Postdoc @MIT_CSAIL | previously: @Cornell @Stanford @IITKgp
Cambridge, MA
Joined April 2013
- Tweets853
- Following2.6K
- Followers1K
- Likes16.3K
Pinned Tweet
Play2Perfect learns contact-rich, precise assembly skills like screwing.
The best part is watching the rollouts slowed down: the policy makes tiny corrections and recoveries that are essential for fast, dexterous manipulation.
Check out Tyler’s thread for more details 👇
Kushal retweeted
What can a robot hand learn from seconds of human motion?
It turns out: dexterous skills that generalize!
Excited to share AdaDex: Learning Adaptable Dexterous Skills via Hybrid Dexterous Mimicking.
From one short hand–object clip per skill, AdaDex learns an adaptable dexterous skill, with skills spanning functional grasping, object reposing, in-hand reorientation, and throw-and-catch.
Project page: yuxuank.com/AdaDex/
🧵👇 (1/8)
High-quality, high resolution tactile sensing tailor made for your robot hand.
Really awesome work enabled by the hardware expertise of Arik. Packaged in an agentic system, it works out of the box for any robot!
Kushal retweeted
How should two agents collaborate if they hold different information?
We are 95% certain they should communicate their uncertainty.
🚨In our new NeurIPS paper we show that communicating calibrated probabilities can benefit collaborative decision-making!
arxiv.org/abs/2610.19215
Kushal retweeted
@EkaRobotics has raised a $150M Series A.
Generality v/s Performance is the fundamental tension at the frontier of robotics. Generality without performance is just a prototype. Performance without generality is just a single-purpose tool.
Customers shouldn’t have to choose. Eka breaks the tradeoff. Powered by our Vision-Force-Action (VFA) model, we deliver the performance and versatility customers need to put robots to work every day.
Eka is built on three convictions:
⚡️Force is the language of manipulation.
🚀 Reinforcement learning is how dexterous manipulation scales.
🎂 System 1—fast, reliable physical execution—is the bottleneck to solve.
With this funding, we will scale our model and accelerate customer deployments.
Grateful to be building alongside a dream team and investors who share our ambition.
Kushal retweeted
❓How can humanoids learn by watching humans?
Naive retargeting can lose crucial contacts: like bracing a hand on a table while bending to pick an object off the floor.
🚀 Introducing VICAR: preserve key contacts from human videos, practice in simulation, deploy on hardware
1/n
Kushal retweeted
I’ve been lucky to overlap with Toni at MIT. It’s always a real pleasure working with her. She’s sharp, thoughtful, and a fantastic collaborator. If you’re hiring robotics research scientists, talk to her!
10/ On a personal note, this project is the final one of my PhD. It's been the most incredible privilege to work with folks at MIT and beyond over the past 6 years, and I can't wait for what's next! I'm on the job market for robotics research scientist positions starting in 2027 - so if you're hiring, I'd love to talk! tonibronars.github.io/
Kushal retweeted
Training for blind manipulation pressures your policy to learn an intelligent "searching" mechanism in context!
Every robot already carries contact sensors: its own joints. When a robot touches something, its joints drift slightly off their commands, and the encoders pick it up. Legged robots and humanoids already leverage this signal to walk blind - but is it enough for blind precise assembly?
We find that it is! Our policies can insert pegs, mesh gears, and thread nuts using only the robot's joint positions - no camera, no tactile, and no object poses. And the peg policy works zero-shot in the real world! 🧵
w/ my amazing co-first author @younghyo_park, and advised by @kushalk_ and @pulkitology.
Precise assembly is an important benchmark for robotics. Standard industrial automation fails to generalize, but robot learning struggles with contacts.
In this work, we extract force signals from proprioception and generalize well beyond the training distribution. Check it out!
Every robot already carries contact sensors: its own joints. When a robot touches something, its joints drift slightly off their commands, and the encoders pick it up. Legged robots and humanoids already leverage this signal to walk blind - but is it enough for blind precise assembly?
We find that it is! Our policies can insert pegs, mesh gears, and thread nuts using only the robot's joint positions - no camera, no tactile, and no object poses. And the peg policy works zero-shot in the real world! 🧵
w/ my amazing co-first author @younghyo_park, and advised by @kushalk_ and @pulkitology.
Kushal retweeted
Can a robot assemble tight-clearance parts using only its proprioception? No camera, no tactile sensor, no F/T sensor, no joint torque sensing, no priors over the object geometries or poses: just joint position readings.
