-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
842 lines (785 loc) · 52.8 KB
/
index.html
File metadata and controls
842 lines (785 loc) · 52.8 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
<!DOCTYPE html>
<html lang="en">
<head>
<title>Kelin Yu</title>
<meta charset="utf-8">
<meta http-equiv="X-UA-Compatible" content="IE=edge">
<meta name="viewport" content="width=device-width, initial-scale=1, shrink-to-fit=no">
<meta name="description" content="Kelin Yu is a Master student at Georgia Tech constructing learning tools for Robots.">
<meta name="keywords" content="Kelin Yu,Robotics,Computer Vision,Reinforcement Learning,Imitation Learning">
<link rel="stylesheet" href="https://stackpath.bootstrapcdn.com/bootstrap/4.1.1/css/bootstrap.min.css" integrity="sha384-WskhaSGFgHYWDcbwN70/dfYBj47jz9qbsMId/iRN3ewGhXQFZCSftd1LZCfmhktB" crossorigin="anonymous">
<style type="text/css">
body{margin:0;padding:0;background-color:#eee;}
a{color:#4e613e}
a:hover{color:#566c7f}
#background{position:fixed;top:0;overflow:hidden;height:0}
#wave-canvas{display:block;position:relative;top:0;left:5%;width:55%;min-width:850px;height:150px;border:0}
#main{z-index:2;margin:0 auto;position:absolute;top:5%;width:100%;min-width:240px;text-align:center;}
#main-content{background:rgba(238,238,238,0.95);padding:0;text-align:center;cursor:default;text-align:center;}
#main-content{margin-left:6.5104167%;width:86.9791667%;}
@media (min-width: 768px) {
#main-content{width:66%;min-width:668px;margin-left:50px}
}
@media (min-width: 1200px) {
#main-content{width:66%;max-width:900px;margin-left:100px}
}
/* Avatar, Name, etc. */
#main-content-container{margin:0 auto;padding:15px 0em 15px 0em;max-width:1080px;}
@media (min-width: 768px) {
#main-content-container{padding:15px 0 15px 2em;}
}
#main-content-container .col-avatar{text-align:center}
@media (min-width: 768px) {
#main-content-container .col-avatar{text-align:right}
}
#main-content-container .col-avatar .avatar{z-index:100;width:150px;height:150px;margin-top:0px;border-radius:100%;box-shadow:0 0 0 0.3em #fff;opacity:1;}
#main-content-container .col-info{height:100%;align-items:center;display:-webkit-flex;height:150px}
#main-info-container{padding-top:15px}
#main-info-container .col-name{text-align:center}
#main-info-container .col-name .textbox-info{display:inline-block;margin:0 auto;text-align:left}
#main-info-container .col-name .name{margin:0;color:#566c7f;font-size:1.85em;font-family:"Source Sans Pro",Helvetica,sans-serif}
#main-info-container .col-name .email{color:#666; font-size:14px}
#main-info-container .col-link{text-align:center;margin:-5px 0 -25px;}
#main-info-container .col-link ul {margin:0px auto auto;padding:0;list-style:none;text-align:center}
#main-info-container .col-link li {display:inline-block;margin:auto;padding:0 10px;color:#ccc;line-height:10px}
#main-info-container .col-link li a{width:45px;height:45px;color:#ccc;padding:0;font-size:45px;line-height:45px}
#main-info-container .col-link li a:focus,
#main-info-container .col-link li a:hover{text-decoration:none;cursor:hand}
#main-info-container .col-link li a.border{border:1px solid #ccc;border-radius:100%;}
#main-info-container .col-link li a.padding{padding:0px 0px;font-size:22px}
/* Other */
#main-more-container{background:#ffffff;width:100%;padding:0 30px}
#main-bio-container{margin:auto;padding: 0 0em;max-width:1080px;text-align:left;padding-top:30px;max-width:1080px;font-size:1rem;line-height:1.4}
#main-bio-container .subtitle{margin-top:30px;color:#566c7f}
#main-pub-container{margin:auto;padding: 0 0em;max-width:1080px;text-align:left;padding-top:30px;max-width:1080px}
#main-pub-container .subtitle{color:#566c7f;}
#main-pub-container .subtitle a {font-weight:400;font-size:18px}
#main-pub-container .subtitle a.activated {color:#333333}
#main-pub-container .subtitle a.activated:focus,
#main-pub-container .subtitle a.activated:hover {text-decoration:none;cursor:initial}
#main-pub-container .move-up {margin-top:-0.3cm}
#main-pub-container .move-up-more {margin-top:-0.6cm}
#main-pub-container .move-down {margin-top:0.8cm}
#main-pub-container .subtitle-aux {font-size:1rem}
#main-pub-container .subtitle-aux a.activated {color:#333333}
#main-pub-container .subtitle-aux a.activated:focus,
#main-pub-container .subtitle-aux a.activated:hover {text-decoration:none;cursor:initial}
#main-pub-container .subtitle-aux .note{margin-left:1em;color:#999999;text-decoration:none;font-size:14px}
#main-pub-card-container{}
#main-pub-card-container>h5{margin-bottom: 20px}
#main-pub-card-container .pub-card{width: 100%;padding-bottom:15px;font-size:1rem}
#main-pub-card-container .pub-card .col-l{text-align:center;padding-top:10px}
