-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
867 lines (813 loc) · 38.2 KB
/
index.html
File metadata and controls
867 lines (813 loc) · 38.2 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
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<meta name="description" content="DARP: Difference-Aware Retrieval Policies for Imitation Learning. A novel framework for robust robot manipulation and visuomotor control.">
<meta name="keywords" content="DARP, Imitation Learning, Robotics, Machine Learning, ICLR 2026, Computer Science, University of Washington, Diffusion Policy, RAG for Imitation Learning, RAG for Behavior Cloning, Retrieval Augmented Generation for BC, Retrieval-Augmented Behavior Cloning, RAG Imitation Learning, RAG for Robotics">
<meta name="author" content="Quinn Pfeifer, Ethan Pronovost, Paarth Shah, Khimya Khetarpal, Siddhartha Srinivasa, Abhishek Gupta">
<meta name="robots" content="index, follow">
<title>DARP</title>
<link rel="shortcut icon" href="https://www.cs.washington.edu/wp-content/themes/uw_wp_theme/favicon.ico">
<link rel="stylesheet" href="style.css">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.5.1/css/all.min.css">
<script id="MathJax-script" async src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js"></script>
<script type="importmap">
{
"imports": {
"three": "https://cdn.jsdelivr.net/npm/three@0.160.0/build/three.module.js",
"three/addons/": "https://cdn.jsdelivr.net/npm/three@0.160.0/examples/jsm/"
}
}
</script>
<script>
window.MathJax = {
tex: {
inlineMath: [['$', '$'], ['\\(', '\\)']],
displayMath: [['$$', '$$'], ['\\[', '\\]']]
}
};
</script>
</head>
<body>
<!-- Progress Sidebar -->
<nav class="progress-sidebar" aria-label="Page progress">
<div class="progress-track">
<div class="progress-line"></div>
<div class="progress-line-fill"></div>
<div class="progress-item" data-section="hero">
<a href="#">
<span class="progress-dot"></span>
<span class="progress-label">Intro</span>
</a>
</div>
<div class="progress-item" data-section="architecture">
<a href="#architecture">
<span class="progress-dot"></span>
<span class="progress-label">Architecture</span>
</a>
</div>
<div class="progress-item" data-section="results">
<a href="#results">
<span class="progress-dot"></span>
<span class="progress-label">Results</span>
</a>
</div>
<div class="progress-item" data-section="divergence">
<a href="#divergence">
<span class="progress-dot"></span>
<span class="progress-label">Divergence Analysis</span>
</a>
</div>
<div class="progress-item" data-section="multimodal">
<a href="#multimodal">
<span class="progress-dot"></span>
<span class="progress-label">Multimodal Distributions</span>
</a>
</div>
<div class="progress-item" data-section="diffusion">
<a href="#diffusion">
<span class="progress-dot"></span>
<span class="progress-label">Diffusion Policy</span>
</a>
</div>
<div class="progress-item" data-section="bibtex">
<a href="#bibtex">
<span class="progress-dot"></span>
<span class="progress-label">BibTeX</span>
</a>
</div>
</div>
</nav>
<div class="main-content">
<div class="hero-flow">
<div class="hero-text">DARP</div>
<div class="sub-hero-text">Difference-Aware Retrieval Policies for Imitation Learning</div>
<div class="authors">
<a href="https://quinn.computer" target="_blank">Quinn Pfeifer</a><sup>1</sup>,
<a href="https://www.linkedin.com/in/epronovost/" target="_blank">Ethan Pronovost</a><sup>1</sup>,
<a href="https://www.linkedin.com/in/paarth-shah-bb6743b1/" target="_blank">Paarth Shah</a><sup>2</sup>,
