-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.html
681 lines (649 loc) · 40.5 KB
/
index.html
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
<html>
<head>
<!-- Google tag (gtag.js) -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-VTKHNTKBM4"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag(){dataLayer.push(arguments);}
gtag('js', new Date());
gtag('config', 'G-VTKHNTKBM4');
</script>
<meta charset="utf-8" />
<title>Botany-Bot</title>
<!-- TODO double check these are updated -->
<!-- Website Metadata -->
<meta content="Hierarchical grouping in 3D by training a scale-conditioned affinity field from multi-level masks"
name="description" />
<meta content="GrowSplat: Constructing Temporal Digital Twins of Plants with
Gaussian Splats" property="og:title" />
<meta content="4D reconstruct moving object parts from a monocular video using a scanned object model and imitate part motion with a robot"
property="og:description" />
<meta content="http://robot-see-robot-do.github.io/data/preview_card.png" property="og:image" />
<meta content="GrowSplat: Constructing Temporal Digital Twins of Plants with Gaussian Splats" property="twitter:title" />
<meta content="4D reconstruct moving object parts from a monocular video using a scanned object model and imitate part motion with a robot"
property="twitter:description" />
<meta content="http://robot-see-robot-do.github.io/data/preview_card.png" property="twitter:image" />
<meta property="og:type" content="website" />
<meta content="summary_large_image" name="twitter:card" />
<meta name="viewport" content="width=device-width, initial-scale=1, minimum-scale=1" />
<!-- Fonts -->
<link href="https://fonts.googleapis.com" rel="preconnect" />
<link href="https://fonts.gstatic.com" rel="preconnect" crossorigin="anonymous" />
<script src="https://ajax.googleapis.com/ajax/libs/webfont/1.6.26/webfont.js" type="text/javascript"></script>
<script
type="text/javascript">WebFont.load({ google: { families: ["Lato:100,100italic,300,300italic,400,400italic,700,700italic,900,900italic", "Montserrat:100,100italic,200,200italic,300,300italic,400,400italic,500,500italic,600,600italic,700,700italic,800,800italic,900,900italic", "Ubuntu:300,300italic,400,400italic,500,500italic,700,700italic", "Open Sans:300,300italic,400,400italic,600,600italic,700,700italic,800,800italic", "Changa One:400,400italic", "Varela Round:400", "Bungee Shade:regular", "Roboto:300,regular,500", "Bungee Outline:regular"] } });</script>
<!--[if lt IE 9]><script src="https://cdnjs.cloudflare.com/ajax/libs/html5shiv/3.7.3/html5shiv.min.js" type="text/javascript"></script><![endif]-->
<!-- JQuery, scripts etc -->
<script src="https://ajax.googleapis.com/ajax/libs/jquery/1/jquery.min.js"></script>
<script src="script.js" type="text/javascript"></script>
<script src="js/carousel_utils.js" type="text/javascript"></script>
<!-- Stylesheets; tabler icons, fonts, ...-->
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/@tabler/icons@latest/iconfont/tabler-icons.min.css">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/4.7.0/css/font-awesome.min.css">
<link href="style.css" rel="stylesheet" type="text/css" />
<link href="data/botanybot.png" rel="shortcut icon" type="image/x-icon" />
</head>
<body>
<div>
<!-- Title, Subtitle, Authors -->
<div>
<h1 class="title">
<span style="text-wrap: nowrap">GrowSplat: Constructing Temporal Digital Twins of Plants with </span> <span style="text-wrap: nowrap"> Gaussian Splats</span>
<!-- <span style="text-wrap: nowrap">Robot Do 🦾</span> -->
</h1>
<!-- <h1 class="subheader">Imitating Articulated Object Manipulation with Monocular 4D Reconstruction</h1> -->
<!-- Author list -->
<div class="base-row">
<!-- Refer to how author wrapping is performed in https://brentyi.github.io/tilted/ -->
<div style="display: flex; flex-wrap: wrap; justify-content: center;">
<div>
<a href="https://simeonoa.github.io/" target="_blank" class="author-text">
Simeon Adebola
</a><sup>1</sup>
<div style="width: 1.25em; display: inline-block"></div>
<a href="https://ehehee.github.io/" target="_blank" class="author-text">
Shangyu Xie
</a><sup>1</sup>
<div style="width: 1.25em; display: inline-block"></div>