#main-pub-card-container .pub-card .col-r{padding-top:10px}
#main-pub-card-container .pub-card img{width:100%;margin:0 auto}
#main-pub-card-container .pub-card .pub-card-body{width:100%;background:transparent;padding-left:0px}
#main-pub-card-container .pub-card .title{font-weight:600;font-size:1rem;margin-bottom:4px;line-height:1.4}
#main-pub-card-container .pub-card .authors{font-weight:400;color:#666666;margin-bottom:4px;line-height:1.4}
#main-pub-card-container .pub-card .authors a{color:#666666}
#main-pub-card-container .pub-card .authors u{color:#333333;font-weight:600;text-decoration:none}
#main-pub-card-container .pub-card .conference{color:#333333;font-weight:600}
#main-pub-card-container .pub-card .note{display:block;color:#999999;text-decoration:none;font-size:12px}
#main-pub-card-container .pub-card .info{margin-bottom:0;}
#main-pub-card-container .row{margin:0}
#main-pub-card-container .col-l,
#main-pub-card-container .col-r{vertical-align:top;display:inline-block;margin:0;padding-left:0px}
/* News */
#news-container{margin:auto;padding:0 0em;max-width:1080px;text-align:left;padding-top:30px}
#news-container .subtitle{color:#566c7f}
.news-scroller{max-height:180px;overflow-y:auto;background:#fff;border:1px solid #e6e6e6;border-radius:10px;padding:10px 14px}
.news-list{list-style:none;margin:0;padding:0}
.news-item{padding:8px 2px;border-bottom:1px dashed #eee;line-height:1.4}
.news-item:last-child{border-bottom:none}
.news-item .date{font-weight:600;color:#566c7f;margin-right:.6em;white-space:nowrap}
.news-item a{text-decoration:none}
.news-item a:hover{text-decoration:underline}
/* Mentoring & Collaborators */
#mentoring-collab-container{margin:auto;padding:0 0em;max-width:1080px;text-align:left;padding-top:30px}
.people-list{list-style:disc;margin:0;padding-left:1.2em}
.people-list li{margin:4px 0;line-height:1.4}
#footer{padding:1.5em 0 1.5em;min-width:240px;background:#f4f4f4;opacity:0.95;font-size:14px;line-height:12px;text-align:left}
#footer .container{max-width:540px;padding:0 45px}
#footer .row{align-items:center}
#footer .stat{width:90px;overflow:hidden;margin:0 auto;padding:0 15px;}
#footer .copyright{flex-grow:1;margin:5px auto 5px;text-align:left;display:block;font-size:14px;line-height:1.4}
#footer a{color:#666666}
#footer a:hover{color:#333333}
.bold{font-weight:600}
.highlight{color:#BB2222;font-weight:600}
.highlight:hover{color:#BB2222;font-weight:600}
.hide{display:none}
/* NEW: Representative works highlighting */
.pub-card.rep{background:#fff8cc;border-radius:10px;padding:10px}
.rep-badge{display:inline-block;font-size:12px;font-weight:600;margin-left:6px;color:#7a6a00}
.legend{font-size:0.95rem;color:#6b6b6b;margin-top:6px}
</style>
</head>
<body>
<div id="main">
<div id="main-content">
<div id="main-content-container" class="container">
<div class="row">
<div class="col-sm-12 col-md-3 col-avatar">
<img src="images/kelin.png" class="avatar" />
</div>
<div class="col-sm-12 col-md-9 col-info">
<div id="main-info-container" class="container">
<div class="row">
<div class="col-md-12 col-xl-6 col-name">
<div class="textbox-info">
<h1 class="name"><span>Kelin Yu</span></h1>
<p class="email">Email: kyu85 [at] umd [dot] edu <br>
<a href="cv.pdf">CV</a>
<a href="https://scholar.google.com/citations?user=zVdZJRwAAAAJ&hl=en">Google Scholar</a>
<a href="https://twitter.com/ColinYu14116982">Twitter/X</a>
<a href="https://github.com/ColinYu1">Github</a>
</div>
</div>
</div>
</div>
</div>
</div>
</div>
<div id="main-more-container">
<div id="main-bio-container">
<p>
I am a second-year Ph.D. student in Computer Science at University of Maryland, where I work with
Prof. <a href="https://ruohangao.github.io/" target="_blank">Ruohan Gao</a>. I also closely work with Prof. <a href="https://prg.cs.umd.edu/" target="_blank">Yiannis Aloimonos</a> and Prof. <a href="https://tokekar.com/" target="_blank">Pratap Tokekar</a>.
I am interested in MultiSensory Robot Learning, Learning from Cross-embodiment Human Videos, and Robot Manipulation. <br /><br />
I received my master's degree in Computer Science at <a href="https://www.cc.gatech.edu/">Georgia Tech</a> in May 2024, advised by
Ph.D. mentor <a href="https://y8han.github.io/" target="_blank">Yunhai Han</a>,
Prof. <a href="https://www.me.gatech.edu/faculty/zhao" target="_blank">Ye Zhao</a> and
Prof. <a href="https://faculty.cc.gatech.edu/~danfei/" target="_blank">Danfei Xu</a>.
Previously, I obtained dual bachelor's degree in Electrical Engineering and Mathematics from Georgia Tech in 2022 and
was a Robotics Software Engineering intern at Amazon. <br /><br />
I'm open to discussions and collaborations, so feel free to drop me an email if you are interested.