It actually can! Blind policies trained in simulation insert pegs, mesh gears, and thread nuts, and the peg policy works zero-shot in the real world. Here's what we found 🧵
Joint work with @BronarsToni, my wonderful co-first author, who's on the job market right now! Advised by @kushalk_ and @pulkitology.
Resets are critical for RL! For dex hands, human demos are a rich source of reset states. RL can learn the dynamics between the states.
Awesome work by @prithwish_dan !!
To get an extremely fast system 1 loop, accurately modeling actuation dynamics is critical.
A cool method to calibrate this sim-to-real gap using realworld data!
(1/9) Same rewards. Less domain randomization. Higher jumps. Stable landings.
Just by improving the actuator model.
Introducing Contact-UAN: a residual world model for humanoids, learned from proprioception. No torque sensing or motion capture.
Project website: contact-uan.csail.mit.edu.
Kushal retweeted
(1/9) Same rewards. Less domain randomization. Higher jumps. Stable landings.
Just by improving the actuator model.
Introducing Contact-UAN: a residual world model for humanoids, learned from proprioception. No torque sensing or motion capture.
Project website: contact-uan.csail.mit.edu.
Compliant, general and so smoooooth!
@venki1899 and @michl_wang have done incredible research and engineering to make this kind of smooth motion plug and play for the entire lab. Seamless human-robot collaboration is imminent. Try it out!
As humanoid robots become part of our everyday lives, they need to do more than just perform impressive motions. They should be safe, fun to interact with, and respond naturally to unexpected contact, without sacrificing performance.
Our latest work, Compliant GMT (C-GMT), brings whole-body compliant responses to general motion tracking.
🌐 More details: venki1899.github.io/c-gmt/
Work done at MIT Improbable AI Lab with @michl_wang, Jin, @pulkitology.
Kushal retweeted
🎓 On-policy distillation has a teacher problem. SCOUT fixes it.
Then irony? OPD lets students learn from reasoning paths they actually visit. But their teachers must provide feedback on paths they rarely take themselves.
As these paths grow longer, even strong teachers struggle to provide reliable supervision.
Our idea is simple: Why not let the teacher and student learn together?
Kushal retweeted
Want to know how well agents can really control robots directly?
In RoboARCade, we systematically evaluate and diagnose frontier agents controlling robots directly using tens of thousands of dollars’ worth of agent trajectories. RoboARCade contains diverse, precise, and long-horizon tasks. Paper, code, assets and every trace are out today. roboarcade.org
I always wondered what happens when we scale RL in sim to 1M parallel environments. Turns out you can do really impressive manipulation behaviors that CANNOT be reliably teleop'd
Awesome work @mateoguaman and team! 👏
(1/N) Who says you can't do contact-rich, precise manipulation with sim-to-real RL?
No demonstrations, no retargeting, no tactile sensors, no crazy reward tuning. Just 1x, reactive policies trained with scaled up RL.
Excited to finally release Success-Guided Sampling (SGS)! More info in @octi_zhang's thread and more thoughts of my own below! 🧵
Website here: sgs-rl.github.io
Kushal retweeted
Human fingers move in coordinated patterns. Can those patterns help robots explore?
Excited to share EigenDEXplore! 🤖✋
Key idea: add exploration noise along directions learned from human hand motion, while leaving robots free to discover motions of their own
See @harshgupta's 🧵
Kushal retweeted
How can we train better behavior from imperfect robot demos?
In NEEDLEWORK, we study how we can get stitching to work on robot data, and more importantly, why these connections help.
Website: needle-work.github.io
Paper: arxiv.org/abs/2610.02339
Kushal retweeted
How should we control multi-fingered robot hands? 🖐️🤖
With RL policies. Using policies trained in simulation, our new 13 DoF hand for Atlas can autonomously rotate a ball between its fingers, spin a drumstick, and thread a nut. 🧵 (1/5)
Robots need to have long-context memory but don't have time to wait for long VLM reasoning traces.
Cool method to amortize the VLM reasoning cost at train-time without hurting fast inference!
There’s lots of buzz around agentic harnesses for robots, particularly for long horizon tasks that require complex reasoning and memory. But what will it really take to turn a reasoning agent into a reactive, reliable and low-latency robot policy?
In our new paper, workspace models, we design a new memory architecture that acts as a latent harness for stronger reasoning models. Here’s why we think this might be the way forward:
(1/9)
