.
</h3>
<div class="column overlay-image has-text-centered">
<img src="media/l123_generalization_02.png">
</div>
<br>
<div class="columns is-vcentered">
<div class="column is-half">
<p class="has-text-justified">
Above figure shows the different levels of generalization we test our approach on.
We visualize levels of generalization, L1 with object pose changes, L2
with diverse table backgrounds and distractors and L3 with novel skill-object
combinations. Next we show how each method performs on these levels of generalization.
In a rigorous evaluation study under, we observe that MT-ACT significantly outperforms
all other
methods especially on harder generalization levels (L3).
</p>
</div>
<div class="column overlay-image has-text-centered">
<img src="media/main_results.png" width="500">
</div>
</div>
<br>
<h3 style="font-family: 'Source Sans 3', sans-serif; font-weight: 400;">
RoboAgent is highly scalable.
</h3>
<div class="columns is-vcentered">
<div class="column is-half">
<p class="has-text-justified">
Next we evaluate how RoboAgent performs with increasing levels of semantic
augmentations.
We evaluate this on one activity (5-skills). Below figure shows that with increased data
(i.e. more augmentations per frame)
the performance significantly improves across all generalization levels. Importantly,
the performance increase is much larger for the harder tasks (L3 generalization).
</p>
</div>
<div class="column overlay-image has-text-centered">
<img src="media/l123_generalization_scale.png" width="400">
</div>
</div>
</div>
</div>
</div>
</section>
<div class="container is-max-desktop">
<div class="content is-centered has-text-centered">
<!-- <br><br> -->
<div class="column has-text-justified is-vcentered">
<h2>RoboAgent can exhibit skills across diverse activities</h2>
<section class="hero is-light is-small">
<div class="hero-body">
<div class="container">
<h5 id="vidtit">
<b>
Baking Prep
</b>
</h5>
<div id="results-carousel" class="carousel results-carousel">
<div class="card" id="slide-open">
<div class="card-image">
<video poster="" controls muted loop playsinline height="100%">
<source
src="media/baking_prep/crop_720w/open_drawer_cropped_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Slide-Open Drawer
</div>
</div>
</div>
<div class="card" id="pick">
<div class="card-image">
<video poster="" controls muted loop playsinline height="100%">
<source
src="media/baking_prep/crop_720w/pick_butter_cropped_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Pick Butter
</div>
</div>
</div>
<div class="card" id="place">
<div class="card-image">
<video poster="" controls muted loop playsinline height="100%">
<source
src="media/baking_prep/crop_720w/place_butter_cropped_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Place Butter
</div>
</div>
</div>
<div class="card" id="slide-close">
<div class="card-image">
<video poster="" controls muted loop playsinline height="100%">
<source
src="media/baking_prep/crop_720w/close_drawer_cropped_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content ">
<div class="item__title" style="text-align: center;">
Slide-Close Drawer
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<section class="hero is-light is-small">
<div class="hero-body">
<div class="container">
<h5 id="vidtit">
<b>
Clean Kitchen
</b>
</h5>
<div id="results-carousel" class="carousel results-carousel">
<div class="card" id="slide-close">
<div class="card-image">
<video poster="" controls muted loop playsinline height="100%">
<source
src="media/clean_kitchen/crop_720w/close_drawer_cropped_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Slide-Close Drawer
</div>
</div>
</div>
<div class="card" id="flap-close">
<div class="card-image">
<video poster="" controls muted loop playsinline height="100%">
<source
src="media/clean_kitchen/crop_720w/close_oven_cropped_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Flap-Close Oven
</div>
</div>
</div>
<div class="card" id="pick">
<div class="card-image">
<video poster="" controls muted loop playsinline height="100%">
<source
src="media/clean_kitchen/crop_720w/pick_towel_cropped_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Pick Towel
</div>
</div>
</div>
<div class="card" id="wipe">
<div class="card-image">
<video poster="" controls muted loop playsinline height="100%">
<source