<a href="https://chungmin99.github.io/" target="_blank" class="author-text">
Chung Min Kim
</a><sup>1</sup>
<div style="width: 1.25em; display: inline-block"></div>
<a href="https://kerrj.github.io/" target="_blank" class="author-text">
Justin Kerr
</a><sup>1</sup>
<div style="width: 1.25em; display: inline-block"></div>
<a href="https://scholar.google.com/citations?user=t_vRU4wAAAAJ&hl=nl/" target="_blank" class="author-text">
Bart van Marrewijk
</a><sup>3</sup>
</div>
<div>
<div style="width: 1.25em; display: inline-block"></div>
<a href="https://berkeleyautomation.github.io/GrowSplat/" target="_blank" class="author-text">
Meike van Vlaardingen
</a><sup>3</sup>
<div style="width: 1.25em; display: inline-block"></div>
<a href="https://research.wur.nl/en/persons/tim-van-daalen" target="_blank" class="author-text">
Tim van Daalen
</a><sup>3</sup>
<div style="width: 1.25em; display: inline-block"></div>
<a href="https://www.wur.nl/en/persons/robert-van-loo-1.htm" target="_blank" class="author-text">
Robert van Loo
</a><sup>3</sup>
<div style="width: 1.25em; display: inline-block"></div>
<a href="https://scholar.google.com/citations?user=vk2qKkYAAAAJ&hl=en" target="_blank" class="author-text">
Jose Luis Susa Rincon
</a><sup>3</sup>
</div>
<div>
<div style="width: 1.25em; display: inline-block"></div>
<a href="https://scholar.google.com.au/citations?user=AvvaaJcAAAAJ&hl=en" target="_blank" class="author-text">
Eugen Solowjow
</a><sup>2</sup>
<div style="width: 1.25em; display: inline-block"></div>
<a href="https://scholar.google.com.au/citations?user=gmrJR2cAAAAJ&hl=en" target="_blank" class="author-text">
Rick van Zedde
</a><sup>3</sup>
<div style="width: 1.25em; display: inline-block"></div>
<a href="https://goldberg.berkeley.edu/" target="_blank" class="author-text">
Ken Goldberg
</a><sup>1</sup>
</div>
</div>
</div>
<div style="text-align: center">
<h1 id="uc-berkeley"><sup>1</sup>UC Berkeley </h1>
<h1 id="uc-berkeley"><sup>2</sup>Siemens Research Lab, Berkeley</h1>
<h1 id="uc-berkeley"><sup>3</sup>Netherlands Plant Eco-phenotyping Centre, Wageningen University
and Research </h1>
<!-- <span class="text-star">*</span>
Denotes Equal Contribution -->
</div>
</div>
<!-- Submission status -->
<div class="title-row">
<!-- <h2 class="subheader">CoRL 2024 (Oral)</h1> -->
<h2 class="subheader">Under Review</h1>
</div>
<!-- Paper / code / data URLs -->
<!-- TODO: Update arxiv/code/data link -->
<div class="base-row add-top-padding">
<!-- Paper -->
<a href="https://berkeleyautomation.github.io/Botany-Bot/" target="_blank" class="link-block">
<figure>
<img src="https://uploads-ssl.webflow.com/51e0d73d83d06baa7a00000f/5cab99df4998decfbf9e218e_paper-01.png"
alt="paper"
srcset="https://uploads-ssl.webflow.com/51e0d73d83d06baa7a00000f/5cab99df4998decfbf9e218e_paper-01-p-500.png"
style="max-height: 4em" />
</figure>
<figcaption>
<strong class="link-labels-text">Paper </strong>
</figcaption>
<figcaption>
<strong class="link-labels-text">Coming Soon </strong>
</figcaption>
</a>
<!-- Code -->
<a href="https://berkeleyautomation.github.io/Botany-Bot/" target="_blank" class="link-block">
<figure>
<img src="https://uploads-ssl.webflow.com/51e0d73d83d06baa7a00000f/5cae3b53b42ebb3dd4175a82_68747470733a2f2f7777772e69636f6e66696e6465722e636f6d2f646174612f69636f6e732f6f637469636f6e732f313032342f6d61726b2d6769746875622d3235362e706e67.png"
alt="code" style="max-height: 4em" />
</figure>
<figcaption>
<strong class="link-labels-text"></Code ></strong>
</figcaption>
<figcaption>
<strong class="link-labels-text">Coming Soon </strong>
</figcaption>
</a>
<!-- Data -->
<a href="https://berkeleyautomation.github.io/Botany-Bot/" target="_blank" class="link-block">
<figure>
<img src="data/database_icon.jpg" alt="data" style="max-height: 4em" />
</figure>
<figcaption>
<strong class="link-labels-text">Data </strong>
</figcaption>
<figcaption>
<strong class="link-labels-text">Coming Soon </strong>
</figcaption>
</a>
</div>
<!-- TL;DR + Teaser video -->
<!-- <div class="section base-row add-top-padding">
<h1 class="tldr">
<b>TL;DR</b>:
GrowSplat uses Gaussian splatting, plant registration and view rendering to create temporal digital twins of plant from industrial scale data.
