DeepMind Unveils Gemini Robotics On Device
On 24 june 2025, Google DeepMind has launched “Gemini Robotics On Device” — a compact, powerful version of its flagship robotics model that runs entirely on a robot’s onboard systems, without the need for cloud connectivity.
This marks a major leap in robotic autonomy and flexibility, especially in offline or low-connectivity environments.
What’s new?
The flagship Gemini Robotics model is designed to help robots complete a wide range of physical tasks, even if it hasn’t been specifically trained on them.
It allows robots to generalize new situations and understand and respond to commands, as well as perform tasks that require fine motor skills.
Key Features
Offline Autonomy: On-device inference means the robot doesn’t need internet access to function, making it more secure and reliable in the field.
High Performance: Despite running locally, the model performs close to DeepMind’s original hybrid Gemini Robotics system, which utilized both cloud and onboard compute.
Generalization & Dexterity: Capable of performing complex and delicate tasks such as folding clothes or unzipping bags—without being trained explicitly on those tasks.
Fine motor skill execution: Capable of precision tasks like unzipping bags or folding clothes in demos.
How it works & Demos:
Trained initially on DeepMind’s ALOHA robot, it was successfully adapted to:
- Apptronik’s humanoid Apollo, The bi-armed Franka FR3 robot.
- With only 50–100 demonstrations, it grasped new tasks quickly.
Developer Tools & SDK
Alongside the model, DeepMind has also introduced an SDK—its first public developer kit for a Vision-Language-Action (VLA) model.
The SDK allows engineers and researchers to:
- Deploy Gemini On‑Device models on different robotic systems.
- Fine-tune or adapt the model for specific tasks.
- Run high-performance robotics AI entirely offline.
Availability:
Currently, the on-device model and SDK are in the hands of a select group of trusted testers, with safety evaluations still underway.
News Gist
On Tuesday, June 24, 2025, DeepMind unveiled Gemini Robotics On Device, empowering robots with real-time, offline perception and action.
With its SDK and performance close to cloud versions, this marks a major leap toward truly autonomous, versatile robots.
Baidu Open-Sources Its Ernie AI Model to Compete with OpenAI
Alibaba Launches AI Model Qwen-VLo
Meta, Anthropic Secure Legal Wins in AI Copyright Battles
Microsoft Launches “Mu” – A Smart Local AI for Your PC Settings
Google Launches Personal Styling App “Doppl”