src="media/clean_kitchen/crop_720w/wipe_towel_cropped_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Wipe Counter
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- <section class="hero is-light is-small">
<div class="hero-body">
<div class="container">
<h5 id="vidtit">
<b>
Heat Soup
</b>
</h5>
<div id="results-carousel" class="carousel results-carousel">
<div class="item item-mask">
<video poster="" controls muted loop playsinline height="100%">
<source src="media/IMG_2715_cropped.mp4" type="video/mp4">
</video>
</div>
<div class="item item-mask">
<video poster="" controls muted loop playsinline height="100%">
<source src="media/IMG_2812_cropped.mp4" type="video/mp4">
</video>
</div>
<div class="item item-mask">
<video poster="" controls muted loop playsinline height="100%">
<source src="media/IMG_2820_cropped.mp4" type="video/mp4">
</video>
</div>
<div class="item item-mask">
<video poster="" controls muted loop playsinline height="100%">
<source src="medicda/IMG_2827_cropped.mp4" type="video/mp4">
</video>
</div>
</div>
</div>
</div>
</section> -->
<section class="hero is-light is-small">
<div class="hero-body">
<div class="container">
<h5 id="vidtit">
<b>
Serve Soup
</b>
</h5>
<div id="results-carousel" class="carousel results-carousel">
<div class="card" id="flap-open">
<div class="card-image">
<video poster="" controls muted loop playsinline height="100%">
<source src="media/serve_souo/crop_720w/open_oven_cropped_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Flap-Open Oven
</div>
</div>
</div>
<div class="card" id="pick">
<div class="card-image">
<video poster="" controls muted loop playsinline height="100%">
<source src="media/serve_souo/crop_720w/pick_bowl_cropped_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Pick Bowl
</div>
</div>
</div>
<div class="card" id="place">
<div class="card-image">
<video poster="" controls muted loop playsinline height="100%">
<source
src="media/serve_souo/crop_720w/place_bowl_cropped_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Place Bowl
</div>
</div>
</div>
<div class="card" id="flap-close">
<div class="card-image">
<video poster="" controls muted loop playsinline height="100%">
<source
src="media/serve_souo/crop_720w/close_oven_cropped_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Flap-Close Oven
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<section class="hero is-light is-small">
<div class="hero-body">
<div class="container">
<h5 id="vidtit">
<b>
Make Tea
</b>
</h5>
<div id="results-carousel" class="carousel results-carousel">
<div class="card" id="uncap">
<div class="card-image">
<video poster="" controls muted loop playsinline height="100%">
<source src="media/make_tea/crop_720w/unlid_cropped_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Uncap Lid
</div>
</div>
</div>
<div class="card" id="pick">
<div class="card-image">
<video poster="" controls muted loop playsinline height="100%">
<source src="media/make_tea/crop_720w/pick_teabag_cropped_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Pick Teabag
</div>
</div>
</div>
<div class="card" id="place">
<div class="card-image">
<video poster="" controls muted loop playsinline height="100%">
<source
src="media/make_tea/crop_720w/place_teabag_cropped_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Place Teabag
</div>
</div>
</div>
<div class="card" id="pick">
<div class="card-image">
<video poster="" controls muted loop playsinline height="100%">
<source src="media/make_tea/crop_720w/picklid_cropped_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Pick Lid
</div>
</div>
</div>
<div class="card" id="cap">
<div class="card-image">
<video poster="" controls muted loop playsinline height="100%">
<source src="media/make_tea/crop_720w/close_lid_cropped_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">
Cap Lid
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<section class="hero is-light is-small">
<div class="hero-body">
<div class="container">
<h5 id="vidtit">
<b>
Stow Bowl
</b>
</h5>
<div id="results-carousel" class="carousel results-carousel">
<div class="card" id="slide-open">
<div class="card-image">
<video poster="" controls muted loop playsinline height="100%">
<source
src="media/putting_away/crop_720w/open_drawer_cropped_720w_compress.mp4"
type="video/mp4">
</video>
</div>
<div class="card-content">
<div class="item__title" style="text-align: center;">