</h1>
<video id="main-video" autobuffer muted autoplay loop controls playsinline>
<source id="mp4" src="data/supp_video.mp4" type="video/mp4">
</video>
</div> -->
<!-- Abstract -->
<div class="section base-row add-top-padding">
<h1>Abstract</h1>
<p class="paragraph">
Accurate temporal reconstructions of plant growth are essential for plant phenotyping and breeding, yet remain challenging due to complex geometries, occlusions, and non-rigid deformations of plants. We present a novel framework for building temporal digital twins of plants by combining 3D
Gaussian Splatting with a robust sample alignment pipeline. Our method begins by reconstructing Gaussian Splats from multi-view camera data, then leverages a two-stage registration approach: coarse alignment through feature-based matching and Fast Global Registration, followed by fine alignment with Iterative Closest Point. This pipeline yields a consistent 4D model of plant development in discrete time steps. We evaluate the approach on data from the Netherlands Plant Eco-phenotyping Center, demonstrating detailed temporal reconstructions of Sequoia and Quinoa species.
</p>
</div>
<!-- <div class="section base-row add-top-padding">
<h1 class="tldr">
Humans imitate manipulation by watching <b>object</b> motion, not <b>hand</b> motion. RSRD does the same. <br>
This enables imitation from a single video robust to orientation and across the human-robot embodiment gap.
</h1>
<img src="data/rsrd_mainfig.jpg" style="max-width: 100%" />
</div>
<div class="section add-top-padding">
<div class="base-row">
<img src="data/favicon.png" style="height: 40px; margin-left: 5px; margin-right: 5px; margin-top: 40px; margin-bottom: 40px;">
<img src="data/favicon.png" style="height: 40px; margin-left: 5px; margin-right: 5px; margin-top: 40px; margin-bottom: 40px;">
<img src="data/favicon.png" style="height: 40px; margin-left: 5px; margin-right: 5px; margin-top: 40px; margin-bottom: 40px;">
</div>
</div> -->
<!-- <div class="section base-row add-top-padding">
<h1>4D Part Reconstruction</h1>
<p class="paragraph">
RSRD takes in <b>1) </b>a multi-view object scan and <b>2)</b> a monocular demonstration video.
By creating part-aware 3D representations using <a href="https://www.garfield.studio/" class="author-text" target="_blank">GARField</a> (parts, toggle for clusters) and
<a href="https://github.com/facebookresearch/dinov2" class="author-text"target="_blank">DINOv2</a> (tracking SE3 pose),
these smartphone-captured inputs can generate these 4D reconstructions:
</p> -->
<!-- <div id="main-results">
<h1 class="tldr"><b>Input demonstration video</b></h1> -->
<!-- <div class="video-carousel-wrapper">
<div class="carousel-container" id="videoCarousel"> -->
<!-- Add your video elements here -->
<!-- <div class="carousel-item" id="bear_video">
<video autoplay muted loop playsinline height="400px">
<source src="data/demo_vids/cal_bear_naked_wave.mp4" type="video/mp4">
</video>
</div>
<div class="carousel-item" id="nerfgun_video">
<video autoplay muted loop playsinline height="400px">
<source src="data/demo_vids/nerfgun_final.mp4" type="video/mp4">
</video>
</div>
<div class="carousel-item" id="redbox_video">
<video autoplay muted loop playsinline height="400px">
<source src="data/demo_vids/buddha_empty_close.mp4" type="video/mp4">
</video>
</div>
<div class="carousel-item" id="scissors_video">
<video autoplay muted loop playsinline height="400px">
<source src="data/demo_vids/scissors2.mp4" type="video/mp4">
</video>
</div> -->
<!-- <div class="carousel-item" id="sunglasses_video">
<video autoplay muted loop playsinline height="400px">
<source src="data/demo_vids/sunglasses_fold.mp4" type="video/mp4">
</video>
</div>
<div class="carousel-item" id="ledlight_video">
<video autoplay muted loop playsinline height="400px">
<source src="data/demo_vids/led_light_unfold3.mp4" type="video/mp4">
</video>
</div>
<div class="carousel-item" id="stapler_video">
