What happened
Inner Sky Labs, the company behind the Miko children's companion robots, has launched a suite of foundation models specifically designed for physical AI. These models handle perception, action planning, and safety checks for machines such as industrial arms, humanoids, and autonomous vehicles. Simultaneously, the startup is closing a ₹300 Cr (approximately $30.9 Mn) funding round, bringing its total funding to $100 Mn. The capital will support data collection, computing infrastructure, and model training as the company expands beyond its consumer robot brand into enterprise and government markets.
Inner Sky Labs, founded in 2015 by IIT Bombay alumni Sneh Vaswani, Prashant Iyengar, and Chintan Raikar, has unveiled a new umbrella brand for its physical AI initiatives. The company, previously known primarily for its Miko children's companion robots, is now launching a suite of foundation models designed to help machines perceive their surroundings, plan physical actions, and enforce safety rules. CEO Sneh Vaswani stated that these models were developed in-house over the past decade and are not fine-tuned from open-source models.
The platform comprises three core functions: perception, action, and safety. Perception models include Drishti for vision, Shruti for audio, and Sparsh for touch. Action models Kriya, Prana, and Karma use these inputs to plan movements and anticipate consequences. A separate safety layer, Kavach, checks proposed actions against more than 20 , such as detecting the presence of people and enforcing speed limits, before allowing execution. The company describes the system as hardware-agnostic, supporting industrial arms, humanoids, mobile robots, quadrupeds, cars, and CCTV systems.
Alongside the product launch, Inner Sky Labs is closing a funding round of ₹300 Cr (approximately $30.9 Mn). Investors include existing backers YourNest Venture Capital and Keshav R Murugesh, as well as new global investors. This round will bring the company's total funding to $100 Mn. The capital is earmarked for real-world data collection, computing infrastructure, and model training to support expansion into enterprise, government, and other machine-making sectors.
The company claims that commercial deployments began two to three quarters before this public launch, with four to five large deployments completed across enterprises and governments, including outside India. However, specific customer names and government applications have not been disclosed. The startup is currently serving enterprise markets in the US and Europe and is exploring opportunities in Africa.
Why it matters
This launch represents a significant shift for Inner Sky Labs from a consumer robotics company to a provider of core AI infrastructure for the physical world. By offering hardware-agnostic foundation models that include a dedicated safety layer, the company addresses a critical gap for manufacturers who cannot afford to build their own AI stacks. The emphasis on 'sovereign AI'—allowing models to run on-premise or in air-gapped environments—positions the technology as a viable option for sensitive government and industrial applications where data privacy and control are paramount. This move could accelerate the adoption of safe, autonomous physical AI in sectors ranging from manufacturing to surveillance.
The launch addresses a significant economic barrier for machine manufacturers. Vaswani argues that developing foundation models is becoming too expensive for individual companies, particularly those with modest production volumes. By providing a common provider of physical AI models, Inner Sky Labs allows multiple manufacturers to adopt advanced AI capabilities without the burden of building their own models from scratch.
The emphasis on 'sovereign AI' is a key differentiator. The platform allows customers to run models within their own infrastructure, including on-premise installations and on-device processing in offline or air-gapped environments. This is particularly relevant for sensitive applications like drones and city surveillance, where reliance on foreign cloud-based vision models may be unacceptable for security or data privacy reasons.
The integration of a dedicated safety layer (Kavach) with over 20 is critical for the adoption of physical AI in shared spaces. By blocking actions that breach predefined rules, such as moving into an area occupied by a person, the system aims to mitigate the risks associated with autonomous machines operating in dynamic environments.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
crm_get_transaction(id='4092').An agent must create a draft calendar event for Tuesday at 2 p.m. Which evidence would establish the requested result?
What to watch next
Monitor the specific performance metrics and safety validation results of the Kriya, Prana, and Karma action models in real-world industrial settings. Watch for the disclosure of named enterprise or government customers, as the current report only confirms four to five large deployments without identifying them. Additionally, track the scaling of the models from their current 4-9 billion range to the planned 12-37 billion parameters, which will determine their capability in handling complex multi-robot coordination and richer sensory inputs.
The roadmap indicates that the action models Kriya, Karma, and Prana will scale from 4B, 4B, and 9B parameters to 12B, 16B, and 37B parameters, respectively. The next generation of models will focus on coordinating multiple robotic arms, handling more complex scenes, and processing richer touch signals from fingertips, fingers, and palms. The success of this scaling will determine the platform's utility in complex industrial scenarios.
Independent verification of the safety claims is necessary. While the company states that the Kavach layer can block unsafe actions, specific test results or third-party audits of these in real-world conditions have not been provided in the source material.
The identity of the four to five large enterprise and government deployments mentioned by the CEO remains undisclosed. Future announcements naming these customers will provide concrete evidence of the platform's commercial viability and adoption in high-stakes sectors.