@@ -108,17 +108,32 @@ <h1>Kay - Liyiming Ke</h1>
108108 <!-- Preview Videos -->
109109 < div class ="list-item highlight previews " data-category ="highlight ">
110110
111- < a href ="https://personalrobotics.github.io/CCIL/ "> < video class ="preview3 " playsinline ="" muted ="" autoplay =""
111+ < a href ="https://goodcherrybot.github.io/ "> < video class ="preview1 " playsinline ="" muted ="" autoplay ="" loop ="">
112+ < source src ="images/20250330-pi05-makebed.mp4 " type ="video/mp4 ">
113+ </ video > </ a >
114+
115+ < a href ="https://goodcherrybot.github.io/ "> < video class ="preview2 " playsinline ="" muted ="" autoplay ="" loop ="">
116+ < source src ="images/20250330-pi05-putsink.mp4 " type ="video/mp4 ">
117+ </ video > </ a >
118+
119+ < a href ="https://goodcherrybot.github.io/ "> < video class ="preview3 " playsinline ="" muted ="" autoplay ="" loop ="">
120+ < source src ="images/20250110-hirobot.mp4 " type ="video/mp4 ">
121+ </ video > </ a >
122+
123+ </ div >
124+ < div class ="list-item highlight previews " data-category ="highlight ">
125+
126+ < a href ="https://personalrobotics.github.io/CCIL/ "> < video class ="preview1 " playsinline ="" muted ="" autoplay =""
112127 loop ="">
113128 < source src ="images/20241101-ccil-applied.mp4 " type ="video/mp4 ">
114129 </ video > </ a >
115130
116- < a href ="https://www.physicalintelligence.company/blog/pi0 "> < video class ="preview1 " playsinline ="" muted =""
131+ < a href ="https://www.physicalintelligence.company/blog/pi0 "> < video class ="preview2 " playsinline ="" muted =""
117132 autoplay ="" loop ="">
118133 < source src ="images/20241101-pizero_bussing_trashpile.mp4 " type ="video/mp4 ">
119134 </ video > </ a >
120135
121- < a href ="https://goodcherrybot.github.io/ "> < video class ="preview2 " playsinline ="" muted ="" autoplay ="" loop ="">
136+ < a href ="https://goodcherrybot.github.io/ "> < video class ="preview3 " playsinline ="" muted ="" autoplay ="" loop ="">
122137 < source src ="images/20230315-cherrybot-granola_retry.mp4 " type ="video/mp4 ">
123138 </ video > </ a >
124139
@@ -173,6 +188,70 @@ <h1>Kay - Liyiming Ke</h1>
173188
174189 <!-- Publications -->
175190
191+ < div class ="list-item publication " data-category ="publication ">
192+ < a href ="https://www.pi.website/blog/pi05 " class ="thumbnail ">
193+ < video playsinline ="" muted ="" autoplay ="" loop ="" width ="180px ">
194+ < source src ="images/20250330-pi05-makebed.mp4 " type ="video/mp4 ">
195+ </ video >
196+ </ a >
197+ < div class ="project-description ">
198+ < h3 > < a href ="https://www.pi.website/blog/pi05 "> π0.5: A Vision-Language-Action
199+ Model with Open World Generalization</ a > </ h3 >
200+ < p >
201+ Kevin Black, Noah Brown, James Darpinian, Karan Dhabalia, Danny Driess, Adnan Esmail, Michael Equi, Chelsea
202+ Finn, Niccolo Fusai, Manuel Y Galliker, Dibya Ghosh, Lachy Groom, Karol Hausman, Brian Ichter, Szymon
203+ Jakubczak, Tim Jones, < b > Liyiming Ke</ b > , Devin LeBlanc, Sergey Levine, Adrian Li-Bell, Mohith Mothukuri,
204+ Suraj Nair, Karl Pertsch, Allen Z Ren, Lucy Xiaoyang Shi, Laura Smith, Jost Tobias Springenberg, Kyle
205+ Stachowicz, James Tanner, Quan Vuong, Homer Walke, Anna Walling, Haohuan Wang, Lili Yu, Ury Zhilinsky< br >
206+ <!--<i>CoRL 2025</i><br>-->
207+ < a href ="https://www.physicalintelligence.company/download/pi05.pdf "> PDF</ a > •
208+ < span class ="excerpt-container ">
209+ < a href ="# " onclick ="event.preventDefault(); "> Summary</ a >
210+ < span class ="excerpt-content ">
211+ We send mobile robots to many AirBnB houses to generalize tasks across diverse, real-world environments.