<a href="https://kkhetarpal.github.io/" target="_blank">Khimya Khetarpal</a><sup>3, 4</sup>,
<a href="https://goodrobot.ai/" target="_blank">Siddhartha Srinivasa</a><sup>1</sup>,
<a href="https://homes.cs.washington.edu/~abhgupta/" target="_blank">Abhishek Gupta</a><sup>1, 2</sup><br>
<span class="affiliation">
<sup>1</sup>Paul G. Allen School of Computer Science & Engineering, University of Washington, <sup>2</sup>Toyota Research Institute, <sup>3</sup>Google DeepMind, <sup>4</sup>Mila
</span>
</div>
<div class="hero-actions">
<a class="cta primary" href="https://openreview.net/pdf?id=9AA27en4go" target="_blank">
<svg width="16" height="16" viewBox="0 0 24 24" fill="currentColor"
style="vertical-align: -2px; margin-right: 6px;">
<path
d="M14 2H6a2 2 0 0 0-2 2v16a2 2 0 0 0 2 2h12a2 2 0 0 0 2-2V8l-6-6zm-1 2l5 5h-5V4zM6 20V4h5v7h7v9H6zm2-6h8v2H8v-2zm0-4h4v2H8v-2zm0 8h8v2H8v-2z" />
</svg>
Read the paper
</a>
<a class="cta dark" href="https://github.com/WEIRDLabUW/darp" target="_blank">
<svg width="18" height="18" viewBox="0 0 24 24" fill="currentColor"
style="vertical-align: -3px; margin-right: 6px;">
<path
d="M12 0c-6.626 0-12 5.373-12 12 0 5.302 3.438 9.8 8.207 11.387.599.111.793-.261.793-.577v-2.234c-3.338.726-4.033-1.416-4.033-1.416-.546-1.387-1.333-1.756-1.333-1.756-1.089-.745.083-.729.083-.729 1.205.084 1.839 1.237 1.839 1.237 1.07 1.834 2.807 1.304 3.492.997.107-.775.418-1.305.762-1.604-2.665-.305-5.467-1.334-5.467-5.931 0-1.311.469-2.381 1.236-3.221-.124-.303-.535-1.524.117-3.176 0 0 1.008-.322 3.301 1.23.957-.266 1.983-.399 3.003-.404 1.02.005 2.047.138 3.006.404 2.291-1.552 3.297-1.23 3.297-1.23.653 1.653.242 2.874.118 3.176.77.84 1.235 1.911 1.235 3.221 0 4.609-2.807 5.624-5.479 5.921.43.372.823 1.102.823 2.222v3.293c0 .319.192.694.801.576 4.765-1.589 8.199-6.086 8.199-11.386 0-6.627-5.373-12-12-12z" />
</svg>
Code
</a>
<a class="cta colab" href="https://drive.google.com/file/d/1N0kBjaT773HkzESaXw884wmsmEpZJjEy/view?usp=sharing" target="_blank">
<svg xmlns:xlink="http://www.w3.org/1999/xlink" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24" preserveAspectRatio="xMidYMid meet" focusable="false" style="pointer-events: none; display: block; width: 32px; height: 32px; vertical-align: -3px; margin-right: 6px;"><g>
<path d="M4.54,9.46,2.19,7.1a6.93,6.93,0,0,0,0,9.79l2.36-2.36A3.59,3.59,0,0,1,4.54,9.46Z" style="" fill="#E8710A"/>
<path d="M2.19,7.1,4.54,9.46a3.59,3.59,0,0,1,5.08,0l1.71-2.93h0l-.1-.08h0A6.93,6.93,0,0,0,2.19,7.1Z" style="" fill="#F9AB00"/>
<path d="M11.34,17.46h0L9.62,14.54a3.59,3.59,0,0,1-5.08,0L2.19,16.9a6.93,6.93,0,0,0,9,.65l.11-.09" style="" fill="#F9AB00"/>
<path d="M12,7.1a6.93,6.93,0,0,0,0,9.79l2.36-2.36a3.59,3.59,0,1,1,5.08-5.08L21.81,7.1A6.93,6.93,0,0,0,12,7.1Z" style="" fill="#F9AB00"/>
<path d="M21.81,7.1,19.46,9.46a3.59,3.59,0,0,1-5.08,5.08L12,16.9A6.93,6.93,0,0,0,21.81,7.1Z" style="" fill="#E8710A"/>
</g></svg>
Google Colab
</a>
</div>
<!-- Hero Pipeline Animation -->
<div class="hero-pipeline-container">
<canvas id="heroPipelineCanvas"></canvas>
</div>
</div>
<!-- DARP Architecture Section -->
<div id="architecture" class="architecture-section">
<div class="tagline">DARP Architecture</div>
<p class="section-description">
Hover over the state space to preview neighbors, then click to lock in a query state and see how DARP processes it through neighbor retrieval,
difference vector calculation, and aggregation. Watch how the predicted action (blue) emerges from averaging neighbor actions. Click again to unlock your query point and enter hover mode again.
