BREAKING: Google DeepMind Robots Now Work Without Internet
June 26, 2025, 11:30 AM, London — Google DeepMind today unveiled Gemini On-Device, a powerful AI model that lets robots function entirely offline. The launch marks a major shift in how autonomous systems operate, with faster decisions, better security, and zero cloud dependency.
DeepMind Brings Gemini AI Fully On-Device
Gemini On-Device is a compact version of DeepMind’s Gemini Robotics model. Unlike its cloud-hybrid predecessor, it runs entirely on local hardware. “We designed it to perform in real-world conditions where internet access is limited or impossible,” said Carolina Parada, DeepMind Robotics Lead.
The model supports advanced tasks like object recognition, manipulation, and adaptive behavior across multiple robot platforms.
📈Real-World Tasks Without the Cloud
Tests show the on-device model performing nearly as well as cloud-based systems. It requires only 50–100 demonstrations to learn complex skills such as tool usage or navigating unpredictable spaces.
“Gemini On-Device’s dexterity and reasoning are impressive for an offline AI,” said Dr. Arjun Kapadia, robotics researcher at Imperial College London.
🛡️ Security and Speed for Critical Industries
Cloudless AI offers privacy advantages in healthcare, defense, and industrial sectors. With zero data transmission to servers, Gemini On-Device helps meet strict data regulations while also reducing latency.
Manufacturers can deploy robots that react instantly — even in network dead zones — ensuring operational continuity.
🧰 SDK Launched for Developers Worldwide
DeepMind announced an SDK rollout for trusted partners. The toolkit allows developers to fine-tune, train, and deploy Gemini On-Device across various robots, including Apollo and Franka FR3.
The SDK is expected to spark rapid experimentation in fields like logistics, agriculture, and eldercare robotics.
FAQs About Gemini On-Device
Q1: What is Gemini On-Device?
A1: It’s a DeepMind AI model that runs directly on robots, enabling them to work offline.
Q2: Can Gemini On-Device replace cloud AI?
A2: It performs nearly as well but is best suited for secure, low-connectivity environments.