</p>
</div>
<div id="news-container">
<h5 class="subtitle">News</h5>
<div class="news-scroller" role="region" aria-label="Latest updates" tabindex="0">
<ul class="news-list">
<li class="news-item">
<span class="date">2026/3</span>
Selected as one of the <a href="https://www.qualcomm.com/research/university-relations/innovation-fellowship/2026-north-america" target="_blank"><span class="highlight">Qualcomm Innovation Fellowship Finalist</span></a> with <a href="https://y8han.github.io/" target="_blank">Yunhai Han</a></span>
</li>
<li class="news-item">
<span class="date">2026/2</span>
LG-SAIL is accepted by TMLR with <span class="highlight">J2C certification</span>, will be presented at NeurIPS 2026</span>
</li>
<li class="news-item">
<span class="date">2026/2</span>
I am going to join Amazon Personal Robotics Group as a Research Intern in June 2026</span>
</li>
<li class="news-item">
<span class="date">2025/10</span>
Invited talk at Shanghai Jiaotong University
</li>
<li class="news-item">
<span class="date">2025/9</span>
Control is accepted by ICCV CDEL Workshop with
<span class="highlight">Oral Presentation</span> —
<a href="https://colinyu1.github.io/controltac/" target="_blank">link</a>
</li>
<li class="news-item">
<span class="date">2025/6</span>
GenFlowRL is accepted by ICCV 2025 —
<a href="https://colinyu1.github.io/genflowrl/" target="_blank">link</a>
</li>
<li class="news-item">
<span class="date">2025/5</span>
GenFlowRL is presented at ICRA FMNS Workshop with
<span class="highlight">Spotlight and Best Paper Nomination</span> —
<a href="https://colinyu1.github.io/genflowrl/" target="_blank">link</a>
</li>
<li class="news-item">
<span class="date">2025/4</span>
Invited talk at UMD Robotics Symposium —
<a href="https://robotics.umd.edu/symposium" target="_blank">link</a>
</li>
<li class="news-item">
<span class="date">2024/8</span>
Graduated from Georgia Tech and started my Ph.D. at UMD with Prof. Ruohan Gao
</li>
<li class="news-item">
<span class="date">2024/8</span>
MimicTouch accepted by CoRL 2024 —
<a href="https://arxiv.org/abs/2310.16917" target="_blank">link</a>
</li>
<li class="news-item">
<span class="date">2024/4</span>
One paper published at IEEE T-Mech —
<a href="https://ieeexplore.ieee.org/document/10552075" target="_blank">link</a>
</li>
<li class="news-item">
<span class="date">2023/12</span>
MimicTouch won the <span class="highlight">Best Paper</span> at NeurIPS TouchProcessing Workshop —
<a href="https://sites.google.com/view/mimictouch/" target="_blank">link</a>
</li>
<li class="news-item">
<span class="date">2023/12</span>
Invited oral talk at NeurIPS TouchProcessing Workshop —
<a href="https://sites.google.com/view/mimictouch/" target="_blank">link</a>
</li>
</ul>
</div>
</div>
<div id="mentoring-collab-container">
<h5 class="subtitle">Mentored Students</h5>
<ul class="people-list">
<li><a href="https://dongyuluo.github.io/" target="_blank">Dongyu Luo</a> — Undergraduate, HKU, 2025–</li>
<li><a href="https://haodezhang.github.io/" target="_blank">Haode Zhang</a> — Undergraduate, SJTU, 2025–</li>
</ul>
<h5 class="subtitle move-up">External Collaborators (Past & Present)</h5>
<ul class="people-list">
<li><a href="https://y8han.github.io/" target="_blank">Yunhai Han</a> — GT Robotics Ph.D. in Prof. Harish Ravichandar's Lab</li>
<li><a href="https://sheng-eatamath.github.io/" target="_blank">Sheng Zhang</a> — UMD CS Ph.D. in Prof. Heng Huang's Lab</li>
<li><a href="https://amirshahid.github.io/" target="_blank">Amir-Hossein Shahidzadeh</a> — UMD CS Ph.D. in Prof. Yiannis Aloimonos's Lab</li>
<li><a href="https://sites.google.com/view/shuocheng" target="_blank">Shuo Cheng</a> — GT CS Ph.D. in Prof. Danfei Xu's Lab</li>
<li>Qixian Wang — UIUC ME Ph.D. in Prof. Mattia Gazzola's Lab</li>
<li><a href="https://sites.google.com/view/vaibhavsaxena" target="_blank">Vaibhav Saxena</a> — GT CS Ph.D. in Prof. Danfei Xu's Lab</li>
</ul>
</div>
<div id="main-pub-container">
<h5 class="subtitle">Publications</h5>
<p class="legend">🟨 Highlighted entries are <b>representative works</b> </p>
<p class="subtitle-aux"><span class="note">(* indicates equal contribution)</span></p>
<div id="main-pub-card-container" class="hide">
<!-- NEW: Afford2Act entry (representative) -->
<!-- OmniTacTune (first; representative) -->
<div class="pub-card rep" data-year="2026" data-selected="true">
<div class="row">
<div class="col-l col-xs-12 col-lg-3">
<img src="images/omnitact.jpg" onerror="this.style.display='none'" width="100%"/>
</div>
<div class="col-r col-xs-12 col-lg-9">
<div class="pub-card-body">
<h5 class="title">OmniTacTune: Policy-Agnostic Real-World RL for Tactile Residual Adaptation of Visual Policies </h5>
<h6 class="authors">
<u>Kelin Yu*</u>,
<a href="https://haodezhang.github.io/">Haode Zhang*</a>,
<a href="https://harishravichandar.com/">Harish Ravichandar</a>,
<a href="https://y8han.github.io/">Yunhai Han</a>,
<a href="https://ruohangao.github.io/">Ruohan Gao</a>,
<a> (Haode is my undergraduate mentee)</a>
</h6>
<p class="info">
<span class="conference">In Submission</span> <br>
</p>
OmniTacTune uses a two-stage RL design that first warm-starts tactile-aware learning from autonomous base-policy rollouts, and then learns a lightweight tactile residual policy through online interaction. This design requires no offline tactile demonstrations or base-policy finetuning, making tactile adaptation efficient and low-cost. OmniTacTune generalizes across diverse contact-rich tasks, visual base policies from both human videos and teleoperated demos, and tactile representations. Across four contact-rich real-world tasks, it improves weak visual base policies from 5-40% success to 85-100% within 40-80 minutes, demonstrating an efficient way to adapt tactile feedback to scalable visual robot policies.