212+ Our robots can perform some household chores like cleaning kitchens in unseen houses.
213+ </ span >
214+ </ span >
215+ </ p >
216+ </ div >
217+ </ div >
218+
219+
220+ < div class ="list-item publication " data-category ="publication ">
221+ < a href ="https://www.physicalintelligence.company/research/hirobot " class ="thumbnail ">
222+ < video playsinline ="" muted ="" autoplay ="" loop ="" width ="180px ">
223+ < source src ="images/20250110-hirobot.mp4 " type ="video/mp4 ">
224+ </ video >
225+ </ a >
226+ < div class ="project-description ">
227+ < h3 > < a href ="https://www.physicalintelligence.company/research/hirobot "> Hi Robot: Open-Ended Instruction
228+ Following
229+ with Hierarchical
230+ Vision-Language-Action Models</ a > </ h3 >
231+ < p >
232+ Lucy Xiaoyang Shi, Brian Ichter, Michael Equi, < b > Liyiming Ke</ b > , Karl Pertsch, Quan Vuong, James Tanner,
233+ Anna Walling, Haohuan Wang, Niccolo Fusai, Adrian Li-Bell, Danny Driess, Lachy Groom, Sergey Levine, Chelsea
234+ Finn< br >
235+ < i > ICML 2025</ i > < br >
236+ < a href ="https://arxiv.org/abs/2502.19417 "> PDF</ a > •
237+ < span class ="excerpt-container ">
238+ < a href ="# " onclick ="event.preventDefault(); "> Summary</ a >
239+ < span class ="excerpt-content ">
240+ We introduce a hierarchical system enabling robots to “think aloud” and deconstruct complex tasks ("make
241+ me a sandwich") into
242+ manageable steps ("pick up bread, pick up tomato, put tomato on the bread ..."). By combining a
243+ low-level action
244+ model for execution and a
245+ high-level
246+ vision-language model for reasoning and interaction with human inputs, we allow robots to follow complex
247+ instructions and perform tasks with high
248+ precision and adaptability.
249+ </ span >
250+ </ span >
251+ </ p >
252+ </ div >
253+ </ div >
254+
176255 < div class ="list-item publication " data-category ="publication ">
177256 < a href ="https://www.physicalintelligence.company/blog/pi0 " class ="thumbnail ">
178257 < video playsinline ="" muted ="" autoplay ="" loop ="" width ="180px ">
@@ -188,10 +267,10 @@ <h3><a href="https://www.physicalintelligence.company/blog/pi0">π0: A Visio
188267 Lachy Groom, Karol Hausman, Brian Ichter, Szymon Jakubczak, Tim Jones, < b > Liyiming Ke</ b > , Sergey Levine,
189268 Adrian Li-Bell, Mohith Mothukuri, Suraj Nair, Karl Pertsch, Lucy Xiaoyang Shi, James Tanner, Quan Vuong,
190269 Anna Walling, Haohuan Wang, Ury Zhilinsky< br >
191- <!--<i>Venue 2024 </i><br>-- >
270+ < i > RSS 2025 </ i > < br >
192271 < a href ="https://www.physicalintelligence.company/download/pi0.pdf "> PDF</ a > •
193272 < span class ="excerpt-container ">
194- < a href ="# "> Summary</ a >
273+ < a href ="# " onclick =" event.preventDefault(); " > Summary</ a >
195274 < span class ="excerpt-content ">
196275 Can you train cross-embodiment robotic policies over many many tasks and expect it to work? We show that
197276 it is promising: a big pre-training model can be finetuned on a single task and outperform
@@ -214,7 +293,7 @@ <h3><a href="https://arxiv.org/abs/2410.20254">Overcoming the Sim-to-Real Gap: L
214293 < i > NeurIPS 2024</ i > < br >
215294 < a href ="https://arxiv.org/abs/2410.20254 "> PDF</ a > •
216295 < span class ="excerpt-container ">
217- < a href ="# "> Summary</ a >
296+ < a href ="# " onclick =" event.preventDefault(); " > Summary</ a >
218297 < span class ="excerpt-content ">
219298 We show that, learning an exploration policy in simulation can boost the real-world reinforcement
220299 learning
@@ -236,11 +315,11 @@ <h3><a href="https://arxiv.org/abs/2405.19307">Data Efficient Behavior Cloning f
236315 Continuity-based Corrective Labels</ a > </ h3 >
237316 < p >
238317 Abhay Deshpande, < b > Liyiming Ke</ b > , Quinn Pfeifer, Abhishek Gupta, Siddhartha S. Srinivasa< br >
239- < i > In submission 2024</ i > < br >
318+ < i > IROS 2024</ i > < br >
240319 < a href ="https://personalrobotics.github.io/CCIL/ "> Webpage</ a > •
241320 < a href ="https://arxiv.org/abs/2405.19307 "> PDF</ a > •
242321 < span class ="excerpt-container ">
243- < a href ="# "> Summary</ a >
322+ < a href ="# " onclick =" event.preventDefault(); " > Summary</ a >
244323 < span class ="excerpt-content ">
245324 We apply CCIL to real world robotic manipulation tasks and it kinda worked after some design tweak. The
246325 most juice comes from setting up trust threshold for the generated labels in a task-agnostic way.