<video autoplay muted loop playsinline height="400px">
<source src="data/demo_vids/stapler_fold.mp4" type="video/mp4">
</video>
</div>
<div class="carousel-item" id="wirecutters_video">
<video autoplay muted loop playsinline height="400px">
<source src="data/demo_vids/wire_cutters_close2.mp4" type="video/mp4">
</video>
</div>
<div class="carousel-item" id="usbplug_video">
<video autoplay muted loop playsinline height="400px">
<source src="data/demo_vids/usb_plug_unplug3.mp4" type="video/mp4">
</video>
</div>
</div>
</div> -->
<!-- <h1 class="tldr"><b>
<img src="data/drag_icon.png" alt="" class="inline-image">Click and move me!<img src="data/drag_icon.png" alt="" class="inline-image">
</b></h1> -->
<!-- <div id="iframe-container" class="iframe-container">
<div class="click-and-move-overlay">
<h1 class="tldr">
<b>
<img src="data/drag_icon.png" alt="" class="inline-image">
Click and move me!
<img src="data/drag_icon.png" alt="" class="inline-image">
</b>
</h1>
</div>
<iframe
id="bear"
class = "iframe"
data-src="https://rsrd-anonymous.github.io/build/?playbackPath=https://rsrd-anonymous.github.io/recordings/bear_animation.viser&initDistanceScale=1.1&gaussianGroupColorShuffleSeed=1"
></iframe>
<iframe
id = "nerfgun"
class = "iframe"
data-src="https://rsrd-anonymous.github.io/build/?playbackPath=https://rsrd-anonymous.github.io/recordings/nerfgun_animation.viser&gaussianGroupColorShuffleSeed=3"
></iframe>
<iframe
id = "redbox"
class = "iframe"
data-src="https://rsrd-anonymous.github.io/build/?playbackPath=https://rsrd-anonymous.github.io/recordings/red_box_animation.viser&gaussianGroupColorShuffleSeed=13"
></iframe>
<iframe
id = "scissors"
class = "iframe"
data-src="https://rsrd-anonymous.github.io/build/?playbackPath=https://rsrd-anonymous.github.io/recordings/scissors_animation.viser&initDistanceScale=1.1&gaussianGroupColorShuffleSeed=4"
></iframe>
<iframe
id = "sunglasses"
class = "iframe"
data-src="https://rsrd-anonymous.github.io/build/?playbackPath=https://rsrd-anonymous.github.io/recordings/sunglasses_animation.viser&initDistanceScale=1.2&gaussianGroupColorShuffleSeed=0"
></iframe>
<iframe
id="ledlight"
class = "iframe"
data-src="https://rsrd-anonymous.github.io/build/?playbackPath=https://rsrd-anonymous.github.io/recordings/led_light.viser&initDistanceScale=1.3&gaussianGroupColorShuffleSeed=5"
></iframe>
<iframe
id="stapler"
class = "iframe"
data-src="https://rsrd-anonymous.github.io/build/?playbackPath=https://rsrd-anonymous.github.io/recordings/stapler.viser&gaussianGroupColorShuffleSeed=4"
></iframe>
<iframe
id="usbplug"
class = "iframe"
data-src="https://rsrd-anonymous.github.io/build/?playbackPath=https://rsrd-anonymous.github.io/recordings/usb_plug.viser&gaussianGroupColorShuffleSeed=3"
></iframe>
<iframe
id="wirecutters"
class = "iframe"
data-src="https://rsrd-anonymous.github.io/build/?playbackPath=https://rsrd-anonymous.github.io/recordings/wirecutters.viser&initDistanceScale=1.4&gaussianGroupColorShuffleSeed=9"
></iframe>
</div>
<div style="position: relative; display: flex;">
<button class="results-slide-arrow" id="results-slide-arrow-prev" onclick="results_slide_left()">
‹
</button>
<div class="results-slide-row" id="results-objs-scroll">
<div data-img-src="data/thumbnails/bear_zoomed.jpg" data-id="bear-thumb" data-label="Bear"></div>
<div data-img-src="data/thumbnails/nerfgun_zoomed.jpg" data-id="nerfgun-thumb" data-label="Nerfgun"></div>
<div data-img-src="data/thumbnails/red_box_zoomed.jpg" data-id="redbox-thumb" data-label="Box"></div>
<div data-img-src="data/thumbnails/scissors_zoomed.jpg" data-id="scissors-thumb" data-label="Scissors"></div>
<div data-img-src="data/thumbnails/sunglasses_zoomed.jpg" data-id="sunglasses-thumb" data-label="Sunglasses"></div>
<div data-img-src="data/thumbnails/led_light_zoomed.jpg" data-id="ledlight-thumb" data-label="LED Light"></div>
<div data-img-src="data/thumbnails/stapler_zoomed.jpg" data-id="stapler-thumb" data-label="Stapler"></div>