</div>
</div>
</div>
</div>
<!-- ControlTac (co-first; representative) -->
<div class="pub-card rep" data-year="2024" data-selected="true">
<div class="row">
<div class="col-l col-xs-12 col-lg-3">
<img src="images/teaser_0.png" width="100%"/>
</div>
<div class="col-r col-xs-12 col-lg-9">
<div class="pub-card-body">
<h5 class="title">ControlTac: Force- and Position-Controlled Tactile Data Augmentation with a Single Reference Image </h5>
<h6 class="authors">
<a href="https://dongyuluo.github.io/">Dongyu Luo*</a>,
<u>Kelin Yu*</u>,
<a href="https://amirshahid.github.io/">Amir-Hossein Shahidzadeh</a>,
<a href="https://users.umiacs.umd.edu/~fermulcm/">Cornelia Fermuler</a>,
<a href="https://robotics.umd.edu/clark/faculty/350/Yiannis-Aloimonos">Yiannis Aloimonos</a>,
<a href="https://ruohangao.github.io/">Ruohan Gao</a>,
<a> (Dongyu is my undergraduate mentee)</a>
</h6>
<p class="info">
<span class="conference">ICCV CDEL Workshop 2025 (<font color="red"> Oral </font>)</span> <br>
<a href="https://arxiv.org/abs/2510.01433">Paper</a> /
<a href="https://dongyuluo.github.io/controltac/">Website</a> /
<span class="conference">In Submission</span>
</p>
Vision-based tactile sensing has been widely used in perception, reconstruction, and robotic manipulation. However, collecting large-scale tactile data remains costly due to the localized nature of sensor-object interactions and inconsistencies across sensor instances. Existing approaches to scaling tactile data, such as simulation and free-form tactile generation, often suffer from unrealistic output and poor transferability to downstream tasks. To address this, we propose ControlTac, a two-stage controllable framework that generates realistic tactile images conditioned on a single reference tactile image, contact force, and contact position.
</div>
</div>
</div>
</div>
<!-- HumanEgo -->
<div class="pub-card" data-year="2026" data-selected="true">
<div class="row">
<div class="col-l col-xs-12 col-lg-3">
<img src="images/humanego.jpg" onerror="this.style.display='none'" width="100%"/>
</div>
<div class="col-r col-xs-12 col-lg-9">
<div class="pub-card-body">
<h5 class="title">HumanEgo: Zero-Shot Robot Learning from Minutes of Human Egocentric Videos </h5>
<h6 class="authors">
<a href="https://tx-leo.github.io/">Zhi Wang</a>,
<a href="https://bottle101.github.io/">Botao He</a>,
<u>Kelin Yu</u>,
<a href="https://sjlee.cc/">Seungjae Lee</a>,
<a href="https://ruohangao.github.io/">Ruohan Gao</a>,
<a href="https://furong-huang.com/">Furong Huang</a>,
<a href="https://robotics.umd.edu/clark/faculty/350/Yiannis-Aloimonos">Yiannis Aloimonos</a>
</h6>
<p class="info">
<a href="https://arxiv.org/abs/2605.24934">Paper</a> /
<a href="https://humanego-ai.github.io/">Website</a> /
<a href="https://github.com/HumanEgo-ai/HumanEgo">Code</a> /
<a href="https://youtu.be/0BMRwF9HACQ">Video</a> /
<span class="conference">In Submission</span>
</p>
HumanEgo learns robot manipulation policies from minutes of raw human egocentric videos by lifting demonstrations into entity-level hand-object interaction representations and training a flow matching policy with dense auxiliary objectives. The framework is robot-data-free, hardware-agnostic, data-efficient, and zero-shot transferable across different robot setups.
</div>
</div>
</div>
</div>
<!-- OdorSpaces -->
<div class="pub-card" data-year="2026" data-selected="true">
<div class="row">
<div class="col-l col-xs-12 col-lg-3">
<img src="images/fire.jpg" onerror="this.style.display='none'" width="100%"/>
</div>
<div class="col-r col-xs-12 col-lg-9">
<div class="pub-card-body">
<h5 class="title">OdorSpaces: Visual-Olfactory Embodied Navigation in 3D Environments </h5>
<h6 class="authors">
<a href="https://changyeon2.github.io/">Changyeon Lee</a>,
<u>Kelin Yu</u>,
<a>Zheng Wei</a>,
<a>Shuna Ni</a>,
<a href="https://ruohangao.github.io/">Ruohan Gao</a>
</h6>
<p class="info">
<span class="conference">In Submission</span> <br>
</p>
We introduce OdorSpaces, a 3D visual-olfactory navigation framework that augments Habitat-Sim with physics-based odor dynamics simulated by FDS (Fire Dynamics Simulator). OdorSpaces defines two physically distinct navigation scenarios: a fire scenario, where buoyant carbon dioxide (CO<sub>2</sub>) rises and forms upper-region odor fields, and a gas-leak scenario, where dense propane gas (C<sub>3</sub>H<sub>8</sub>) settles and spreads near the floor. We train multimodal reinforcement learning agents that use visual observations and olfactory measurements, and show that they outperform chemotaxis, biomimetic plume tracking, and history-based odor-gradient estimation baselines across both scenarios.
