Physical AI and robotics
Machines that see, decide and act in the physical world. The models arrived in 2025; the Indian deployment layer is being built now.
What we believe
The foundation models for robotics changed in 2025: single policies that control a whole body, vision-language-action models released as open weights, and simulation good enough to train on. What has not changed is that a robot is only as good as its perception and its data, and both are local. Indian roads, factories, ports and fields look like nothing in a Western training set.
That makes three things investable at seed in India: the perception layer (Paar Autonomy), the data and teleoperation operations that train and supervise robots for global companies, and inspection and maintenance robotics for the infrastructure India already has: ships, tanks, solar farms, transmission lines.
What must happen next
- Deployment rather than demonstration. The first Indian robotics companies with recurring revenue will be service businesses that own the robot and sell the outcome.
- Simulation and synthetic data for Indian environments as a product in its own right.
- Component supply: actuators, sensors and compute that do not depend on a single import route.
What we look for
- A named physical task with a paying customer and a measurable cost of doing it by hand.
- Founders from robotics, computer vision or controls, with someone who has run field operations.
- Businesses where the data collected in deployment compounds into the model.
What we pass on
Humanoid platforms without a task. Research projects with no customer inside 18 months.