Bipasha Sen
@bipashasen31
Followers
2K
Following
909
Media
35
Statuses
256
π Introducing Frank π€βa whole-body robot control system for day-to-day household chores. Frank has been in works for the past yearβa tightly coupled hardware and a remote teleoperation interface co-led by @michl_wang, @nst372, and @pulkitology. Thanks to @skymanaditya1 and
8
34
244
Three years back I wanted to code on a real robot, and now I own one!
41
51
2K
I'll be starting my Ph.D. at @MIT_CSAIL advised by @pulkitology!. I was fortunate to receive offers from amazing labs and I wished - many times - to clone myself and join each of them -- I wouldn't be here if not for my advisors, Prof. C V Jawahar, @vinaypn, and Madhav Krishna!
30
10
595
Excited to be a part of ConceptGraphs -- a unique representation of the 3D world highly amenable to roboticsπ€πΎ! . Beyond capturing semantics of and between objects -- ConceptGraphs is also integrated with powerful LLMs enabling highly complex queries! π
4
17
112
My personal highlight of #ICRA2023! . ICRA was grand in every aspect, be it the exhibitions showcasing the state of the art robots, the amazing posters, and mind-blowing keynotes and plenary talks. And interestingly. this is still just the beginning for robotics!
3
14
95
π’π’ Introducing EDMP - Ensemble-of-costs-guidance diffusion for motion planning. EDMP combines the strengths of deep learning and classical planning to solve complex plans DIRECTLY AT INFERENCE without ANY specialized training!! . Webpage: π§΅.
1
19
78
Excited to share "HyP-NeRF" recently accepted to #NeurIPS2023! . HyP-NeRF doesn't do just one thing -- it does many things!!. It can generate NeRFs (directly the parameters!) from text, single-view, and multi-view (occluded/non-occluded) images! . Page: π
1
13
74
Our technical report is here: ππ SOUND ON!! .Here's a clip of you coming back home to π€ Frank π€setting up your dinner table π«ππ²π₯π’
π Introducing Frank π€βa whole-body robot control system for day-to-day household chores. Frank has been in works for the past yearβa tightly coupled hardware and a remote teleoperation interface co-led by @michl_wang, @nst372, and @pulkitology. Thanks to @skymanaditya1 and
0
7
66
I'm presenting HyP-NeRF at #NeurIPS2023 at booth 438 now along with @melonmusk42 and @skymanaditya1 ! Come check us out!
Excited to share "HyP-NeRF" recently accepted to #NeurIPS2023! . HyP-NeRF doesn't do just one thing -- it does many things!!. It can generate NeRFs (directly the parameters!) from text, single-view, and multi-view (occluded/non-occluded) images! . Page: π
0
3
53
Had such a great last few weeks at @UMontreal / @Mila_Quebec, it feels really sad to leave! . I not only got to work on really cool real-world robotic setups, but also make so many memories! Especially, Montreal, it is unlike any other city I've lived in -- Au revoir! πΎπ€
2
1
41
We presented our paper "SCARP: Shape Completion in ARbitrary Poses for Improved Grasping" yesterday at #ICRA2023! . Along with @skymanaditya1, @melonmusk42, at @ieee_ras_icra
0
4
37
I will be presenting EDMP at #CoRL2023 Pretraining for Robot Learning Workshop!.
π’π’ Introducing EDMP - Ensemble-of-costs-guidance diffusion for motion planning. EDMP combines the strengths of deep learning and classical planning to solve complex plans DIRECTLY AT INFERENCE without ANY specialized training!! . Webpage: π§΅.
1
3
35
Check out our #IROS2024 work on diffusion-based Constrained 6-DoF Grasp Generation led by @melonmusk42 !.