</p>
<div class="architecture-container">
<!-- Left: State Space & Retrieval -->
<div class="arch-panel">
<div class="panel-title">1. Retrieve Neighbors & Visualize Inputs</div>
<canvas id="architectureCanvas" class="arch-canvas"></canvas>
<div class="arch-legend">
<div class="legend-row">
<span class="legend-dot query-dot"></span>
<span>Query State (s<sub>q</sub>)</span>
</div>
<div class="legend-row">
<span class="legend-dot neighbor-dot"></span>
<span>Neighbor State (s*)</span>
</div>
<div class="legend-row">
<svg width="24" height="12" style="vertical-align: middle;">
<defs>
<marker id="legend-action" markerWidth="6" markerHeight="6" refX="5" refY="3" orient="auto">
<polygon points="0 0, 6 3, 0 6" fill="#dc2626" />
</marker>
</defs>
<line x1="0" y1="6" x2="24" y2="6" stroke="#dc2626" stroke-width="1.5" marker-end="url(#legend-action)" />
</svg>
<span>Action (a*)</span>
</div>
<div class="legend-row">
<svg width="24" height="12" style="vertical-align: middle;">
<defs>
<marker id="legend-delta" markerWidth="6" markerHeight="6" refX="5" refY="3" orient="auto">
<polygon points="0 0, 6 3, 0 6" fill="#16a34a" />
</marker>
</defs>
<line x1="0" y1="6" x2="24" y2="6" stroke="#16a34a" stroke-width="1.5" stroke-dasharray="4,2" marker-end="url(#legend-delta)" />
</svg>
<span>Difference (Δ)</span>
</div>
<div class="legend-row">
<svg width="24" height="12" style="vertical-align: middle;">
<defs>
<marker id="legend-prediction" markerWidth="6" markerHeight="6" refX="5" refY="3" orient="auto">
<polygon points="0 0, 6 3, 0 6" fill="#2563eb" />
</marker>
</defs>
<line x1="0" y1="6" x2="24" y2="6" stroke="#2563eb" stroke-width="1.5" marker-end="url(#legend-prediction)" />
</svg>
<span>Predicted Action (â<sub>q</sub>)</span>
</div>
</div>
</div>
<!-- Middle: Processing -->
<div class="arch-panel">
<div class="panel-title">2. Process Each Neighbor: f<sub>θ</sub>(s*, a*, Δ) → a'</div>
<div id="neighborCards" class="neighbor-cards">
<div class="empty-state">
<svg width="48" height="48" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="1.5">
<circle cx="12" cy="12" r="10"></circle>
<path d="M12 6v6l4 2"></path>
</svg>
<p>Click a query state to begin</p>
</div>
</div>
</div>
<!-- Right: Aggregation -->
<div class="arch-panel">
<div class="panel-title">3. Aggregate: â<sub>q</sub> = g<sub>ψ</sub>({a'<sub>1</sub>, ..., a'<sub>k</sub>})</div>
<div id="aggregationPanel" class="aggregation-panel">
<div class="empty-state">
<svg width="48" height="48" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="1.5">
<path d="M9 5H7a2 2 0 00-2 2v12a2 2 0 002 2h10a2 2 0 002-2V7a2 2 0 00-2-2h-2M9 5a2 2 0 002 2h2a2 2 0 002-2M9 5a2 2 0 012-2h2a2 2 0 012 2"></path>
</svg>
<p>Awaiting neighbor predictions</p>
</div>
</div>
</div>
</div>
<div class="architecture-explanation">
<h3>How DARP Works</h3>
<div class="explanation-grid">
<div class="explanation-item">
<div class="explanation-number">1</div>
<div class="explanation-content">
<strong>Retrieve Neighbors & Compute Differences</strong>
<p>For query state <code>s<sub>q</sub></code>, find the k closest states from expert demonstrations. For each neighbor, calculate the difference vector <code>Δ<sub>i</sub> = s*<sub>i</sub> - s<sub>q</sub></code> and form input tuple <code>(s*<sub>i</sub>, a*<sub>i</sub>, Δ<sub>i</sub>)</code>.</p>
</div>
</div>
<div class="explanation-item">
<div class="explanation-number">2</div>
<div class="explanation-content">
<strong>Predict Action Per Neighbor</strong>
<p>Pass each tuple through network <code>f<sub>θ</sub></code> to get action candidate: <code>a'<sub>i</sub> = f<sub>θ</sub>(s*<sub>i</sub>, a*<sub>i</sub>, Δ<sub>i</sub>)</code>. This processes each neighbor independently.</p>
