Yash Jangir (@off_jangir) header
Yash Jangir (@off_jangir) profile picture

Generally Curious Agent. Ph.D with @mangahomanga. Previously, MS Robotics with @katerinafragiad and @ybisk

Pittsburgh, PA, USA
Joined August 2020
Photo by Yash Jangir (@off_jangir)Photo by Yash Jangir (@off_jangir)Photo by Yash Jangir (@off_jangir)Photo by Yash Jangir (@off_jangir)Photo by Yash Jangir (@off_jangir)Photo by Yash Jangir (@off_jangir)
Yash Jangir retweeted
Excited to share Grasping the Future! We taught a model something people do effortlessly: handling objects. It generates the whole interaction in 3D, both hands, their fingers and the object, as one motion. It learned from internet videos. grasping-the-future.github.i… 🧵 (1/8)
1
5
2
14
972
We had similar experiences where the reviewers keep responding same suggestion what GPT does i.e copy paste my rebuttal as a prompt. Seems like peer review is really dead : )
We are dealing with an interesting #ICLR2027 reviewer. They changed their ratings from 1 ➡️ 3 ➡️ 2 ➡️ 1 ➡️ 2 ➡️ 1 ➡️ 2. Today, the reviewer raised their confidence from 3 ("I am familiar with this topic & its relevant literature") to 4 ("I have substantial direct research experience on this topic”)! No surprises: we were able to reproduce their review and responses using GPT with a very simple prompt in a private chat with web search turned off.
11
1,487
I believe adapting robot policies using just a handful of demonstrations will be crucial for robots and Physical AI to work alongside humans, allowing them to learn new tasks and quickly adapt to variations of familiar ones. Check out exciting new work led by @BillyYYan and @irmakkguzey !
As we get closer to a world where we live among robots, it’s becoming more and more valuable for robots to learn quickly with and from people. I’m very excited to share Steertuning, a novel approach that adapts robots to new tasks from 5 human demonstrations in under 5 minutes, from demonstration to deployment. Website: steertuning.github.io/ Code: github.com/billy-yibo-yan/st…
1
5
587
Excited to share a preview @corl_conf of our upcoming paper on human-like grasping from human videos via 3D reconstruction and physics simulation.
Robots that learn should generalize. Locomotion should show generalization across terrain; manipulation across object categories. Here is a video of “Pick Anything”, demonstrating human like grasps for 66 different objects, ranging in size and shape from air pods to dumbbells to pipe wrenches, using a *single* policy. The system uses the stereo camera head for vision (no wrist cameras) & Sharpa Wave hands for manipulation. It was trained using video and RL (zero teleoperation data). This is research from our Dex-Hands team at Amazon FAR and we will be doing live demos of the system at CoRL 2026 ( Demo 12, Nov. 9 &10). Please come by!
2
4
51
2,158
What are the right failures to collect for training robot reward models? @g0kul_6 can tell you!
What are the right failures to collect for training robot reward models? We analyze failures from kinesthetic teaching, policy rollouts, human videos, and real2sim and their impact on reward quality and downstream real-world RL. rightkindofwrong.github.io 🧵
1
6
461
For Halloween, @geetachandra21 got our Unitree G1 to crawl like a spider! Give Geeta (co-advised w/ @HaiminHu) a follow if you're interested in more humanoid demos. We're doubling down on whole-body locomanipulation research at @HopkinsDSAI @HopkinsEngineer @JHUCompSci -- reach out if interested in working with us! (credit to Addison Johnson, who helped design the mounts for the hands, and filmed this; and to Geeta who took a full-stack approach to ensuring that no humans or humanoids were harmed during the production of this video! and to @jloganolson for last year's humanoid crawling deo) sciphylab.org/
4
10
4
85
13,769
Alex, his insights, and his work have always been amazing. Check out his latest work!
In language models, turning a generative model into a critic is simple: pass in the output and prompt for a score. Robot policies can't do this - they output physical actions & there's no built-in vocabulary to predict scores. So when it comes to RL on robot policies, we typically train a separate critic from sparse rewards - blind to everything the policy learned from the actions in the pretraining data. We think this has it backwards - can a critic learn from those actions the way the policy does? (1/n)
1
4
1,015
That clip of testing robustness is toptier. 😂
We present OpenTac, an open, full-stack ecosystem for 3D tactile sensing on any robot surface. Anyone can design and fabricate high-density tactile skins for any part of their robot, then read them out, simulate them, and share data across different sensors.
1
3
4
489
Tactile sensing is going to be a crucial piece of robot learning. To truly understand physical dynamics, robots need to learn not just what happens visually, but also how forces and contacts evolve during interaction. OpenTac is here! Check out this amazing work!
Your robot's shape shouldn't limit its sense of touch. Meet OpenTac: an open ecosystem for custom tactile skins that sense normal + shear forces. Very excited about this collaborative project! Can’t wait to see what the community builds with it. 🧵👇 opentac.github.io
5
11
760
Yash Jangir retweeted
We present OpenTac, an open, full-stack ecosystem for 3D tactile sensing on any robot surface. Anyone can design and fabricate high-density tactile skins for any part of their robot, then read them out, simulate them, and share data across different sensors.
14
47
15
286
29,341
Yash Jangir retweeted
🍂 Fall break came a little early for @_krishna_murthy, @jmin__cho, @HaiminHu and my group! We had our first lab retreat in Shenandoah National Park this weekend. We went hiking, played board games and enjoyed some great food. @JHUCompSci @HopkinsDSAI
2
3
78
4,713
Check out our latest work :) What 30,000 Hours of Egocentric Video Do Not Teach? It learned exactly where the hands go. What happens to the things they touch, much less so. fidelity-gap.github.io/ Amazing collaboration with @JiahuaDong @ptokmakov @Miccooper9 @mzubairirshad @ZakharovSergeyN @mangahomanga @vitorguizilini @ToyotaResearch
[1/7] We trained a world model on 30,000 hours of first-person video. It learned exactly where the hands go. What happens to the things they touch, much less so. Going from 300 to 30,000 hours did not narrow the gap, and a 4× larger model barely did either.
6
15
901
What 30,000 Hours of Egocentric Video Do Not Teach? Check out our fun analysis paper on egocentric world models, led by @JiahuaDong and @ptokmakov with a focus on hand-object interactions in daily activities! fidelity-gap.github.io
[1/7] We trained a world model on 30,000 hours of first-person video. It learned exactly where the hands go. What happens to the things they touch, much less so. Going from 300 to 30,000 hours did not narrow the gap, and a 4× larger model barely did either.
8
55
4,202
Wooowwww! What a teaser. @mangahomanga
NEVER LOOK BACK! Introducing LEDGER: persistent 3D object memory from egocentric videos. Remember where things were, when they moved, and what they held -> then answer questions without looking back at the video! ledger-3d.github.io
7
604
Bro really said CEO mode on. Look @KJHMiao is shining 🌟
We’re on a mission to build a billion AI models, not trained by us, but by you. Introducing @bryel_labs we help your company build its in-house AI lab and give you the software to make model development and continual improvement simpler. We are backed by @ycombinator and have trained models at Apple, Framer, and F500s. We’ve enabled companies to train: • Classification models that are 20x faster • Search models that cost 5x less. • Long horizon agent models that match the frontier at 5-8x lower cost. Are you interested in partnering with us? Reach out. bryel.ai/
2
307