Kushal

@kushalk_

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

Cambridge, MA

kushal2000.github.io

Replies Reposts Media

Pinned
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 👇
00:00
Tyler Lum
@tylerlum23
Jul 1
00:29
🤖 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! 🔊 🧵👇
Kushal
@kushalk_
Sep 21
Love this workshop idea! The stuff that actually makes things work in robotics deserves an audience beyond the appendix
Younghyo Park
@younghyo_park
Sep 21
Robot-learning folks: have you ever found that a seemingly minor system choice beneath the policy—controller tuning, hardware selection, or inference infrastructure—made a much bigger difference to your robot’s performance than an architecture or algorithm tweak? How do you
1
Kushal
@kushalk_
Sep 10
Setting up new tasks with Play2Perfect is super easy! Here are 2 additional envs + assets that Opus 5 generated from SCRATCH. The same RL training code works without any modifications. Go play with the awesome demo @tylerlum23 created :) play2perfect.github.io
00:00
Tyler Lum
@tylerlum23
Sep 10
00:35
Excited to share that Play2Perfect was accepted to #CoRL2026! We also added an interactive demo to our project website 🖥️ Watch Play2Perfect perform precise, contact-rich assembly in MuJoCo, right in your browser! Zero-shot sim-to-sim transfer from Isaac Sim to MuJoCo.
2
Kushal
@kushalk_
Jul 1
Let humans be humans, and let robots be robots. Our paper at #ICRA2026, X-Diffusion views human data as noisy counterparts of robot data. Building on the Ambient Diffusion framework, we selectively integrate human data in the high-noise timesteps of diffusion policy training 👇
Maximus Pace
@maxpace1_
Jul 1
Teaching a robot shouldn't require humans to act like robots. Human demonstrations contain valuable signal for robot manipulation, but they aren’t directly transferable to robots. X-Diffusion learns from noisy human demonstrations while staying within the robot’s capabilities.
00:00
1
Kushal
@kushalk_
Jun 18
Cool work! As we deploy robots in the real world, there will be tonnes of noisy interaction data to leverage. But throwing all of that data into the model naively won't cut it. Big fan of methods that can show positive transfer from noisy data!
Adam Wei
@adamwei_
Jun 18
🤖 We introduce Ambient Diffusion Policy, a simple and principled method for training policies with suboptimal data in robotics. Suboptimal data is everywhere in robotics… ❌ Data filtering is wasteful ❌ Co-training learns both good and bad features ✅ Ambient Diffusion Policy
00:00