</div>
</div>
</div>
</div>
<div class="pub-card" data-year="2025" data-selected="true">
<div class="row">
<div class="col-l col-xs-12 col-lg-3">
<img src="images/afford2act.png" onerror="this.style.display='none'" width="100%"/>
</div>
<div class="col-r col-xs-12 col-lg-9">
<div class="pub-card-body">
<h5 class="title">Afford2Act: Affordance-Guided Automatic Keypoint Selection for Generalizable and Lightweight Robotic Manipulation </h5>
<h6 class="authors">
<a href="https://anukritisinghh.github.io/">Anukriti Singh</a>,
<a>Kasra Torshizi</a>,
<a>Khuzema Habib</a>,
<u>Kelin Yu</u>,
<a href="https://ruohangao.github.io/">Ruohan Gao</a>,
<a href="https://tokekar.com/">Pratap Tokekar</a>
</h6>
<p class="info">
<a href="https://arxiv.org/abs/2505.20498">Paper</a> /
<a href="https://afford2act.github.io/">Website</a> /
<span class="conference">In Submision</span> <br>
</p>
AFFORD2ACT, an affordance-guided framework that distills a minimal set of semantic 2D keypoints from a text prompt and a single image. AFFORD2ACT follows a three-stage pipeline: affordance filtering, category-level keypoint construction, and transformer-based policy learning with embedded gating to reason about the most relevant keypoints, yielding a compact 38-dimensional state policy that can be trained in 15 minutes, which performs well in real-time without proprioception or dense representations.
</div>
</div>
</div>
</div>
<!-- Continual Robot Learning (MOVED to 3rd; UPDATED to TMLR 2026; workshops removed; highlighted) -->
<div class="pub-card rep" data-year="2024" data-selected="true">
<div class="row">
<div class="col-l col-xs-12 col-lg-3">
<img src="images/lgsail.jpg" width="100%"/>
</div>
<div class="col-r col-xs-12 col-lg-9">
<div class="pub-card-body">
<h5 class="title">Continual Robot Learning via Language-Guided Skill Acquisition </h5>
<h6 class="authors">
<a>Zhaoyi Li*</a>,
<u>Kelin Yu*</u>,
<a href="https://sites.google.com/view/shuocheng">Shuo Cheng*</a>,
<a href="https://faculty.cc.gatech.edu/~danfei/">Danfei Xu</a>
</h6>
<p class="info">
<span class="conference">TMLR 2026 (<font color="red"> J2C Certification, Top 6.3%</font>), present at NeurIPS 2026</span></span> <br>
<a href="https://openreview.net/forum?id=oYRNxxGN9u&referrer=%5BAuthor%20Console%5D(%2Fgroup%3Fid%3DTMLR%2FAuthors%23your-submissions)"> Openreview</a> /
<a href="https://sites.google.com/view/continuallearning">Website</a>
</p>
We developed LG-SAIL (Language Models Guided Sequential, Adaptive, and Incremental Robot Learning), a framework that leverages Large Language Models (LLMs) to harmoniously integrate TAMP and DRL for continuous skill learning in long-horizon tasks. Our framework achieves automatic task decomposition, operator creation, and dense reward generation for efficiently acquiring the desired skills.
</div>
</div>
</div>
</div>
<!-- GenFlowRL (first; representative) -->
<div class="pub-card rep" data-year="2024" data-selected="true">
<div class="row">
<div class="col-l col-xs-12 col-lg-3">
<img src="images/main.png" width="100%"/>
</div>
<div class="col-r col-xs-12 col-lg-9">
<div class="pub-card-body">
<h5 class="title">GenFlowRL: Shaping Rewards with Generative Object-Centric Flow in Visual Reinforcement Learning </h5>
<h6 class="authors">
<u>Kelin Yu*</u>,
<a href="https://sheng-eatamath.github.io/">Sheng Zhang*</a>,
<a>Harshit Soora</a>,
<a href="https://furong-huang.com/">Furong Huang</a>,
<a href="https://www.cs.umd.edu/~heng/">Heng Huang</a>,
<a href="https://tokekar.com/">Pratap Tokekar</a>,
<a href="https://ruohangao.github.io/">Ruohan Gao</a>
</h6>
<p class="info">
<span class="conference">ICCV 2025</span> <br>
<span class="conference">ICRA FMNS Workshop 2025 (<font color="red"> Spotlight, Best Paper Nomination </font>)</span> <br>
<a href="https://arxiv.org/abs/2508.11049">Paper</a> /
<a href="https://colinyu1.github.io/genflowrl/">Website</a> /
<a href="https://youtu.be/ohhBsjapZpE">Video</a>
</p>
Recent advances have demonstrated the potential of video generation models to guide robot learning by deriving effective robot actions through inverse dynamics. However, these methods heavily depend on the quality of generated data and struggle with fine-grained manipulation due to the lack of environment feedback. While video-based reinforcement learning improves policy robustness, it remains constrained by the artifacts of generated video and the challenges of collecting large-scale in-domain robot datasets for training diffusion models. Motivated by the above, we propose GenFlowRL, which derives shaped rewards from generated flow trained from cross-embodiment datasets.
