Harsh Gupta

@HarshGupta

Robotics @Stanford, advised by @SongShuran. Working on dexterous manipulation 🤖

California

hgupt3.github.io

Joined December 2024

442 Following

879 Followers

Replies Reposts Media

Harsh Gupta
@HarshGupta
Oct 7
Per-joint Gaussian noise is a terrible way to explore a 20+ DoF hand We introduce EigenDEXplore, which explores along coordinated patterns from human hand motion, while keeping full control of every joint Better learning, from tool use 🔨 to reorientation 🔄 to bimanual 👐
20
Harsh Gupta
@HarshGupta
Jun 18
Very true! Push-T was our first task before scaling up. The more interesting open problems are getting the right intent⇔control interfaces (flow/MANO/pixel/latent) and more robust sim2real manipulation policies. Hoping to explore these more in the coming months.
Animesh Garg
@animesh_garg
Jun 18
Replying to @harshgupta
It is likely that you dont need to focus much on this. The large scale pretraining from industry labs or larger video models will get a very good intent model. the major focus is getting the control model. And ensuring the sim2real works.
Harsh Gupta
@HarshGupta
Jun 18
We trained a dexterous hand 🤖 from internet-scale human videos 📺 with ZERO real-world robot or teleoperation data We pair a world model 🌎 that predicts task intent with a generalist sensorimotor policy ✋ trained sim2real to zero-shot fulfill ANY intent Introducing LUCID:
9
Harsh Gupta
@HarshGupta
Jun 1
I’ll be presenting UMI-on-Air ✈️ at #ICRA2026 🤖 📍 Hall C, Interactive Session 3 🗓️ Wed June 3, 9:00–10:30 AM Come by and say hi!
Harsh Gupta
@HarshGupta
Oct 8, 2025
✈️🤖 What if an embodiment-agnostic visuomotor policy could adapt to diverse robot embodiments at inference with no fine-tuning? Introducing UMI-on-Air, a framework that brings embodiment-aware guidance to diffusion policies for precise, contact-rich aerial manipulation.
Harsh Gupta
@HarshGupta
Dec 1, 2025
🚀 Code release! Our UMI-on-Air simulation suite is now live: github.com/LeCAR-Lab/UMI-… Excited to see how the community builds on our work for cross-embodied manipulation research.
GIF
Harsh Gupta
@HarshGupta
Oct 8, 2025
00:16
✈️🤖 What if an embodiment-agnostic visuomotor policy could adapt to diverse robot embodiments at inference with no fine-tuning? Introducing UMI-on-Air, a framework that brings embodiment-aware guidance to diffusion policies for precise, contact-rich aerial manipulation.
3

Harsh Gupta

Harsh Gupta’s posts