
Robotics @amazon | PhD @UofIllinois
San Francisco, CA
Joined July 2020
- Tweets123
- Following804
- Followers251
- Likes511
Arjun Gupta retweeted
Excited to share that Whole-Body Dexterity was accepted as a poster at the upcoming CoRL Dexterous Manipulation: Full-Stack Robotics on Learning and Design workshop! 🦾
Looking forward to chatting during the poster session next week!
📍Thursday, Nov 12, 3:30-4:00 PM
Humanoids today mostly manipulate objects with their hands. As humans, we use much more than our hands: we brace boxes against our chests, tuck objects under our arms, and roll balls along our forearms.
Introducing Whole-Body Dexterity (WBD), turning the humanoid body into one large dexterous manipulator.
wholebodydex.com 🧵👇
Arjun Gupta retweeted
So proud of this one! 🤖 Dexterity we usually expect from a multi-finger hand (rolling, sliding, handoffs), now across the whole humanoid body.
Amazing work @cmcgartoll @dangan1693 @pthangeda_ @arjun__gupta @carohiguerarias @mukadammh @ManikantanNambi!
Humanoids today mostly manipulate objects with their hands. As humans, we use much more than our hands: we brace boxes against our chests, tuck objects under our arms, and roll balls along our forearms.
Introducing Whole-Body Dexterity (WBD), turning the humanoid body into one large dexterous manipulator.
wholebodydex.com 🧵👇
Arjun Gupta retweeted
We tend to think of dexterity as a hands problem🤔.
But the torso, neck, and arms are part of the machinery too. That's what lets us juggle multiple objects and move bulky payloads.
🤖📦🏀 Let's bring whole-body dexterity - WBD! 🏀📦🤖
🌐 wholebodydex.com
Humanoids today mostly manipulate objects with their hands. As humans, we use much more than our hands: we brace boxes against our chests, tuck objects under our arms, and roll balls along our forearms.
Introducing Whole-Body Dexterity (WBD), turning the humanoid body into one large dexterous manipulator.
wholebodydex.com 🧵👇
Arjun Gupta retweeted
Humanoids today mostly manipulate objects with their hands. Humans use much more than our hands: we brace boxes against our chests, tuck objects under our arms, and roll balls along our forearms.
Introducing Whole-Body Dexterity (WBD), turning the entire humanoid body into one big dexterous manipulator. Work led by @cmcgartoll, @dangann66, @pthangeda_.
🌐 wholebodydex.com
Arjun Gupta retweeted
We use all of our body to make contact, so should your humanoid: wholebodydex.com
This was a fun project that started off as a lunch convo with @dangan1693 and @cmcgartoll, then us realizing "wait we can do this we've been doing it with our hands sim2real since forever"
Humanoids don’t have to hold everything with their hands. Excited to finally share our new work, led by @cmcgartoll, @dangan1693, and @pthangeda_!
Humanoids today mostly manipulate objects with their hands. As humans, we use much more than our hands: we brace boxes against our chests, tuck objects under our arms, and roll balls along our forearms.
Introducing Whole-Body Dexterity (WBD), turning the humanoid body into one large dexterous manipulator.
wholebodydex.com 🧵👇
Arjun Gupta retweeted
Can a robot see how hard it is squeezing without force sensors?
VisualForce uses gripper deformation to learn force-aware manipulation—from handling berries to inserting plugs—without force or tactile sensors at deployment.
visual-force.github.io/
1/4
Arjun Gupta retweeted
Thrilled to share CARD: Contact-Anchored Retargeting for Dexterous manipulation. So far, collecting robot data, teleop, UMI, kinesthetic teaching, has run into the same two problems: 1) the mapping is largely kinematics only, and 2) the human has to act through the robot and hope the interaction works out. CARD instead treats contact and forces as the interface, from recording to replay, which is what lets it handle dexterous tasks that motion-only methods struggle with.
🧵 below 1/
Can you believe these are open-loop rollouts?
Excited to introduce CARD 🃏 — Contact-Anchored Retargeting for Dexterous Manipulation.
Almost every method today is limited to kinematics alone as the 𝐢𝐧𝐭𝐞𝐫𝐟𝐚𝐜𝐞 between the human and the robot. With CARD, we track real world contact and forces alongside kinematics to perform unprecedented dexterous tasks through tactile reactive replay!
CARD is scalable:
📹 Multiple input sources
🤖 Multiple robot hands
⚡ needs no object models, no simulator, 10 Hz
Arjun Gupta retweeted
🔥HERO live demo and code is available now! Play it at at
hero-humanoid.github.io/#int…
🔥Check out the code at
github.com/RunpeiDong/HERO
Arjun Gupta retweeted
I’ll be giving a talk at the sensors and actuators IROS workshop at 2:20pm today.
I will talk about what I think are, “the missing ingredients in scaling dexterous manipulation data”.
Workshop link: sensact4dexman-workshop.gith…
Come by and say hi if you’re in person!
