Inner Sky Labs Unveils Kavach, Raises ₹300 Cr to Scale Physical AI

Inner Sky Labs' approach to physical AI focuses on enabling machines to sense, decide, and act safely in environments shared with humans. Their perception models, Drishti for vision, Shruti for audio, and Sparsh for touch, are designed to give machines a comprehensive understanding of their surroundings, moving beyond basic sensor input to interpret scenes and sounds more akin to human perception. This capability is presented as fundamental for machines to operate effectively and safely in complex, dynamic spaces.

The company's action models, Kriya, Prana, and Karma, aim to provide machines with generative intelligence for planning and executing physical movements, such as reaching or gripping. These models also incorporate predictive capabilities, allowing a machine to anticipate the consequences of its actions, like whether a box might slip or a person nearby might move. This anticipatory judgment is intended to embed foresight into a machine's decision-making process, a capability that typically relies on human instinct.

A key component of the system is Kavach, an independent safety layer designed to validate every action a machine is about to take. This system acts as a final veto, checking for factors like clear space, safe speed, and adherence to defined operational zones. If any safety criterion is not met, Kavach blocks the action before it begins, regardless of the machine's primary decision. This independent safety mechanism, developed from the company's prior work with the Miko companion robot, is positioned as crucial for building and maintaining trust in machines operating in close proximity to people.