Harsh Gupta

This account is verified. Learn more

@HarshGupta

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

California hgupt3.github.io Joined December 2024

440 Following 881 Followers 72 Posts

Latest posts 5

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 👐

0:54

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 Verified account @animesh_garg Jun 18

@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.
View quoted post on X

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: Learning Embodiment-Agnostic Intent Models from Unstructured Human Videos for Scalable Dexterous Robot Skill Acquisition 🧵

1:02

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 Verified account @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.

View quoted post on X

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.

Harsh Gupta Verified account @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.

View quoted post on X