<div data-img-src="data/thumbnails/wirecutter_zoomed.jpg" data-id="wirecutters-thumb" data-label="Wirecutters"></div>
<div data-img-src="data/thumbnails/usb_plug_zoomed.jpg" data-id="usbplug-thumb" data-label="USB Plug"></div>
</div>
<button class="results-slide-arrow" id="results-slide-arrow-next" onclick="results_slide_right()">
›
</button>
</div>
</div>
<p class="tldr">These 4D reconstructions are rendered in-browser! If you think that's cool, check out <a href="https://viser.studio/latest/" class="author-text" target="_blank">Viser</a>!</p>
</div>
<div class="section add-top-padding">
<div class="base-row">
<img src="data/favicon.png" style="height: 40px; margin-left: 5px; margin-right: 5px; margin-top: 40px; margin-bottom: 40px;">
<img src="data/favicon.png" style="height: 40px; margin-left: 5px; margin-right: 5px; margin-top: 40px; margin-bottom: 40px;">
<img src="data/favicon.png" style="height: 40px; margin-left: 5px; margin-right: 5px; margin-top: 40px; margin-bottom: 40px;">
</div>
</div> -->
<!-- <div class="section base-row add-top-padding">
<h1>Robot Motion Retargeting</h1>
<p class="paragraph">
After recovering 3D part motion, RSRD optimizes grasps and robot motions to reproduce the 4D reconstructions.
</p>
<div id="iframe-container" class="iframe-container">
<div class="click-and-move-overlay">
<h1 class="tldr">
<b>
<img src="data/drag_icon.png" alt="" class="inline-image">
Click and move me!
<img src="data/drag_icon.png" alt="" class="inline-image">
</b>
</h1>
</div>
<iframe class="iframe show"
src="https://rsrd-anonymous.github.io/build/?playbackPath=https://rsrd-anonymous.github.io/recordings/robot_bear.viser&initDistanceScale=1.5&gaussianGroupColorShuffleSeed=1"
></iframe>
</div>
<p class="paragraph">
These can be physically executed on a real robot to produce the demonstrated motion:
</p>
<div style="position: relative; display: flex;">
<button class="results-slide-arrow" id="results-slide-arrow-prev" onclick="vid_slide_left()">
‹
</button>
<div class="video-scroll" id="result-video-scroll">
<div class="video-scroll-card">
<video autoplay muted loop playsinline>
<source src="data/bear_robot_1_trimmed.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
<div class="video-scroll-card">
<video autoplay muted loop playsinline>
<source src="data/nerfgun_robot_1_trimmed.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
<div class="video-scroll-card">
<video autoplay muted loop playsinline>
<source src="data/scissors_white_out.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
<div class="video-scroll-card">
<video autoplay muted loop playsinline>
<source src="data/scissors_red_out.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
<div class="video-scroll-card">
<video autoplay muted loop playsinline>
<source src="data/red_box_out.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
<div class="video-scroll-card">
<video autoplay muted loop playsinline>
<source src="data/sunglasses_out.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</div>
<button class="results-slide-arrow" id="results-slide-arrow-next" onclick="vid_slide_right()">
›
</button>
</div>
<p class="paragraph">
RSRD's visual imitation is <b>object-centric</b>, allowing it to adapt to <em>different</em> object orientations with the <em>same</em> demo:
<div class="video-container">
<div class="video-item">
<p><b>0 degrees rotated</b></p>
<video autoplay muted loop playsinline>
<source src="data/bear_robot_2_trimmed.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
<div class="video-item">
<p><b>180 degrees rotated</b></p>
<video autoplay muted loop playsinline>
<source src="data/bear_robot_1_trimmed.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</div>
<div class="video-container">
<div class="video-item">
<p><b>0 degrees rotated</b></p>
<video autoplay muted loop playsinline>
<source src="data/nerfgun_straight_out.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