@@ -259,12 +338,12 @@ <h3><a href="https://arxiv.org/abs/2310.12972v1">CCIL: Continuity-based Data Aug
259338 Imitation Learning</ a > </ h3 >
260339 < p >
261340 < b > Liyiming Ke*</ b > , Yunchu Zhang*, Abhay Deshpande, Siddhartha Srinivasa, Abhishek Gupta< br >
262- < i > International Conference on Learning Representations ( ICLR) 2024</ i > < br >
341+ < i > ICLR 2024</ i > < br >
263342 < a href ="https://personalrobotics.github.io/CCIL/ "> Webpage</ a > •
264343 < a href ="https://github.com/personalrobotics/CCIL "> Code</ a > •
265344 < a href ="https://arxiv.org/abs/2310.12972v1 "> PDF</ a > •
266345 < span class ="excerpt-container ">
267- < a href ="# "> Summary</ a >
346+ < a href ="# " onclick =" event.preventDefault(); " > Summary</ a >
268347 < span class ="excerpt-content ">
269348 Enhances robustness of imitation learning by generating synthetic corrective labels:
270349 The trick is to leverage local continuity in the environment dynamics - and for regions that are
@@ -285,11 +364,11 @@ <h3><a href="https://arxiv.org/abs/2310.12972v1">CCIL: Continuity-based Data Aug
285364 < h3 > < a href ="https://goodcherrybot.github.io/ "> Cherry Picking with Reinforcement Learning</ a > </ h3 >
286365 < p >
287366 Yunchu Zhang*, < b > Liyiming Ke*</ b > , Abhay Deshpande, Abhishek Gupta, Siddhartha Srinivasa< br >
288- < i > Robotics Science and Systems ( RSS) 2023</ i > < br >
367+ < i > RSS 2023</ i > < br >
289368 < a href ="https://goodcherrybot.github.io/ "> Webpage</ a > •
290369 < a href ="https://arxiv.org/abs/2303.05508 "> PDF</ a > •
291370 < span class ="excerpt-container ">
292- < a href ="# "> Summary</ a >
371+ < a href ="# " onclick =" event.preventDefault(); " > Summary</ a >
293372 < span class ="excerpt-content ">
294373 Use reinforcement learning to learn fine motor skills: pick up slippery cherries with chopsticks under
295374 wind or human disturbances. And I refuse to do parameter sweeping or random seed cherry picking.
@@ -308,11 +387,11 @@ <h3><a href="https://sites.google.com/view/real-orl">Real World Offline Reinforc
308387 Data Sources</ a > </ h3 >
309388 < p >
310389 Gaoyue Zhou*, < b > Liyiming Ke*</ b > , Siddhartha Srinivasa, Abhinav Gupta, Aravind Rajeswaran, Vikash Kumar< br >
311- < i > IEEE International Conference on Robotics and Automation ( ICRA) 2023</ i > < br >
