DharaniLabs.ai

Dharani Labs · Physical AI

From human hands
to robot hands.

We build the whole pipeline for physical AI: wearables that record the force and touch cameras can't see, an operator network collecting demonstrations across India, annotation pipelines that turn raw video into training-ready data, and training and deploying robot action models on real hardware.

Request a sample dataset →
AURA glove · live data
Live visualization of data from the Dharani AURA gloves: fingertips light up with measured contact pressure.

01 · The pipeline

The whole company in six steps.

01 Wearables Open-source rigs and in-house designs like AURA: fingertip tactile arrays, finger bend, wrist IMU and a wrist camera. 02 Collect Partner vendors and our in-house operations team record demonstrations across India. 03 Annotate Raw data becomes robot-learning datasets and environments, shipped in standard training formats. 04 Simulate Raw data from real factories becomes realistic simulation assets for validating trained policies. 05 Model training We take state-of-the-art models and train them for specific use cases with our datasets. 06 Deploy On real robots in real environments. Our team also helps deploy customer robots in our partner factories.

02 · Collection & annotation

Operators collect it. Pipelines make it trainable.

Partner vendors and our in-house operations team record manipulation demonstrations across India, with AURA gloves, UMI grippers, and other open-source rigs.

Every episode ships as robot-learning datasets and environments, in the standard formats teams already train on.

Skeleton and hand joints recovered on egocentric footage from exocentric views.

03 · Simulation track

Your real objects, running in simulation.

Ship a camera, not a robot. Photograph the objects at your site and we return watertight, physics-ready twins for Isaac Sim (USD + PBR), MuJoCo, Gazebo, or plain GLB. We also use raw data from real factories to build realistic simulation assets for validating trained manipulation policies.

Multi-view photos of a real household object captured by walking around it with a phone.
1 · A handheld walk around the real object. No rig, no studio.
2 · The same object reconstructed: a watertight, textured 3D model.

04 · From data to deployed policy

Our diffusion trajectory policy learns manipulation directly from egocentric human video. No teleoperation, no lab capture rig.

Talk to us about a pilot →
A robot opening a drawer using the learned policy.
The learned policy opening a drawer on a real robot.

05 · Team

A small team, growing.

Kavya Shankar Kavya Shankar Founder
Robot Learning hiring now
Simulation hiring now
Operations hiring now
This could be you write to us

Research · India

We're looking for ambitious people who take ownership: researchers in robot learning and simulation across India. If that's you, write to us.

Partners · SF & everywhere

We collaborate with robotics teams and labs on pilots, data, and research, in San Francisco or any other city. Deployments run in our partner factories in India.