In a major leap for autonomous robotics, Google DeepMind has launched Gemini Robotics On-Device, a compact AI model that enables robots to operate without internet connectivity, revolutionizing how machines function in real-world, low-connectivity environments.
Offline, Yet Exceptionally Capable
Unlike its cloud-based predecessor, the on-device model runs entirely on a robot’s internal hardware, allowing it to process vision, language, and action inputs locally. Despite its smaller size, Gemini Robotics On-Device can perform complex tasks like folding clothes, unzipping bags, and assembling parts, even in remote or secure zones.
“It’s small and efficient enough to run directly on a robot,” said Carolina Parada, Head of Robotics at Google DeepMind.
Fast Learning and Cross-Platform Flexibility
The model can learn new tasks from just 50 to 100 demonstrations, thanks to its multimodal reasoning and task generalization capabilities. Initially trained on Google’s ALOHA robot, it has since been adapted to platforms like the Apptronik Apollo humanoid and Franka FR3 dual-arm system.
Ideal for Privacy-Sensitive Applications
By processing all data locally, Gemini Robotics On-Device enhances data security and operational reliability, making it ideal for healthcare, industrial automation, and military-grade robotics. It also eliminates latency issues common in cloud-based systems, ensuring real-time responsiveness.
Developer Access and Safety Measures
Google is offering a Gemini Robotics SDK and MuJoCo simulator to select developers for testing and fine-tuning. While the on-device model lacks built-in semantic safety tools, developers are encouraged to integrate Gemini Live APIs and low-level safety controllers to ensure responsible deployment.
🔁 Share this article to explore how Google’s Gemini AI is redefining the future of offline robotics and intelligent automation.