</div>
</div>
</div>
</div>
<!-- MimicTouch (first; representative) -->
<div class="pub-card rep" data-year="2024" data-selected="true">
<div class="row">
<div class="col-l col-xs-12 col-lg-3">
<img src="data/projects/mimictouch.png" width="100%"/>
</div>
<div class="col-r col-xs-12 col-lg-9">
<div class="pub-card-body">
<h5 class="title">MimicTouch: Leveraging Multi-modal Human Tactile Demonstrations for Contact-rich Manipulation </h5>
<h6 class="authors">
<u>Kelin Yu*</u>,
<a href="https://y8han.github.io/">Yunhai Han*</a>,
<a>Qixian Wang</a>,
<a href="https://sites.google.com/view/vaibhavsaxena">Vaibhav Saxena</a>,
<a href="https://faculty.cc.gatech.edu/~danfei/">Danfei Xu</a>,
<a href="https://sites.google.com/site/yezhaout">Ye Zhao</a>
</h6>
<p class="info">
<span class="conference">CoRL 2024</span> <br>
<span class="conference">NeurIPS Touch Processing Workshop 2023 (<font color="red"> Best Paper Award </font>)</span> <br>
<span class="conference">CoRL Deployable Workshop 2023</span> <br>
<a href="https://arxiv.org/abs/2310.16917">Paper</a> /
<a href="https://sites.google.com/view/mimictouch/%E9%A6%96%E9%A1%B5">Website</a> /
<a href="https://youtu.be/RWHoZOUtQvg?si=OXFG8drB90v7rZ8L">Video</a>
</p>
We introduce MimicTouch, a novel framework for learning policies directly from demonstrations provided by human users with their hands. The key innovations are i) a human tactile data collection system which collects multi-modal tactile dataset for learning human's tactile-guided control strategy, ii) an imitation learning-based framework for learning human's tactile-guided control strategy through such data, and iii) an online residual RL framework to bridge the embodiment gap between the human hand and the robot gripper. Through comprehensive experiments, we highlight the efficacy of utilizing human's tactile-guided control strategy to resolve contact-rich manipulation tasks.
</div>
</div>
</div>
</div>
<!-- Vision-Tactile Transformer (co-first; representative) -->
<div class="pub-card rep" data-year="2024" data-selected="true">
<div class="row">
<div class="col-l col-xs-12 col-lg-3">
<img src="data/projects/transformer.png" width="100%"/>
</div>
<div class="col-r col-xs-12 col-lg-9">
<div class="pub-card-body">
<h5 class="title">Learning Generalizable Vision-Tactile Robotic Grasping Strategy for Deformable Objects via Transformer </h5>
<h6 class="authors">
<a href="https://y8han.github.io/">Yunhai Han*</a>,
<u>Kelin Yu*</u>,
<a>Rahul Batra</a>,
<a>Nathan Boyd</a>,
<a>Chaitanya Mehta</a>,
<a href="https://www.isye.gatech.edu/users/tuo-zhao">Tuo Zhao</a>,
<a href="https://www.purduemars.com/">Yu She</a>,
<a href="https://faculty.cc.gatech.edu/~seth/">Seth Hutchinson</a>,
<a href="https://sites.google.com/site/yezhaout">Ye Zhao</a>
</h6>
<p class="info">
<span class="conference">Transactions on Mechatronics 2024 (Long Version)</span> <br>
<span class="conference">AIM 2023 (Short Version)</span> <br>
<a href="https://ieeexplore.ieee.org/document/10552075">Paper</a> /
<a href="https://github.com/GTLIDAR/DeformableObjectsGrasping/tree/master/src/grasping_framework">Code</a>
</p>
Reliable robotic grasping with deformable objects remains a challenging task due to underactuated contact interactions with a gripper, unknown object dynamics, and variable object geometries. In this study, we propose a Transformer-based robot grasping framework for rigid grippers that leverage tactile information from a GelSight sensor for safe object grasping. The Transformer network learns physical feature embeddings from visual & tactile feedback and predict a final grasp through a multilayer perceptron (MLP) with grasping strength. Using these predictions, the gripper is commanded with an optimal force for safe grasping tasks.
</div>
</div>
</div>
</div>
<div class="pub-card" data-year="2023" data-selected="true">
<div class="row">
<div class="col-l col-xs-12 col-lg-3">
<img src="data/projects/drl.png" width="100%"/>
</div>
<div class="col-r col-xs-12 col-lg-9">
<div class="pub-card-body">
<h5 class="title">Evolutionary Curriculum Training for DRL-Based Navigation Systems</h5>
<h6 class="authors">
<u>Kelin Yu*</u>,
<a>Max Asselmeier*</a>,
<a>Zhaoyi Li*</a>,
<a href="https://faculty.cc.gatech.edu/~danfei/">Danfei Xu</a>
</h6>
<p class="info">
<span class="conference">RSS MultiAct Workshop 2023</span> /
<a href="https://sites.google.com/view/hierarchical-navigation">Website</a> /
<a href="https://arxiv.org/abs/2306.08870">Paper</a> /
</p>
We introduce a novel approach called evolutionary curriculum training to tackle these challenges. The primary goal of evolutionary curriculum training is to evaluate the collision avoidance model's competency in various scenarios and create curricula to enhance its insufficient skills. The paper introduces an innovative evaluation technique to assess the DRL model's performance in navigating structured maps and avoiding dynamic obstacles. Additionally, an evolutionary training environment generates all the curriculum to improve the DRL model's inadequate skills tested in the previous evaluation. We benchmark the performance of our model across five structured environments to validate the hypothesis that this evolutionary training environment leads to a higher success rate and a lower average number of collisions.