</div>
</div>
<div class="explanation-item">
<div class="explanation-number">3</div>
<div class="explanation-content">
<strong>Aggregate Predictions</strong>
<p>Combine all action candidates via permutation-invariant function: <code>â<sub>q</sub> = g<sub>ψ</sub>({a'<sub>1</sub>, ..., a'<sub>k</sub>})</code>. The final prediction is typically an average of individual predictions.</p>
</div>
</div>
</div>
</div>
<p class="demo-disclaimer">Note: This is a simplified simulation for illustration purposes and does not use an actual trained DARP model.</p>
</div>
<!-- Results Section -->
<div id="results" class="results-section">
<div class="tagline">Results</div>
<!-- Domain Selector -->
<div class="domain-selector">
<button class="domain-btn active" data-tab="robocasa">
<span class="icon is-small">
<i class="fas fa-home"></i>
</span>
<span>Robocasa</span>
</button>
<button class="domain-btn" data-tab="robosuite">
<span class="icon is-small">
<i class="fas fa-robot"></i>
</span>
<span>Robosuite</span>
</button>
<button class="domain-btn" data-tab="mujoco">
<span class="icon is-small">
<i class="fas fa-horse"></i>
</span>
<span>MuJoCo</span>
</button>
</div>
<!-- MuJoCo Content -->
<div id="mujoco-content" class="domain-content">
<div class="results-layout">
<!-- Carousel Side -->
<div class="carousel-side">
<div class="results-carousel" id="carousel-mujoco">
<!-- Slide 1: Hopper -->
<div class="carousel-item active">
<p class="slide-title">Task: Hopper</p>
<div class="comparison-grid">
<div class="comparison-video">
<span class="video-tag bc-tag">BC</span>
<video autoplay muted loop playsinline>
<source src="./static/videos/mujoco/hopper_bc_fail.mp4" type="video/mp4">
</video>
</div>
<div class="comparison-video">
<span class="video-tag darp-tag">DARP</span>
<video autoplay muted loop playsinline>
<source src="./static/videos/mujoco/hopper_darp_success.mp4" type="video/mp4">
</video>
</div>
</div>
</div>
<!-- Slide 2: Ant -->
<div class="carousel-item">
<p class="slide-title">Task: Ant</p>
<div class="comparison-grid">
<div class="comparison-video">
<span class="video-tag bc-tag">BC</span>
<video autoplay muted loop playsinline>
<source src="./static/videos/mujoco/ant_bc_fail.mp4" type="video/mp4">
</video>
</div>
<div class="comparison-video">
<span class="video-tag darp-tag">DARP</span>
<video autoplay muted loop playsinline>
<source src="./static/videos/mujoco/ant_darp_success.mp4" type="video/mp4">
</video>
</div>
</div>
</div>
<!-- Slide 3: Walker -->
<div class="carousel-item">
<p class="slide-title">Task: Walker</p>
<div class="comparison-grid">
<div class="comparison-video">
<span class="video-tag bc-tag">BC</span>
<video autoplay muted loop playsinline>
<source src="./static/videos/mujoco/walker_bc_fail.mp4" type="video/mp4">
</video>
</div>
<div class="comparison-video">
<span class="video-tag darp-tag">DARP</span>
<video autoplay muted loop playsinline>
<source src="./static/videos/mujoco/walker_darp_success.mp4" type="video/mp4">
</video>
</div>
</div>
</div>
<!-- Slide 4: HalfCheetah -->
<div class="carousel-item">
<p class="slide-title">Task: HalfCheetah</p>
<div class="comparison-grid">
<div class="comparison-video">
<span class="video-tag bc-tag">BC</span>
<video autoplay muted loop playsinline>
<source src="./static/videos/mujoco/halfcheetah_bc_fail.mp4" type="video/mp4">
</video>
</div>
<div class="comparison-video">
<span class="video-tag darp-tag">DARP</span>
<video autoplay muted loop playsinline>
