Google Launches Gemini Robotics On-Device AI: Robots Go Offline, Stay Smart

Google DeepMind has unveiled a powerful new AI model — Gemini Robotics On-Device — designed to run directly on robots without the need for internet connectivity. Announced on June 24, the cutting-edge model brings real-time intelligence and agility to robots in offline or low-connectivity environments.

AI for a Cloud-Free Future

Unlike most AI models that rely on cloud processing, Gemini Robotics On-Device is optimized for latency-sensitive tasks and privacy-first deployments. It runs locally on the robot’s hardware, delivering near cloud-level performance in real-world scenarios while minimizing data transmission delays.

Despite being built for low-compute systems, the model retains robust capabilities, such as understanding natural language instructions, executing multi-step commands, and performing complex physical tasks.

Dexterous Tasks Without the Delay

Google demonstrated the AI’s ability to perform intricate activities like folding clothes, unzipping bags, and assembling parts, all while processing inputs directly on-device. This shift enables robots to respond instantly and operate securely in sensitive or remote environments.

Cross-Platform Compatibility with Franka FR3 and Apollo

Initially developed using ALOHA robotic arms, Gemini’s capabilities have now been extended to more advanced systems such as the Franka FR3 bi-arm robot and the Apollo humanoid platform. In tests, robots powered by Gemini On-Device successfully followed voice commands, interacted with unfamiliar objects, and executed complex tasks like industrial belt assembly — without internet support.

Apollo, notably, was able to generalize movement patterns and manipulate new objects in uncontrolled environments, a capability traditionally reserved for cloud-dependent systems.

Developer Access Now Available via SDK

To expand adoption, Google has released a Software Development Kit (SDK) for Gemini Robotics On-Device, allowing developers to integrate the model into compatible robotics platforms and build custom applications.

A Competitive AI Race in Robotics

As Google accelerates in the field of offline robotic AI, competitors like NVIDIA (with its Groot N1 model) and Hugging Face are also advancing efforts to bring generative AI to physical machines. The launch of Gemini Robotics On-Device signals a major turning point — where speed, security, and local intelligence converge to define the next generation of robotics.