Miko Maker Inner Sky Labs Unveils Foundation Models For Physical AI

AI startup Inner Sky Labs, the maker of children’s companion robots under the Miko brand, has launched a suite of foundation models designed to help machines perceive their surroundings, plan and execute physical actions, and check those actions against safety rules.

Alongside the launch, the startup is closing a ₹300 Cr (about $30.9 Mn) funding round involving existing backers YourNest Venture Capital and Keshav R Murugesh, as well as new global investors, cofounder and CEO Sneh Vaswani told Inc42.

Once completed, the round would take the business’s total funding to $100 Mn.

The startup plans to use the fresh capital for real-world data collection, computing infrastructure, and model training as it expands sales to enterprises, governments, and other machine makers.

The expansion builds on technology developed for Miko, which has been the business’ public-facing consumer brand since its early years. Founded in 2015 by IIT Bombay alumni Vaswani, Prashant Iyengar, and Chintan Raikar, the startup focused on children’s companion robots while developing the underlying AI models.

Vaswani said the broader physical AI initiative had been developed in stealth over the past decade. Inner Sky Labs is the newly unveiled umbrella brand for that business, operating under the same legal entity that owns Miko.

“We built these AI models ourselves, with a strong focus on safety and sovereignty. Now, we are scaling these models beyond Miko, taking them into enterprises, government applications that require sovereign AI, and other areas,” Vaswani told Inc42.

Taking Miko’s AI To Other Machines

The CEO said that the expansion had been planned since inception. During investor pitches in 2017, the team was already discussing developing AI models, with Miko as their first application before expanding into enterprise, software, and government use cases.

Limited demand for standalone AI models at the time led the team to focus on Miko. Its robots, designed to talk, play, and learn with children, provided an environment for deploying and testing the technology, particularly conversational safety.

With Inner Sky Labs, the startup is opening those capabilities to businesses building robots, industrial machines, and other systems. Customers can adopt individual models or the full platform, which is available to qualified partners by invitation.

Vaswani argued that developing foundation models is becoming too expensive for individual machine makers, particularly at modest production volumes. A common provider would allow multiple manufacturers to adopt the technology without each building their own models.

The startup describes its platform as hardware agnostic, supporting industrial arms, humanoids, mobile and companion robots, quadrupeds, cars, and CCTV systems.

AI For Sensing, Movement, And Safety

The platform combines three functions needed to operate these systems — perception, action, and safety.

Its perception models comprise Drishti for vision, Shruti for audio, and Sparsh for touch. Action models Kriya, Prana, and Karma use those inputs to help machines plan movements and anticipate their consequences.

Kavach, the independent safety layer, checks proposed actions before execution and can block those that breach predefined rules. The startup says the system has more than 20 guardrails, including checks on the presence of people, movement speeds, and permitted operating areas.

In a factory, for example, the models could help a robotic arm identify, grip, and move a component, while blocking movement if a person enters its operating area.

Vaswani said the models draw on almost a decade of physical-world data collection from factories, Miko’s manufacturing ecosystem, and other environments. He said they were developed in-house: “This is not fine-tuned on an open source model. It is our own model.”

The startup’s roadmap shows Kriya, Karma, and Prana scaling from 4 Bn, 4 Bn, and 9 Bn parameters to 12 Bn, 16 Bn, and 37 Bn, respectively.

According to Inner Sky Labs, the first generation is already being used in industrial applications, combining vision, information about a robot’s position and movement, and force and torque readings.

The next generation will focus on coordinating multiple robotic arms, handling more complex scenes, and processing richer touch signals from fingertips, fingers, and palms.

Sovereign AI And Enterprise Deployments

Beyond the models themselves, Inner Sky Labs is pitching control over deployment and operating rules as a selling point for enterprises and governments.

The startup positions its offering as sovereign AI, saying customers can run the models within their own infrastructure and define the rules governing their behaviour. Deployment options include cloud APIs, on-premise installations, and on-device processing, including offline and air-gapped environments isolated from external networks.

This allows sensitive operational data to be processed locally, with Vaswani citing drones and surveillance systems as potential applications.

“If you have CCTV cameras in a city, those cameras need to monitor exactly what has happened. You would not rely on a vision model coming from another country for any purpose. So, you would use a vision model from us,” he said.

Commercial deployments began two to three quarters before the public launch, Vaswani said. The startup has completed four to five large deployments across enterprises and governments, including outside India.

However, he did not disclose customer names or specific government applications. The startup claims to serve enterprise markets in the US and Europe and is exploring opportunities in Africa.

The launch comes amid growing activity in India’s physical AI ecosystem. While Addverb is developing humanoid robots for industrial applications, Mowito is building AI models that enable standard industrial robotic arms to learn tasks from human operators. 

Edited by Vinaykumar Rai

The post Miko Maker Inner Sky Labs Unveils Foundation Models For Physical AI appeared first on Inc42 Media.