Arjun Gupta retweeted
I’ve defended my PhD in Computer Science from @UofIllinois and joined @teslasi to work on @Tesla_Optimus!
Grateful to my advisors, Lana Lazebnik and @YunzhuLiYZ, and everyone who supported me.
Excited to keep building robots that can work in the real world.
Arjun Gupta retweeted
Thrilled to announce that HERO has been accepted to @corl_conf 2026🤠! Congrats to every contributor!
I will be attending CoRL in person, and see you guys at Texas and let's talk about humanoids!
Want to ask your humanoid to get novel objects you want from a novel table? We introduce HERO, first achieving open-vocabulary visual loco-manipulation using human language queries! Now we can ask the humanoid to see and grasp the target object (eg, a carrot instead of a tissue box) with whole-body coordination.
How do we do it? Key designs and findings in this work:
We propose a residual-aware end-effector tracking policy that tracks the target accurately in a closed loop.
We find that the robot's forward kinematics are quite inaccurate, and thus we propose residual neural forward models that correct the end-effector FK and base leg odometry.
We design a modular system powered by visual foundation models that achieves generalizable grasping capabilities with 83% success rate on novel daily objects and daily scenes.
Check out our new project:
Learning Humanoid End-Effector Control for Open-Vocabulary Visual Loco-Manipulation
Project page: hero-humanoid.github.io/
Paper: arxiv.org/pdf/2602.16705
Arjun Gupta retweeted
ContactMimic augments keypoint tracking with explicit, part-level contact commands. Contact plays a crucial role in both locomotion and manipulation tasks. How can we incorporate contact labels into a tracking policy and make contact explicitly controllable? Check out our work!
Is keypoint tracking all a humanoid needs?
Introducing ContactMimic (CoRL 2026): test-time contact controllability for humanoids.
Same keypoints, flip one binary label → contact on or off.
lixinyao11.github.io/contact…
w/ @Xialin_He @RunpeiDong @ Saurabh Gupta
See you in Austin! 🤠
Arjun Gupta retweeted
Is keypoint tracking all a humanoid needs?
Introducing ContactMimic (CoRL 2026): test-time contact controllability for humanoids.
Same keypoints, flip one binary label → contact on or off.
lixinyao11.github.io/contact…
w/ @Xialin_He @RunpeiDong @ Saurabh Gupta
See you in Austin! 🤠
Arjun Gupta retweeted
Meet ContactMimic, a contact-conditioned motion tracker that makes object interaction work as you command it! I'm pleased to see it accepted to @corl_conf 2026, see you guys in Texas in Nov and let's talk!
Is keypoint tracking all a humanoid needs?
Introducing ContactMimic (CoRL 2026): test-time contact controllability for humanoids.
Same keypoints, flip one binary label → contact on or off.
lixinyao11.github.io/contact…
w/ @Xialin_He @RunpeiDong @ Saurabh Gupta
See you in Austin! 🤠
Arjun Gupta retweeted
We present Anchor-Align, a simple strategy to retain VLM generalization during VLA training. The surprising part: it works better than the standard recipe of co-training with vision-language data.
"The generalization you want in your VLA has to be in the expert demos!" folks say.
"Didn't we train VLM backbones on trillions of tokens already for it?", we ask.
We set out to reclaim some of this (free) OOD generalization in VLAs. Just need to think beyond BC! 🥲
anchoralignvla.github.io
Two edits on top of BC. Same demos. No extra data.
Arjun Gupta retweeted
After 4 wonderful years, I'm signing off from NVIDIA. Working at NVIDIA's Seattle Robotics Lab was an absolute pleasure. Amazing collaborators, a great research environment, and friendships I'll keep for life.
Before I start at @UMRobotics next year, I'm joining @proceptionAI to lead the effort to build the dexterous manipulation learning stack. Proception is a full-stack
robotics startup that is already shipping robot hands and data-collection gloves.
I believe solving dexterous manipulation requires co-designing the entire stack: hardware, software, and AI. Proception is one of the few places where you can
actually do that.
We're building the best dexterous manipulation system in the world — hands, data, and policies together. If that's the problem you want to work on, come build with us. Links below.
Arjun Gupta retweeted
Thrilled to share that I'll be joining the University of Michigan as an Assistant Professor in Fall 2027!
My lab will work on Robot Learning, Dexterous Manipulation, Robot Foundation Models & 3D Perception. I'm looking for students to join me. Please apply and reach out.
(1/2)
Arjun Gupta retweeted
Can robots perform complex assembly tasks, like motherboard assembly, by composing simple behaviors?
Yes!
Introducing CoStream: Composing Simple Behaviors for Generalizable Complex Manipulation
costream-simple.github.io/
Arjun Gupta retweeted
[1/5] 📢 Super excited about our new paper: "Thinking in Boxes"🎉
Here's the idea: fit 3D boxes around the objects in a single "real" photo, then move the boxes. The objects translate and rotate; two of them can swap places; the camera moves; and the scene holds up -- even in areas that may be hidden in the original.
