π’Excited to share that our paper CGDF has been accepted to #IROS2024π. π Title: Constrained 6-DoF Grasp Generation on Complex Shapes for Improved Dual-Arm Manipulation. β¨ CGDF is a grasp diffusion model that generates targeted grasps on object parts efficiently. π. π§΅1/n
0
2
30
Excited to share that our work, SCARP: 3D Shape Completion in ARbitrary Poses for Improved Grasping has been accepted at #ICRA2023 @ieee_ras_icra! . Get started with SCARP at More details to be out soon!. @drsrinathsridha, @skymanaditya1, @melonmusk42
0
5
23
Our recent work on "Uncovering hidden biases against Indian independent artists in the perception of music quality by Indian Audience" accepted at the International Conference on Music Perception and Cognition (ICMPC) is very different from my usual line of work!. A thread π§΅β¬οΈ.
3
1
22
I'll be officially representing @ieee_ras_icra (#ICRA2023) as a Social Media Ambassador! . This is super duper exciting in many ways - first time at a robotics conference, my first robotics paper, and my first in-person major conference! See you at ICRA 2023 in London! @ieeeras.
1
1
21
FaceOff tries to address a critical problem faced by the movie industry π¬π₯ -- automating the task of swapping an actors' face and expression on a double's body COSTING MILLIONS OFπ²!. Explore our code below!. Refer to to learn more about the problem!.
Happy to announce the code release of our #WACV2023 project that brings the task of video-to-video face-swapping πto the world of moviemaking!π₯.ποΈπ Lights, camera, face swap! π¬π«.Give it a go here- .Check out the page for interesting future directions!
0
2
19
Perhaps, @CVPR should consider making the author rebuttals longer than just one page? . So many points are missed and not thoroughly explained because of the 1-page limitation to answer complex questions posed by 3-4 reviewers?.
1
0
15
After the super grueling (and fantastic) 3 months of PhD applications and interviews, and visa process, finally going to experience the "mini" fruit of "grad visit days"! . Excited to see the labs from up close!
0
0
14
Building great hardware is underrated. The right hardware is deeply problem-specificβif there's no strong fit, trusting an off-the-shelf solution can either bottleneck performance or lead to overengineering for a simple problem.
Companies who want to bring AI into the physical world are quickly realizing hardware isnβt a commodity. Building great functioning robot hardware is incredibly difficult. Its no longer a story of great AI & dumb hardware. This is a story of Great Hardware + Great AI.
0
1
15
Presenting "INR-V" at the Indian Conference on Computer Vision, Graphics and Image Processing! Thank you @iitgn for the amazing platform and hosting the amazing audience! . Check our work here - Code will be out soon! . @skymanaditya1 and @vinaypn
1
0
10
I'm in disbelief. Chandler Bing is my all time favorite character in FRIENDS and I always wanted to be like him. Now knowing that Mathew Perry is no more, FRIENDS won't be the same for me anymore. π I hope he has the most amazing afterlife.
0
1
10
Our autonomous policies are dropping soon! π. The global "neck" camera π¦ guides the local "wrist" cameras ββglobal gives context (like finding the object π), while local handles the details (like when to grab π€ and fine-tuning movements). Stay tuned!. [3/6]
1
0
8
Excited to dive deep into discussions on humanoids, robot arms, and general-purpose robotics with anyone around Palo Alto/SF! . If you're passionate about hardware, software, or AI in robotics and have hard opinions, let's brainstorm and share insights. @Stanford @UCBerkeley.
1
1
10
Wohoo! @benjamin_s70415 and @shankaravaidy are two of the most brilliant minds I know, and I am fortunate to have worked with them!! . Super excited to see where this goes π.
Robots can do flips and play chess, but they still canβt grab a snack or clean your table. Teleoperation is the key to unlocking real dexterity. Itβs not a compromiseβitβs a proven, powerful approach that combines human intuition with robotic precision. Weβre not just using
0
0
9
Whole-body remote teleoperation isnβt easy! π€. βοΈHow can the operator perceive the environment intuitively?. β
Our robot's 5-DoF "neck" lets teleoperators look around just like a humanβpeeking, scanning, and spotting items with ease!πππ€. [2/6]
1
0
8
π₯ production teams needed to remove cameras and equipment reflections from the mirrors in the shot! . Looks like now they'll need to do the same from the retinas too!.