<div class="video-item">
<p><b>30 degrees rotated</b></p>
<video autoplay muted loop playsinline>
<source src="data/nerfgun_half_turned_out.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
<div class="video-item">
<p><b>45 degrees rotated</b></p>
<video autoplay muted loop playsinline>
<source src="data/nerfgun_robot_2_trimmed.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</div>
</p>
</div> -->
<!-- Just for vertical spacing... -->
<!-- <div class="section add-top-padding">
<div class="base-row">
<img src="data/favicon.png" style="height: 40px; margin-left: 5px; margin-right: 5px; margin-top: 40px; margin-bottom: 40px;">
<img src="data/favicon.png" style="height: 40px; margin-left: 5px; margin-right: 5px; margin-top: 40px; margin-bottom: 40px;">
<img src="data/favicon.png" style="height: 40px; margin-left: 5px; margin-right: 5px; margin-top: 40px; margin-bottom: 40px;">
</div>
</div>
</div>
<div class="section base-row add-top-padding">
<h1>How it works</h1>
<div class="separator"></div>
<p class="tldr"><b>4D Differentiable Part Models</b></p>
<p class="paragraph">
4D-DPM decomposes objects into parts with <a href="https://www.garfield.studio/" class="author-text" target="_blank">GARField</a>, and trains part-centric feature fields
on top of these. Each part is assigned a trainable 6D pose parameter which is optimized with gradient descent.
DINO improves dramatically over photometric tracking as a more robust feature target, and allows reconstructing a broad range of open-world objects.
</p>
<img src="data/4ddpm_fig.jpg" style="max-width: 100%; padding-bottom: 30px" />
<p class="paragraph">
Because 4D-DPM uses gradient descent, any differentiable prior is easily incorporated like temporal smoothness and rigidity.
</p>
<img src="data/4ddpm_ablation.jpg" style="max-width: 90%" />
</div> -->
<!-- <div class="section base-row add-top-padding">
<h1>Retargeting Robot Trajectory</h1>
<p class="paragraph">
With the recovered 3D part motion and the object placed in the robot workspace,
RSRD now can <i>do</i> the motion demonstrated in the video.
Motions can be retargeted regardless of object pose!
Three main takeaways are:
<ol class="paragraph">
<li>
<b>Hand-Guided Part Selection</b>:
From the 4D motion reconstruction, RSRD must automatically detect which parts need to be manipulated to reproduce it.
Not all moving parts are relevant, like the wooden figurine's hand.
We use hand position as prior for part selection (using <a class="author-text" href="https://geopavlakos.github.io/hamer/">HaMeR</a>),
but <i>do not</i> use the hand contact points, as explained in (2).
<img src="data/rsrd_hands.jpg" style="max-width: 90%" />
</li><br>
<li>
<b>Part-centric Grasps</b>:
We cannot use detected finger contact locations as grasp points, as they may either jitter or become kinematically unreachable.
Also, a robot must remain rigidly attached to the object part during the entire motion, whileas
humans can do so much more — change contact points by shuffling fingers, or do prehensile motions.
<div class="video-container" style="max-width: 600px;">
<div class="video-item">
<video autoplay muted loop playsinline>
<source src="data/nerfgun_graspsearch.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</div>
</li><br>
<li>
<b>Bimanual Robot Pose Optimization</b>:
We exhaustively search for collision-free, kinematically feasible robot trajectories: we first use
<a href="https://github.com/brentyi/jaxls" class="author-text" target="_blank">jaxls</a>
to optimize a robot trajectory to fit the robot end-effector waypoints using a Levenberg-Marquardt solver.
Then, we use
<a href="https://curobo.org/" class="author-text" target="_blank">cuRobo</a>
to plan collision-free robot approach motions and for all collision avoidance checks.
<br>
Below we vary the object pose, and visualize below some bimanual IK solutions for each object motion.
