Kushal

This account is verified. Learn more

@kushalk_

Postdoc @MIT_CSAIL | previously: @Cornell @Stanford @IITKgp

Cambridge, MA kushal2000.github.io Joined April 2013

2.6K Following 1.1K Followers 871 Posts

Latest posts 5

Pinned post

Kushal @kushalk_ Jul 1

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 👇

0:36

Tyler Lum @tylerlum23 Jul 1

🤖 How can we teach dexterous robots to perform precise, contact-rich assembly?

Introducing Play2Perfect: first learn to play with objects, then perfect the policy for tight insertion, multi-part assembly, and screwing.

Sound on! 🔊

🧵👇

View quoted post on X

Kushal @kushalk_ 2d

Makes me so happy to see it in the wild!!

Tyler Lum @tylerlum23 2d

Play2Perfect is already being reproduced and extended by @microagi at #CoRL2026!

Check out their demo combining our framework with an agentic planner!

View quoted post on X

Kushal @kushalk_ 2d

@tylerlum23 and I have been busy bringing Play2Perfect to new robots and simulators! Here’s a sneak peek at two robot setups in IsaacLab and MJLab 👇

Come chat with us at our poster today at 5pm at #CoRL2026!

0:10

Jeannette Bohg @leto__jean 3d

Play2Perfect is nominated for Best Paper at #CoRL2026

Come to the spotlight today at the 10:30am session and poster 126
View quoted post on X

Kushal @kushalk_ Nov 5

Doing manipulation with high inertia robots limits fine dexterity.

@younghyo_park and @BronarsToni have built up a solid line of work analyzing the science of robot learning so we can scale more efficiently! Great work 👏

Younghyo Park ✈️CoRL Verified account @younghyo_park Nov 5

Robot arms can be extremely heavy compared to the objects they are asked to manipulate, especially with their metal links and highly geared motors.

But is that actually a problem for robot learning, or are learning algorithms powerful enough to handle it? Are we paying any hidden costs for using heavy robots? Here's what we found 🧵

View quoted post on X

Kushal @kushalk_ Oct 29

Very cool paper! There are a lot of interesting questions to answer when environment interactions are scaled up in GPU-parallelized sims.

Immediately improved the performance of my RL runs after reading their reccommendations :)

Michael Beukman @mcbeukman Mar 31

1/ As compute continues to grow and simulators continue to improve, it is becoming feasible to train RL agents for billions or trillions of timesteps. However, this is only useful if agents can continue learning over such long training horizons, which is far from given 👇

View quoted post on X