Love this problem statement! Reconstruct the world from retinal reflections: The original paper from 2004 by Nishino and Nayar was a super inspiring example of creative research for me: Great to see a 3D version of it!!
0
0
8
Our code is out at - Shout-out to my collaborators - @aks1812, Madhav Krishna, @skymanaditya1, @VishalMandadi, @_ksaha, @Jayaram_Robot10, @Ajitsrikanth !.
0
1
7
An actuated neck π₯ helps localize the viewpoint, making it way easier for the teleoperator to perform complex and dexterous manipulation--like picking up a thin plate! . [4/6]
1
0
7
Our work "FaceOff: A Video-to-Video Face Swapping System" has been accepted at WACV 2023! This work opens a new direction of research that can potentially save millions of dollars in movie processing! Read on to know more!
1
2
7
At this point, chatGPT knows me so well, I am finding it very hard to use DeepSeek. First mover's advantage?.
4
0
9
Frank is leveling up fast, and weβre working hard to get it ready for deployment in your home π π€!. [5/6]
3
0
6
We have seen extensive work on "Image Inversion"; but what is "Video Inversion?". In our latest work, INR-V, accepted at @TmlrOrg, we propose a novel video .representation space that can be used to invert videos (complete and incomplete!). Project page:
INR-V: A Continuous Representation Space for Video-based Generative Tasks. Bipasha Sen, Aditya Agarwal, Vinay P Namboodiri, C.V. Jawahar.
1
1
6
@ManusMeta gloves are, without a doubt, the best finger-tracking gloves Iβve ever used. Pairing them with @PSYONICinc's Ability Hand made dexterous manipulation feel seamless. Excited to see these advancements paving the way for a robots-filled future we once only.
What if robots could peek around corners or adjust their viewpoint like humans? . @MITβs new research uses MANUS gloves and an actuated neck for intuitive #teleoperation, enabling robots to learn and adapt like never before. Read the full paper:
0
2
5
This is what we did at @icvgip2022 more or less :D. With. @skymanaditya1, @melonmusk42, and @gsc_2001
0
0
5
The most interesting aspect of ConceptGraphs for me is how easy it is to extend and adapt to any use case! . Do you need to store 6D object bounding boxes? Or customized task specific grasp poses? Everything and anything is possible with ConceptGraphs!
1
0
6
@siddancha This is a @RealmanRobotics. It is pretty good for its costβ5kg payload for a 7.8kg arm! + i know a lot of people who use these arms and had good things to say about it.
0
1
4
We came 3rd in #icra2022's Open Cloud Tabletop Rearrangement and Planning Challenge (OCRTOC - . Here is an article about us that summarizes our journey:
0
2
4
At the time of inference. we incorporate scene context as a set of cost functions that can be used to modify the trajectories sampled from the diffusion prior at each of the intermediate diffusion steps. !
1
0
4
Joint work with @vinaypn and @skymanaditya1.
INR-V: A Continuous Representation Space for Video-based Generative Tasks (TMLR 2022). Authors: Bipasha Sen, Aditya Agarwal, Vinay P Namboodiri, C.V. Jawahar. #neuralfieldsoftheday
0
1
4
Work done in collaboration with @qiaogu1997, @alihkw_, @_krishna_murthy, @SachaMori, @skymanaditya1, @duckietown_coo, and @florian_shkurti! . Find out more at .
0
0
5
@pulkitology @MIT_CSAIL @vinaypn Thank you! :):):) Looking forward to working with you and the group!.