<br>
</div>
</div>
</div>
</div>
<div class="pub-card" data-year="2023" data-selected="true">
<div class="row">
<div class="col-l col-xs-12 col-lg-3">
<img src="data/projects/nlp.png" width="100%"/>
</div>
<div class="col-r col-xs-12 col-lg-9">
<div class="pub-card-body">
<h5 class="title">Temporal Video-Language Alignment Network for Reward Shaping in Reinforcement Learning</h5>
<h6 class="authors">
<u>Kelin Yu*</u>,
<a>Ziyuan Cao*</a>,
<a>Reshma Ramachandra*</a>
</h6>
<p class="info">
<span class="conference">Technical Report 2022</span> /
<a href="https://arxiv.org/abs/2302.03954">Paper</a> /
</p>
Designing appropriate reward functions for Reinforcement Learning (RL) approaches has been a significant problem. Utilizing natural language instructions to provide intermediate rewards to RL agents in a process known as reward shaping can help the agent in reaching the goal state faster. In this work, we propose a natural language-based reward-shaping approach that maps trajectories from Montezuma's Revenge game environment to corresponding natural language instructions using an extension of the LanguagE-Action Reward Network (LEARN) framework. These trajectory-language mappings are further used to generate intermediate rewards which are integrated into reward functions that can be utilized to learn an optimal policy for any standard RL algorithms.
<br>
</div>
</div>
</div>
</div>
<h5 class="subtitle">Work Experience<div data-year="2023" data-selected="true">
<div class="pub-card" data-year="2023" data-selected="true">
<div class="row">
<div class="col-l col-xs-12 col-lg-3">
<img src="data/projects/Amazon.png" width="100%"/>
</div>
<div class="col-r col-xs-12 col-lg-9">
<div class="pub-card-body">
<h5 class="title">Robotics Software Engineering intern
<p class="info">
<span class="conference">Amazon Robotics AI</span> /
</p>
Designed and built Calibration Drift Detector for our Industrial manipulator with Python, Open3D, OpenCV, and machine learning classifier.
Each multi-pick detected as a single pick costs 12$, and MEP/DEP package costs $0.05. My system saves potential thousands of dollars every day.
Used AWS tool (S3) to get past images and point clouds and implemented advanced computer vision algorithms and applied ML classifier to detect calibration drift.
</div>
</div>
</div>
</div>
</div>
<h5 class="subtitle">Projects<div data-year="2023" data-selected="true">
<div class="pub-card" data-year="2023" data-selected="true">
<div class="row">
<div class="col-l col-xs-12 col-lg-3">
<img src="data/projects/parkinglot.png" width="100%"/>
</div>
<div class="col-r col-xs-12 col-lg-9">
<div class="pub-card-body">
<h5 class="title">iValet An Intelligent Parking Lot Management System and Interface</h5>
<h6 class="authors">
<u>Kelin Yu*</u>,
<a>Faiza Yousuf*</a>,
<a>Wei Xiong Toh*</a>,
<a>Yunchu Feng*</a>,
</h6>
<p class="info">
<span class="conference">Senior Design 2022</span> /
<a href="https://eceseniordesign2022spring.ece.gatech.edu/sd22p37/">Website</a> /
</p>
The iValet intelligent parking lot management system automatically directs drivers to the nearest vacant parking spot upon entering a crowded parking lot. The system consists of a camera, machine learning development board, a PostgreSQL server, and a user interface (web application). The camera is used to take photos of the entire parking lot, the development board runs segmentation and classification algorithms on those photos, the SQL server contains data about each parking spot that is written to by the image processing models, a path-planning algorithm, and the UI, while the web application shows end users the directions to the empty parking spots based on the location of the parking lot entrance. <br>
</div>
</div>
</div>
</div>
<h5 class="subtitle">Awards<div data-year="2023" data-selected="true">
<div class="pub-card" data-year="2023" data-selected="true">
<div class="row">
<a>Qualcomm Innovation Fellowship Finalist 2026</a>,
</div>
<div class="row">
<a>ICRA FMNS Workshop Best Paper Nomination 2025</a>,
</div>
<div class="row">
<a>Dean’s Fellowship, University of Maryland, College Park, 2024-2026</a>,
</div>
<div class="row">
<a>NeurIPS TouchProcessing Workshop Best Paper Award 2023</a>,
</div>
<div class="row">
<a>CoRL Travel Grant 2023</a>,
</div>
<div class="row">
<a>Graduate with Highest Honor at Georgia Tech 2022</a>,
</div>
</div>
</div>
<h5 class="subtitle">Professional Service<div data-year="2023" data-selected="true">
<div class="pub-card" data-year="2023" data-selected="true">
<div class="row">
<a>ICLR 2025, 2026</a>,
Reviewer
</div>
<div class="row">
<a>RA-L</a>,
Reviewer
</div>
<div class="row">
<a>ICCV 2025</a>,
Reviewer
</div>
<div class="row">
<a>NeurIPS 2024, 2025</a>,
Reviewer
</div>
<div class="row">
<a>RSS 2025</a>,
Reviewer
</div>
<div class="row">