<source src="./static/videos/mujoco/halfcheetah_darp_success.mp4" type="video/mp4">
</video>
</div>
</div>
</div>
</div>
<button class="carousel-nav prev" data-carousel="mujoco">‹</button>
<button class="carousel-nav next" data-carousel="mujoco">›</button>
</div>
<!-- Table Side -->
<div class="table-side">
<div class="results-box">
<h4 class="box-title">Performance (Return)</h4>
<table class="results-table">
<thead>
<tr>
<th>Task</th>
<th>BC</th>
<th>DARP (Ours)</th>
</tr>
</thead>
<tbody>
<tr>
<td>Hopper</td>
<td>2313.65 ± 203.75</td>
<td><strong>3545.57 ± 3.54</strong></td>
</tr>
<tr>
<td>Ant</td>
<td>2376.20 ± 339.43</td>
<td><strong>4383.28 ± 266.37</strong></td>
</tr>
<tr>
<td>Walker</td>
<td>2658.40 ± 274.08</td>
<td><strong>4894.01 ± 75.12</strong></td>
</tr>
<tr>
<td>HalfCheetah</td>
<td>1063.23 ± 371.08</td>
<td><strong>5515.41 ± 841.33</strong></td>
</tr>
</tbody>
</table>
<p class="table-note">Scores averaged across 100 trials with 95% confidence intervals.</p>
</div>
</div>
</div>
</div>
<!-- Robosuite Content -->
<div id="robosuite-content" class="domain-content">
<div class="results-layout">
<!-- Carousel Side -->
<div class="carousel-side">
<div class="results-carousel" id="carousel-robosuite">
<!-- Slide 1: Stack -->
<div class="carousel-item active">
<p class="slide-title">Task: Stack</p>
<div class="comparison-grid">
<div class="comparison-video">
<span class="video-tag bc-tag">BC</span>
<video autoplay muted loop playsinline>
<source src="./static/videos/robosuite/stack_bc_fail.mp4" type="video/mp4">
</video>
</div>
<div class="comparison-video">
<span class="video-tag darp-tag">DARP</span>
<video autoplay muted loop playsinline>
<source src="./static/videos/robosuite/stack_darp_success.mp4" type="video/mp4">
</video>
</div>
</div>
</div>
<!-- Slide 2: Threading -->
<div class="carousel-item">
<p class="slide-title">Task: Threading</p>
<div class="comparison-grid">
<div class="comparison-video">
<span class="video-tag bc-tag">BC</span>
<video autoplay muted loop playsinline>
<source src="./static/videos/robosuite/thread_bc_fail.mp4" type="video/mp4">
</video>
</div>
<div class="comparison-video">
<span class="video-tag darp-tag">DARP</span>
<video autoplay muted loop playsinline>
<source src="./static/videos/robosuite/thread_darp_success.mp4" type="video/mp4">
</video>
</div>
</div>
</div>
<!-- Slide 3: Peg Insertion -->
<div class="carousel-item">
<p class="slide-title">Task: Peg Insertion</p>
<div class="comparison-grid">
<div class="comparison-video">
<span class="video-tag bc-tag">BC</span>
<video autoplay muted loop playsinline>
<source src="./static/videos/robosuite/square_bc_fail.mp4" type="video/mp4">
</video>
</div>
<div class="comparison-video">
<span class="video-tag darp-tag">DARP</span>
<video autoplay muted loop playsinline>
<source src="./static/videos/robosuite/square_darp_success.mp4" type="video/mp4">
</video>
</div>
</div>
</div>
</div>
<button class="carousel-nav prev" data-carousel="robosuite">‹</button>
<button class="carousel-nav next" data-carousel="robosuite">›</button>
</div>
<!-- Table Side -->
<div class="table-side">
<div class="results-box">
<h4 class="box-title" style="margin-top: 2rem;">Success Rate (%) w/ Image Embeddings</h4>
<table class="results-table">
<thead>
<tr>
<th>Task</th>
<th>BC</th>
<th>DARP (Ours)</th>
</tr>
</thead>
<tbody>
<tr>
<td>Stack</td>
<td>44%</td>
<td><strong>75%</strong></td>
</tr>
<tr>
<td>Threading</td>
<td>38%</td>
<td><strong>76%</strong></td>
</tr>
<tr>
<td>Peg Insertion</td>
<td>17%</td>
<td><strong>52%</strong></td>
</tr>
</tbody>