0
0
2
Our training consists of only a set of kinematically valid trajectories -- what does that mean? . we let a diffusion model figure that part out. In a simplistic sense, our model learns a "prior" of smooth 7DoF plans between two joint configurations WITHOUT any scene context.
1
0
2
I think the entire approach of sorting students based on their qualities (brave, courageous, smart, etc.) into different houses (in Harry Potter) is flawed! . Where is the diversity?.
0
0
3
@thephotonguy Absolutely. I use it all the time for everything starting from my notes, to ideas, brainstorming, team meeting MOMs, task list, etc. 100% recommend it.!!.
0
0
3
Nothing more rewarding than getting our work recognized!! . Thank you, @WeeklyRobotics, for featuring SCARP #ICRA2023 as the "publication of the week"!. Find the paper at
Excited to share that our recent work on "SCARP: 3D Shape Completion in ARbitrary Poses" #ICRA2023 has been featured as the "Publication of the Week" at @WeeklyRobotics. Joint work with @NerdNess3195 @melonmusk42 @drsrinathsridha
0
0
2
@ylecun It wasn't just a one way street though, they gained immensely from the open source contributions and the tools and technologies developed by others. Giving birth to a tons of startup, was a parallel outcome!.
0
0
2
Long way we have come, long way to go!.
0
0
2
@ShubhamC526 I have heard so many such horror stories here. I am really lucky to have gotten the best advisors in India. I wish (if not all, but at least most) profs change the mindset of quick publications and actually start caring about the deeper problems and have a long-term vision.
1
0
2
Wohoo way to go robotics! (Although the intent of the post was a bit different).
Checking ICLR final decisions, it made me wonder why audio research is treated so poorly in ML conferences? π₯². I did a simple keyword search on OpenReivew and here is what I found! π. @iclr_conf #ICLR2023
0
0
2
. segregate the Indian indie artists into exclusive categories and playlists (ex. Indian indie artists) that may cause the listener's bias to creep in! . More details about the work to be out soon!. Work in collaboration with @skymanaditya1 and Prof. Vinoo Alluri.
0
0
1
Work in collaboration with @drsrinathsridha, @skymanaditya1, @melonmusk42, @rohithagaram, and K. Madhava Krishna! πΎπ€. Keep an eye out on our code - we are releasing the full version soon!.
0
1
2
Sounds like a really exciting opportunity!!.
If you have experience in generative modeling and differentiable rendering and are looking to join a fun team, I've recently co-founded a stealth startup in this space and we're looking for 1-2 ML experts still. Reach out w/ email & summary of what you've worked on in the past!.
0
0
1
@aks1812 I think this one thing that computer vision communities got right. People don't release code and also the tasks are often so niche that it is hard to find the exact paper to compare with. Every paper has their own task definition.
0
0
1
@GeorgiaChal Thank you for bringing his up @GeorgiaChal . looking forward to chatting with you at CORL :).
0
0
1
@Stone_Tao I'm sorry about the paper. Although I do agree with you pov, i do think presentation is important. Because, afterall, the point of a paper is to allow researchers to read and learn from the paper. And thus i would think it is important to get the main points across.
1
0
1
@d3lue @bchesky @GocmenMurat Not true. reviews help guests more than it does the hosts. I rely very much on reviews when selecting a place. so it is more like we are doing each other the favor.
2
0
1
0
0
1
0
0
1
@Manish__999 @MIT_CSAIL @pulkitology @vinaypn I think, in the end, it is about finding what is most mentally stimulating for you and what excites you the most.
0
0
1
n my #wacv2023 accepted work with @skymanaditya1 , "FaceOff: A Video-to-Video Face Swapping System", we automate this process! With our model, one can face-swap the actor's face on the double face with preserving the actor's expressions and the double's pose and scene context !!.
1
0
1
0
0
0
. when the audience is tricked and are told that they are listening to international musicians (while they are played the Indian indie music), the audience rated it to be of higher quality!. This is an important finding given that major music platforms like Spotify. .
2
0
1