Google’s DeepMind has unveiled Gemini Robotics On-Device, a powerful new large language model (LLM) designed to run entirely offline on robotic platforms. Launched on June 24, the model is tailored for latency-sensitive and low-connectivity environments, enabling robots to function autonomously without relying on the cloud. This advancement ensures faster response times, enhanced data privacy, and improved reliability in real-world settings.
Gemini On-Device Brings Cloud-Level Intelligence to Offline Robots
Unlike traditional cloud-based AI systems, Gemini On-Device is optimized for low-compute hardware yet retains advanced capabilities such as natural language understanding, multi-step task execution, and complex manipulation. Google claims the model can handle delicate tasks like folding clothes, unzipping bags, and assembling mechanical parts — all processed locally without cloud support. Remarkably, it delivers performance comparable to cloud models in accuracy and versatility, making it a game-changer for mobile and industrial robotics.
Originally developed on ALOHA robots, Gemini’s compatibility now extends to sophisticated platforms such as the Franka FR3 and Apollo humanoid robot. These systems have successfully demonstrated voice-command execution, interaction with unfamiliar objects, and industrial tasks like belt assembly — showcasing Gemini’s adaptability across robotic platforms.
Google Empowers Developers with Gemini SDK as AI Race in Robotics Heats Up
To foster innovation, Google has released an SDK (software development kit) for Gemini Robotics On-Device, allowing developers to experiment with and deploy the model across compatible machines. The SDK opens up opportunities for developers to bring AI-powered autonomy to diverse robotic applications without needing constant internet access.
As Google pushes forward in this domain, competitors like NVIDIA (with Groot N1) and Hugging Face are also accelerating efforts in generative AI for robotics, signalling a broader shift in the industry. With Gemini Robotics On-Device, Google is positioning robots to be faster, smarter, and fully independent of the cloud — a pivotal leap for the future of autonomous systems.
Summary:
Google’s DeepMind has launched Gemini Robotics On-Device, an AI model that runs entirely offline on robots, ensuring low-latency performance and privacy. Optimized for low-compute hardware, it supports complex tasks and voice commands. With an SDK for developers and rivals like NVIDIA entering the space, robotics AI is evolving rapidly.