</table>
<h4 class="box-title">Success Rate (%) w/ Low-Dimensional State</h4>
<table class="results-table">
<thead>
<tr>
<th>Task</th>
<th>BC</th>
<th>DARP (Ours)</th>
</tr>
</thead>
<tbody>
<tr>
<td>Stack</td>
<td>47%</td>
<td><strong>72%</strong></td>
</tr>
<tr>
<td>Threading</td>
<td>37%</td>
<td><strong>63%</strong></td>
</tr>
<tr>
<td>Peg Insertion</td>
<td>46%</td>
<td><strong>62%</strong></td>
</tr>
</tbody>
</table>
<p class="table-note">Success rate over 100 trials.</p>
</div>
</div>
</div>
</div>
<!-- Robocasa Content -->
<div id="robocasa-content" class="domain-content active">
<div class="results-layout">
<!-- Carousel Side -->
<div class="carousel-side">
<div class="results-carousel" id="carousel-robocasa">
<!-- Slide 1: Close Drawer -->
<div class="carousel-item active">
<p class="slide-title">Task: Close Drawer</p>
<div class="comparison-grid">
<div class="comparison-video">
<span class="video-tag bc-tag">BC</span>
<video autoplay muted loop playsinline>
<source src="./static/videos/robocasa/drawer_bc_fail.mp4" type="video/mp4">
</video>
</div>
<div class="comparison-video">
<span class="video-tag darp-tag">DARP</span>
<video autoplay muted loop playsinline>
<source src="./static/videos/robocasa/drawer_darp_success.mp4" type="video/mp4">
</video>
</div>
</div>
</div>
<!-- Slide 2: Close Door -->
<div class="carousel-item">
<p class="slide-title">Task: Close Door</p>
<div class="comparison-grid">
<div class="comparison-video">
<span class="video-tag bc-tag">BC</span>
<video autoplay muted loop playsinline>
<source src="./static/videos/robocasa/door_bc_fail.mp4" type="video/mp4">
</video>
</div>
<div class="comparison-video">
<span class="video-tag darp-tag">DARP</span>
<video autoplay muted loop playsinline>
<source src="./static/videos/robocasa/door_darp_success.mp4" type="video/mp4">
</video>
</div>
</div>
</div>
<!-- Slide 3: Turn on Stove -->
<div class="carousel-item">
<p class="slide-title">Task: Turn on Stove</p>
<div class="comparison-grid">
<div class="comparison-video">
<span class="video-tag bc-tag">BC</span>
<video autoplay muted loop playsinline>
<source src="./static/videos/robocasa/stove_bc_fail.mp4" type="video/mp4">
</video>
</div>
<div class="comparison-video">
<span class="video-tag darp-tag">DARP</span>
<video autoplay muted loop playsinline>
<source src="./static/videos/robocasa/stove_darp_success.mp4" type="video/mp4">
</video>
</div>
</div>
</div>
</div>
<button class="carousel-nav prev" data-carousel="robocasa">‹</button>
<button class="carousel-nav next" data-carousel="robocasa">›</button>
</div>
<!-- Table Side -->
<div class="table-side">
<div class="results-box">
<h4 class="box-title">Success Rate (%)</h4>
<table class="results-table">
<thead>
<tr>
<th>Task</th>
<th>BC</th>
<th>DARP (Ours)</th>
</tr>
</thead>
<tbody>
<tr>
<td>Close Drawer</td>
<td>54%</td>
<td><strong>85%</strong></td>
</tr>
<tr>
<td>Close Door</td>
<td>29%</td>
<td><strong>45%</strong></td>
</tr>
<tr>
<td>Turn on Stove</td>
<td>28%</td>
<td><strong>43%</strong></td>
</tr>
</tbody>
</table>
<p class="table-note">Success rate over 100 trials.</p>
</div>
</div>
</div>
</div>
</div>
<!-- Divergence Analysis Section - Insert this after the co-training section -->
<div id="divergence" class="divergence-section">
<div class="tagline">Divergence Analysis</div>
<p class="section-description">
Watch how BC and DARP behave when they drift out-of-distribution. While both agents' states become unlikely under the expert distribution,
DARP's difference vectors to nearest neighbors remain in-distribution, enabling stable recovery.
