Google unveils Gemini Robotics 2 to power humanoid robots without internet

Google DeepMind has introduced Gemini Robotics 2, a next-generation artificial intelligence system designed to enable humanoid robots to perform complex physical tasks autonomously without requiring an internet connection.

The company said the new model marks a significant advance in robotics by combining language understanding, visual perception and full-body motion control into a single AI system capable of operating in real-world environments.

At the core of Gemini Robotics 2 is what Google calls "intelligent whole-body control," allowing robots to interpret their surroundings, make decisions and execute coordinated physical actions with minimal human intervention.

In a demonstration released by Google, the humanoid robot was shown carrying out a range of household tasks, including picking up a watering can, installing a light bulb and tying a garbage bag.

Carolina Parada, head of robotics at Google DeepMind, said most robots currently deployed are designed for fixed, repetitive tasks and struggle to adapt when conditions change.

"Gemini Robotics 2 enables a full humanoid robot to understand natural language instructions and coordinate its entire body to complete tasks in dynamic environments," she said.

The system integrates a vision-language model with two vision-language-action models, enabling robots not only to understand spoken instructions but also to interpret visual information, recognize objects, navigate unfamiliar environments and carry out physical actions.

Google DeepMind described the technology as a major step toward physical artificial general intelligence (Physical AGI)—AI systems capable of applying human-like reasoning and adaptability in the physical world.

"Our long-term goal is to build AI that can work alongside people to solve complex problems in real-world settings," Parada said.

The announcement builds on earlier comments by Google DeepMind Chief Executive Demis Hassabis, who outlined plans to develop a common AI platform for robots similar to Android's role in the smartphone ecosystem.

Gemini Robotics 2 is expected to serve as a shared intelligence layer for humanoid robots developed by different manufacturers, allowing hardware companies to focus on robot design while Google's AI platform manages perception, decision-making and motion control.

Google also said the model runs entirely on-device, eliminating the need for continuous cloud connectivity. The capability could enable deployment in environments where internet access is unavailable or unreliable, including industrial facilities, remote locations and future space missions.