390+ < i > ICRA 2023</ i > < br >
312391 < a href ="https://sites.google.com/view/real-orl "> Webpage</ a > •
313392 < a href ="https://arxiv.org/abs/2210.06479 "> PDF</ a > •
314393 < span class ="excerpt-container ">
315- < a href ="# "> Summary</ a >
394+ < a href ="# " onclick =" event.preventDefault(); " > Summary</ a >
316395 < span class ="excerpt-content ">
317396 Eval offline RL in real-world: emphasize on data being "kinda good" but not perfect.
318397 </ span >
@@ -330,10 +409,10 @@ <h3><a href="https://personalrobotics.cs.washington.edu/publications/ke2021grasp
330409 Chopsticks: Combating Covariate Shift in Model-free Imitation Learning for Fine Manipulation</ a > </ h3 >
331410 < p >
332411 < b > Liyiming Ke</ b > , Jingqiang Wang, Tapomayukh Bhattacharjee, Byron Boots, Siddhartha S. Srinivasa< br >
333- < i > IEEE International Conference on Robotics and Automation ( ICRA) 2021</ i > < br >
412+ < i > ICRA 2021</ i > < br >
334413 < a href ="https://arxiv.org/abs/2011.06719 "> PDF</ a > •
335414 < span class ="excerpt-container ">
336- < a href ="# "> Summary</ a >
415+ < a href ="# " onclick =" event.preventDefault(); " > Summary</ a >
337416 < span class ="excerpt-content ">
338417 Teach a robot to use chopsticks for precise manipulation tasks through human demonstrations: Addresses
339418 covariate shift in imitation learning by noise-injection, object-centric transformation and
@@ -354,10 +433,10 @@ <h3><a href="https://personalrobotics.cs.washington.edu/publications/ke2020teleo
354433 < p >
355434 < b > Liyiming Ke</ b > , Ajinkya Kamat, Jingqiang Wang, Tapomayukh Bhattacharjee, Christoforos Mavrogiannis,
356435 Siddhartha S. Srinivasa< br >
357- < i > IEEE International Conference on Intelligent Robots and Systems ( IROS) 2020</ i > < br >
436+ < i > IROS 2020</ i > < br >
358437 < a href ="https://arxiv.org/abs/2008.00101 "> PDF</ a > •
359438 < span class ="excerpt-container ">
360- < a href ="# "> Summary</ a >
439+ < a href ="# " onclick =" event.preventDefault(); " > Summary</ a >
361440 < span class ="excerpt-content ">
362441 Built a chopsticks robot and a fun human-interactive demo collection interface: turns out that tracking
363442 a
@@ -376,10 +455,10 @@ <h3><a href="https://personalrobotics.cs.washington.edu/publications/ke2020teleo
376455 < h3 > < a href ="https://arxiv.org/abs/1905.12888 "> Imitation Learning as f-Divergence Minimization</ a > </ h3 >
377456 < p >
378457 < b > Liyiming Ke</ b > , Sanjiban Choudhury, Matt Barnes, Wen Sun, Gilwoo Lee, Siddhartha Srinivasa< br >
379- < i > International Workshop on the Algorithmic Foundations of Robotics ( WAFR) 2020</ i > < br >
458+ < i > WAFR 2020</ i > < br >
380459 < a href ="https://arxiv.org/abs/1905.12888 "> PDF</ a > •
381460 < span class ="excerpt-container ">
382- < a href ="# "> Summary</ a >
461+ < a href ="# " onclick =" event.preventDefault(); " > Summary</ a >
383462 < span class ="excerpt-content ">
384463 A unified theoretical framework for imitation learning! Turns out some SOTA algorithms are using
385464 f-divergence. We show how different divergence measures lead to different imitation learning approaches.
@@ -401,11 +480,11 @@ <h3><a
401480 < p >
402481 < b > Liyiming Ke</ b > , Xiujun Li, Yonatan Bisk, Ari Holtzman, Zhe Gan, Jingjing Liu, Jianfeng Gao, Yejin Choi,
403482 Siddhartha Srinivasa< br >
404- < i > IEEE Conference on Computer Vision and Pattern Recognition ( CVPR) 2019</ i > < br >
483+ < i > CVPR 2019</ i >
405484 < font color ="49bf9 "> < i > ★ Oral Presentation, CVPR (5.6%) ★</ i > </ font > < br >
406485 < a href ="https://arxiv.org/abs/1903.02547 "> PDF</ a > •
407486 < span class ="excerpt-container ">
408- < a href ="# "> Summary</ a >
487+ < a href ="# " onclick =" event.preventDefault(); " > Summary</ a >
409488 < span class ="excerpt-content ">
410489 Baking Search and Planning into ML-based navigation: We propose a new framework for VL navigation,
411490 enabling agents to recover from mistakes by maintaining internal search tree and returning to previous
@@ -426,10 +505,10 @@ <h3><a href="https://ojs.aaai.org/index.php/AAAI/article/view/10061">Behavioral
426505 Evasion</ a > </ h3 >
427506 < p >
428507 < b > Liyiming Ke</ b > , Bo Li, Yevgeniy Vorobeychik< br >
429- < i > AAAI Conference on Artificial Intelligence (AAAI) 2016</ i > < br >
508+ < i > AAAI 2016</ i > < br >
430509 < a href ="https://ojs.aaai.org/index.php/AAAI/article/view/10061 "> PDF</ a > •