</p>
<div class="divergence-canvas-container">
<canvas id="divergenceCanvas"></canvas>
<div class="divergence-controls">
<button id="divergenceRestart" class="divergence-btn">
<svg width="16" height="16" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
<polyline points="1 4 1 10 7 10"></polyline>
<path d="M3.51 15a9 9 0 1 0 2.13-9.36L1 10"></path>
</svg>
Restart
</button>
<button id="divergencePlay" class="divergence-btn primary">
<svg width="16" height="16" viewBox="0 0 24 24" fill="currentColor">
<polygon points="5 3 19 12 5 21 5 3"></polygon>
</svg>
Play
</button>
</div>
<div class="divergence-legend">
<div class="legend-item">
<div class="legend-dot expert-dot"></div>
<span>Expert Demonstrations</span>
</div>
<div class="legend-item">
<div class="legend-dot bc-dot"></div>
<span>BC Agent (OOD State)</span>
</div>
<div class="legend-item">
<div class="legend-dot darp-dot"></div>
<span>DARP Agent (OOD State, ID Δ)</span>
</div>
</div>
</div>
<div class="divergence-metrics">
<div class="metric-card">
<div class="metric-label">BC State Likelihood</div>
<div class="metric-bar-container">
<div id="bcStateLikelihood" class="metric-bar bc-metric"></div>
</div>
<div class="metric-value" id="bcStateValue">100%</div>
</div>
<div class="metric-card">
<div class="metric-label">DARP State Likelihood</div>
<div class="metric-bar-container">
<div id="darpStateLikelihood" class="metric-bar darp-metric"></div>
</div>
<div class="metric-value" id="darpStateValue">100%</div>
</div>
<div class="metric-card highlight">
<div class="metric-label">DARP Δ Likelihood</div>
<div class="metric-bar-container">
<div id="darpDeltaLikelihood" class="metric-bar delta-metric"></div>
</div>
<div class="metric-value" id="darpDeltaValue">100%</div>
</div>
</div>
<div class="divergence-explanation">
<p><strong>Key Insight:</strong> As the animation progresses, BC (red) drifts away from the expert demonstrations and
its state likelihood drops significantly. DARP (green) experiences similar perturbations but <strong>recovers and stays near the expert manifold</strong>
because its difference vectors <strong>Δ = s* - s<sub>q</sub></strong> to nearest neighbors remain in-distribution (~90%).
These difference vectors are similar to the distances between expert demonstration states, enabling DARP to make reliable predictions and maintain stability.</p>
</div>
<p class="demo-disclaimer">Note: This is a simplified simulation for illustration purposes and does not use an actual trained DARP model. See section 3.4 in the paper for plots generated with trained DARP models.</p>
</div>
<!-- Multimodal Aggregation Section -->
<div id="multimodal" class="multimodal-section">
<div class="tagline">Handling Multimodal Action Distributions</div>
<div class="section-intro">
<p>
While simple averaging works well for unimodal action distributions, many robotic tasks
require representing multimodal behaviors. DARP can be extended to use more complex
permutation-invariant aggregation functions like Set Transformers or Deep Sets to handle
complex, multimodal action distributions.
