Arhan

engineer by profession, musician by hobby |
research intern @iiit_hyderabad

[Arhan @__arh4n__]

@PrathamJainAI Yeahh experienced it

[Arhan @__arh4n__]

Maniskill is so goated for rl !!!

[Arhan @__arh4n__]

Online RL it is

[Arhan @__arh4n__]

Very cool!!! Was fun making this

[Sahil Apage @sahilsapage]

i love working on cool hardware:)

here, letting a rL agent control the hardware to play atari games, the agent only sees the game through camera

[Ashish Nerlekar @ashish4225]

Cup stacking with Lerobot teleoperation.

[Arhan @__arh4n__]

Cad!!!!

[Omkar Nanajkar @nomkar24]

Knee pitch joint...

[Arhan @__arh4n__]

@lazybanananann @yyw2000 🐐

[Arhan @__arh4n__]

@BerntBornich Crazy!!!

[Arhan @__arh4n__]

This is so cool !!!

[Robots Digest 🤖 @robotsdigest]

NoContactNoWorries tackles a core limitation in dexterous manipulation: most robots rely on tactile sensors to know when contact happens.

Instead, it estimates contact directly from vision + proprioception, removing the need for dedicated touch hardware while still enabling robust in-hand manipulation.

[Arhan @__arh4n__]

@k7agar @paradigm Congratulations 🎉🎉🎉

[Robots Digest 🤖 @robotsdigest]

DexJoCo is a strong step forward for dexterous manipulation benchmarks.

Instead of simple pick-and-place tasks, it focuses on functional interaction: using tools, pressing buttons, operating articulated objects, and bimanual coordination.

The most interesting part is how clearly it exposes current VLA weaknesses in precision, contact handling, and long-horizon manipulation.

[Robots Digest 🤖 @robotsdigest]

DexJoCo also exposes the embodiment gap in robotics foundation models.
Most VLAs were pretrained on gripper-based data. Dexterous hands introduce high-DOF control, coupled finger motion, and precise contact dynamics that current action heads struggle to model.

Scaling alone will not solve dexterity.

[Abhishek Anand @levelheaded_94]

I don’t think people are realising how differentiated our recent release is for physical ai -

1. Collected on commodity hardware
2 High fidelity multimodal data with rgb, dense depth maps, imu, room mesh, hand tracking, 6dof poses and dense action labels
3. Fully reproducible by anyone through Stera SDK.

The data engine for egocentric data is now open source.

https://t.co/4ASv70TgEA

[Arhan @__arh4n__]

Exiting !!!

[FPV Labs @fpv_labs]

Introducing Project Stera by FPV Labs, an open data infra for embodied AI research.

Project Stera includes Stera-10M, with 10M+ frames of long-horizon data with persistent state tracking, and an open-source pipeline that converts raw data into training-ready formats.

[Arhan @__arh4n__]

Yam arms are really cool !!!

[vrushtee @vruga_]

deployed a diffusion policy on yams the day before

the task
roll a tape, horizontal to vertical

the num_steps for denoising in the policy server made a huge difference in the actions of the bot
denoising samples more towards the action distribution

[Arhan @__arh4n__]

Banger workshop!!!!

[SRA VJTI @SRAVJTI]

We started the year with the basics: c programming, mechanical design ,embedded systems and computer vision

The best way to conclude the year is with our legacy 3dof manipulator based on ROS2 :
MARIO

200 First year students, 30 mentors and a weekend full of fun and learning

[Arhan @__arh4n__]

Pixels done and dusted

[SRA VJTI @SRAVJTI]

150+ students. 1 mission: make Computer Vision actually accessible using our oss codebase

At our PIXELS workshop , we didn’t just “teach OpenCV"

we broke everything down from scratch :
from pixels → perception.
From writing C++ to building real CV pipelines

[atharva ☆ @k7agar]

lmao this is unusually beautiful

[Arhan @__arh4n__]

@5ahilapage 🔥🔥🔥

[Arhan @__arh4n__]

Currently working on dextrous manipulation
Combined 2 wonderful open source projects So100 arm and amazing hand !!

Both great robots @huggingface @pollenrobotics 🔥🔥