<a>ICML 2025</a>,
Reviewer
</div>
<div class="row">
<a>ICRA 2025</a>,
Reviewer
</div>
<div class="row">
<a>CoRL Deployable Workshop, 2023</a>,
Reviewer
</div>
</div>
</div>
<h5 class="subtitle">Teaching Experience<div data-year="2023" data-selected="true">
<div class="pub-card" data-year="2023" data-selected="true">
<div class="row">
<a>CMSC 848M: Multimodal Computer Vision, University of Maryland</a>,
Spring 2025
<a>CMSC 421, Introduction to Artificial Intelligence, University of Maryland</a>,
Fall 2024
<a>CS 4476/6476, Computer Vision, Georgia Tech</a>,
Spring 2024, Fall 2023, Spring 2023, Fall 2022
<a>ECE 3741, Instrumentation and Electronics Laboratory, Georgia Tech</a>,
Spring 2021
</div>
</div>
</div>
</div>
</div>
</tbody></table>
<table class="sub-table" style="width: 200px;height: 100px;" align="center">
<script type="text/javascript" id="clustrmaps" src="//clustrmaps.com/map_v2.js?d=BejUQM27W93Et5QRR_mUdcbiOv65k0Dn7Qn8D6teTqw&cl=ffffff&w=a"></script>
</table>
</tbody>
</table>
<div id="footer">
<div class="container">
<div class="row">
<div class="copyright"; align="center">
Website template from <a href="https://yilundu.github.io/">Yilun Du</a>
</div>
</div>
</div>
</div>
</div>
<!-- jQuery first, then Tether, then Bootstrap JS. -->
<script src="https://code.jquery.com/jquery-3.1.1.min.js" crossorigin="anonymous"></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/tether/1.4.0/js/tether.min.js"
integrity="sha384-DztdAPBWPRXSA/3eYEEUWrWCy7G5KFbe8fFjk5JAIxUYHKkDx6Qin1DkWx51bBrb" crossorigin="anonymous"></script>
<script src="https://maxcdn.bootstrapcdn.com/bootstrap/4.0.0-alpha.6/js/bootstrap.min.js"
integrity="sha384-vBWWzlZJ8ea9aCX4pEW3rVHjgjt7zpkNpZk+02D9phzyeVkE+jo0ieGizqPLForn" crossorigin="anonymous"></script>
<script type="text/javascript" src="static/js/jquery.color.min.js"></script>
<script type="text/javascript">
$.fn.isInViewport = function() {
var elementTop = $(this).offset().top;
var elementBottom = elementTop + $(this).outerHeight();
var viewportTop = $(window).scrollTop();
var viewportBottom = viewportTop + $(window).height();
return elementBottom > viewportTop && elementTop < viewportBottom;
};
var allPublications = null;
$(function() {
getRealSize = function(bgImg) {
var img = new Image();
img.src = bgImg.attr("src");
var width = img.width,
height = img.height;
return {
width: width,
height: height
}
};
getRealWindowSize = function() {
var winWidth = null,
winHeight = null;
if (window.innerWidth) winWidth = window.innerWidth;
else if ((document.body) && (document.body.clientWidth)) winWidth = document.body.clientWidth;
if (window.innerHeight) winHeight = window.innerHeight;
else if ((document.body) && (document.body.clientHeight)) winHeight = document.body.clientHeight;
if (document.documentElement && document.documentElement.clientHeight && document.documentElement.clientWidth) {
winHeight = document.documentElement.clientHeight;
winWidth = document.documentElement.clientWidth
}
return {
width: winWidth,
height: winHeight
}
};
fullBg = function() {
var bgImg = $("#background");
var mainContainer = $("#main");
var firstFire = null;
function resizeImg() {
var realSize = getRealSize(bgImg);
var imgWidth = realSize.width;
var imgHeight = realSize.height;
if (imgWidth == 0 || imgHeight == 0) {
setTimeout(function() {
resizeImg();
}, 200);
}
console.log(realSize);
var realWinSize = getRealWindowSize();
var winWidth = realWinSize.width;
var winHeight = realWinSize.height;
var widthRatio = winWidth / imgWidth;
var heightRatio = winHeight / imgHeight;
console.log(realWinSize);
if (widthRatio > heightRatio) {
bgImg.width(imgWidth * widthRatio + 'px').height(imgHeight * widthRatio + 'px').css({'top':
-(imgHeight * widthRatio - winHeight) / 10 * 5 + 'px', 'left': '0'})
} else {
bgImg.width(imgWidth * heightRatio + 'px').height(imgHeight * heightRatio + 'px').css({'left':
-(imgWidth * heightRatio - winWidth) / 10 * 3 + 'px', 'top': '0'})
}
}
resizeImg();
window.onresize = function() {
if (firstFire === null) {
firstFire = setTimeout(function() {
resizeImg();
firstFire = null
}, 100)
}
}
};
targetColor = $("#main-content-container .col-info .name").css("color");
animatedLink = function(speed) {
$("#main-content-container .col-link li").hover(function() {
$(this).find('.caption').animate({
color: targetColor
})
}, function() {
$(this).find('.caption').animate({
color: '#cccccc'
})
})
};
fullBg();
animatedLink(400);
allPublications = $("#main-pub-card-container .pub-card");
allTopicsLink = $("#main-pub-container .subtitle-aux a");
allTopics = [];
for (var topicId = 0; topicId < allTopicsLink.length; topicId++) {
allTopics.push({name: $(allTopicsLink[topicId]).data("topic"), title: $(allTopicsLink[topicId]).html()});
}
$("#main-pub-card-container").removeClass("hide");
});
</script>
</body>
</html>