431510 < span class ="excerpt-container ">
432- < a href ="# "> Summary</ a >
511+ < a href ="# " onclick =" event.preventDefault(); " > Summary</ a >
433512 < span class ="excerpt-content ">
434513 Studies how humans attempt to evade email spam filters.
435514 Provides insights into adversarial behavior and implications for security system design.
@@ -444,6 +523,14 @@ <h3><a href="https://ojs.aaai.org/index.php/AAAI/article/view/10061">Behavioral
444523 Some of my slides can be found <a href="https://slides.com/andyzeng">here</a>
445524 </div> -->
446525
526+ < div class ="list-item talk " data-category ="talk ">
527+ < p class ="date "> 2025</ p > RSS Workshop (upcoming)
528+ </ div >
529+
530+ < div class ="list-item talk " data-category ="talk ">
531+ < p class ="date "> </ p > Actuate (upcoming)
532+ </ div >
533+
447534 < div class ="list-item talk " data-category ="talk ">
448535 < p class ="date "> 2024</ p > OpenAI Reading Group
449536 </ div >
@@ -471,7 +558,8 @@ <h3><a href="https://ojs.aaai.org/index.php/AAAI/article/view/10061">Behavioral
471558 </ div >
472559
473560 < div class ="list-item talk " data-category ="talk ">
474- < p class ="date "> </ p > Stanford University, < a href ="https://iprl.stanford.edu/ "> Interactive Perception and Robot
561+ < p class ="date "> </ p > Stanford University, < a href ="https://iprl.stanford.edu/ "> Interactive Perception and
562+ Robot
475563 Learning Lab</ a >
476564 </ div >
477565
@@ -498,7 +586,18 @@ <h3><a href="https://ojs.aaai.org/index.php/AAAI/article/view/10061">Behavioral
498586
499587
500588 < div class ="list-item misc " data-category ="misc ">
501- < p class ="date "> </ p > Reviewer of AAMAS, CoRL, HRI, ICLR, ICRA, IJRR, IROS, NeurIPS, RA-L
589+ < p class ="date "> </ p > Reviewer of RSS, CoRL, ICLR, NeurIPS, ICRA, IJRR, IROS, RA-L, HRI, AAMAS, IJCAI
590+ </ div >
591+
592+ < div class ="list-item misc " data-category ="misc ">
593+ < p class ="date "> 2025</ p > We < a href ="https://github.com/Physical-Intelligence/openpi "> open source
594+ π 0 on Github</ a > to empower the community by sharing our foundation models.
595+ </ div >
596+
597+ < div class ="list-item misc " data-category ="misc ">
598+ < p class ="date "> 2024</ p > Our first generalist robotic model, π 0,
599+ is featured on < a href ="https://www.nytimes.com/2024/11/04/business/dealbook/selzer-poll-trump-trade.html ">
600+ New York Times</ a > and < a href ="https://www.wired.com/story/physical-intelligence-home-robot/ "> Wired</ a > .
502601 </ div >
503602
504603 < div class ="list-item misc " data-category ="misc ">
@@ -538,13 +637,15 @@ <h3><a href="https://ojs.aaai.org/index.php/AAAI/article/view/10061">Behavioral
538637 < li style ="display: inline; "> • </ li >
539638 < li style ="display: inline; "> < a href ="https://colah.github.io/ "> Colah's Blog</ a > </ li >
540639 < li style ="display: inline; "> • </ li >
640+ <!--
541641 <li style="display: inline;"><a href="http://neuralnetworksanddeeplearning.com/">Michael Nielsen</a></li>
542642 <li style="display: inline;"> • </li>
543643 <li style="display: inline;"><a href="https://karpathy.github.io/">Andrej Karpathy</a></li>
544644 <li style="display: inline;"> • </li>
545645 <li style="display: inline;"><a href="https://danielseita.github.io/">Seita's Place</a></li>
546646 <li style="display: inline;"> • </li>
547647 <li style="display: inline;"><a href="https://www.ruder.io/">Sebastian Ruder</a></li>
648+ -->
548649 </ ul >
549650 </ div >
550651
0 commit comments