<div class="video-container">
<div class="video-item">
<video autoplay muted loop playsinline>
<source src="data/nerfgun_posevary.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
<div class="video-item">
<video autoplay muted loop playsinline>
<source src="data/bear_posevary.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
<div class="video-item">
<video autoplay muted loop playsinline>
<source src="data/scissors_posevary.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</div>
</li>
</ol>
</p>
</div>
<!-- Just for vertical spacing... -->
<!-- <div class="section add-top-padding">
<div class="base-row">
<img src="data/favicon.png" style="height: 40px; margin-left: 5px; margin-right: 5px; margin-top: 40px; margin-bottom: 40px;">
<img src="data/favicon.png" style="height: 40px; margin-left: 5px; margin-right: 5px; margin-top: 40px; margin-bottom: 40px;">
<img src="data/favicon.png" style="height: 40px; margin-left: 5px; margin-right: 5px; margin-top: 40px; margin-bottom: 40px;">
</div>
</div> --> -->
<!-- <div class="section base-row add-top-padding">
<h1>Limitations and Failures</h1>
<p class="paragraph">
4D monocular reconstruction is an extremely under-constrained and challenging problem, and 4D-DPM still suffers from sensitivity to demonstration viewpoint, occlusions, reconstruction quality, and more.
It also requires some hyper-parameter tuning of regularizers like the ARAP loss, and can frustratingly fail due to incorrect or incomplete part segmentations. More work is needed to adapt the approach for more in-the-wild videos.
In addition, while we show how the 4D reconstruction enables a robot to imitate with motion planning, RSRD as designed cannot scale to multiple demonstration videos of the same object, it can only mimick motion from one video.
Learning to manipulate from more demonstrations, perhaps with policy learning, is an exciting future direction!
</p>
<p class="paragraph">
<b>Camera Angle Sensitivity:</b> Since RSRD uses only a single video, it can be sensitive to camera pose during the demonstration, as illustrated in the tracking failure below.
The same scan of the laptop can work or fail depending on the camera angle.
</p>
<div class="video-container">
<div class="video-item">
<video autoplay muted loop playsinline>
<source src="data/laptop_trackfail.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
<div class="video-item">
<video autoplay muted loop playsinline>
<source src="data/laptop_tracksucc.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</div>
<p class="paragraph">
<b>Difficulty with feature-less parts:</b> 4D-DPM can struggle with parts which look similar from multiple angles, or have
not enough visual features for DINO to pick up on. In these videos the tail of the plushie and cable of the charger rotate along their major axis, making robot
execution difficult.
</p>
<div class="video-container">
<div class="video-item">
<video autoplay muted loop playsinline>
<source src="data/garfield_tailrotate.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
<div class="video-item">
<video autoplay muted loop playsinline>
<source src="data/mac_charger_cablerotate.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</div>
<p class="paragraph">
<b>Poor segmentation or reconstruction</b>: If the object scan is poor, or the part segmentation is incomplete or severely over-segmented,
4D-DPM can be unstable and lead to catastrophic failure like below.
</p>
<div class="video-container">
<div class="video-item">
<video autoplay muted loop playsinline>
<source src="data/mac_charger_catastrophic.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</div>
<p class="paragraph">
<b>Hand occlusions</b>: hand occlusions can interfere with part motion recovery, such as with the leg of this sculpture
</p>
<div class="video-container">
<div class="video-item">
<video autoplay muted loop playsinline>
<source src="data/hand_occ.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</div>
</div> -->
<!-- <div class="section citation" style="margin-top: 50px">
<h1 id="abstract">Citation </h1>
<p class="paragraph"> If you use this work or find it helpful, please consider citing: (bibtex) </p>
<pre id="codecell0">@inproceedings{kerr2024rsrd,
title={Robot See Robot Do: Imitating Articulated Object Manipulation with Monocular 4D Reconstruction},
author={Justin Kerr and Chung Min Kim and Mingxuan Wu and Brent Yi and Qianqian Wang and Ken Goldberg and Angjoo Kanazawa},
booktitle={8th Annual Conference on Robot Learning},
year={2024},
url={https://openreview.net/forum?id=2LLu3gavF1}
} -->
</pre>
</div>
<footer>
<div class="section" style="margin-top: 40px;">
<p class="paragraph">
The website template is adapted from the <a class="author-text" href="https://robot-see-robot-do.github.io/">Robot See Robot Do</a>
project page.
</p>
</div>
</footer>
</div>
</body>