</p>
</div>
<div class="multimodal-demo">
<div class="demo-canvas-container">
<canvas id="multimodalCanvas"></canvas>
</div>
<div class="demo-controls">
<button id="multimodalRestart" class="control-btn">
<svg width="16" height="16" viewBox="0 0 24 24" fill="currentColor">
<path d="M12 5V1L7 6l5 5V7c3.31 0 6 2.69 6 6s-2.69 6-6 6-6-2.69-6-6H4c0 4.42 3.58 8 8 8s8-3.58 8-8-3.58-8-8-8z"/>
</svg>
Reset
</button>
<button id="multimodalToggle" class="control-btn primary">
<svg width="16" height="16" viewBox="0 0 24 24" fill="currentColor">
<polygon points="5 3 19 12 5 21 5 3"></polygon>
</svg>
Play
</button>
</div>
<div class="aggregation-comparison">
<div class="comparison-item">
<div class="comparison-header">
<div class="method-badge averaging">Simple Averaging</div>
</div>
<div class="comparison-viz">
<div id="avgDistribution" class="distribution-viz"></div>
</div>
<div class="comparison-description">
<p>Averaging neighbor predictions collapses multimodal distributions into a single mode, which is oftentimes not useful.</p>
</div>
</div>
<div class="comparison-item">
<div class="comparison-header">
<div class="method-badge set-transformer">Set Transformer / DeepSets</div>
</div>
<div class="comparison-viz">
<div id="setDistribution" class="distribution-viz"></div>
</div>
<div class="comparison-description">
<p>More complex set-based aggregation preserves multimodality, enabling the model to represent multi-modal expert behaviors.</p>
</div>
</div>
</div>
</div>
</div>
</div>
<div id="diffusion" class="diffusion-section">
<div class="tagline">Compatibility with Diffusion Policy</div>
<div class="results-layout">
<!-- Carousel Side -->
<div class="carousel-side">
<p>
The previous results demonstrated DARP using a standard MLP backbone for the predictor function <em>f<sub>θ</sub></em>.
However, DARP's architecture is not limited to simple neural networks—it naturally extends to more sophisticated
policy classes.
<strong>DARP is fully compatible with diffusion policies</strong>, a state-of-the-art approach for visuomotor
control that has shown impressive results on complex manipulation tasks. By combining DARP's retrieval-based
reparameterization with diffusion's generative modeling, we achieve substantial improvements over
diffusion policy alone.
</p>
</div>
<!-- Table Side -->
<div class="table-side">
<div class="results-box">
<h4 class="box-title">Success Rate (%)</h4>
<table class="results-table">
<thead>
<tr>
<th>Task</th>
<th>Diffusion</th>
<th>Diffusion+DARP</th>
</tr>
</thead>
<tbody>
<tr>
<td>Stack</td>
<td>52%</td>
<td><strong>82%</strong></td>
</tr>
<tr>
<td>Threading</td>
<td>44%</td>
<td><strong>72%</strong></td>
</tr>
<tr>
<td>Peg Insertion</td>
<td>54%</td>
<td><strong>66%</strong></td>
</tr>
</tbody>
</table>
<p class="table-note">Success rate over 100 trials. Note that the amount of data used in these diffusion experiments is less than for the Robosuite results in the Results section to normalize scores.</p>
</div>
</div>
</div>
</div>
<div class="tagline" id="bibtex">BibTeX</div>
<div id="bibtex" class="bibtex-code">
<div class="bibtex-title">Reference</div>
<pre><code>
@inproceedings{pfeiferdifference,
title={Difference-Aware Retrieval Polices for Imitation Learning},
author={Pfeifer, Quinn and Pronovost, Ethan and Shah, Paarth and Khetarpal, Khimya and Gupta, Abhishek and Srinivasa, Siddhartha},
booktitle={The Fourteenth International Conference on Learning Representations}
year={2026}
}
</code></pre>
</div>
</div>
<div class="footer">
© DARP team. Source code is available <a href="https://github.com/WEIRDLabUW/darp-site"
target="_blank">here</a>. Modified from the <a href="https://polaris-evals.github.io/"
target="_blank">PolaRiS</a> website.
</div>
<script type="module" src="https://unpkg.com/@google/model-viewer/dist/model-viewer.min.js"></script>
<script src="index.js"></script>
<script src="hero-pipeline.js"></script>
<script src="divergence-section.js"></script>
<script src="architecture-section.js"></script>
<script src="multimodal-section.js"></script>
</body>
</html>