Google DeepMind has introduced Gemini Robotics On-Device, an advanced artificial intelligence model that allows robots to operate entirely offline, without the need for an internet connection. This development marks a significant step forward in robotics, enabling machines to perform complex, real-time tasks in environments where connectivity is limited or privacy is paramount.

​

Key Features

• offline operation: Robots powered by Gemini Robotics On-Device can function in real time without relying on cloud servers, making them ideal for latency-sensitive applications and remote or secure locations

• natural language understanding: The AI model can interpret and follow spoken commands, handle multi-step instructions, and interact naturally with users, bridging the gap between human intent and robotic action

• cross-platform compatibility: Gemini Robotics On-Device is designed to work seamlessly across different robotic platforms, including ALOHA, Franka FR3, and Apollo humanoid robots, and can adapt to new hardware with minimal retraining

• sdk available: Developers can quickly fine-tune the model for specific tasks using only 50 to 100 demonstrations, making it accessible and customizable for a wide range of applications

​

How Gemini Robotics On-Device Works

Gemini Robotics On-Device is built on a vision-language-action (VLA) architecture, enabling robots to process visual, audio, and textual information locally. This means robots can understand their environment, interpret commands, and execute actions instantly, without the delays associated with cloud processing. The model is compact and efficient, able to run on a robot’s onboard GPU or processor, yet still delivers robust performance close to that of cloud-based systems.

​

The AI enables robots to perform a wide range of dexterous tasks, such as folding clothes, unzipping bags, assembling objects, and picking up items, all through local processing. It has demonstrated adaptability by learning new tasks after seeing only a handful of examples, even on robot types it was not originally trained on. This rapid learning ability is crucial for real-world deployment, where robots often encounter unfamiliar objects and scenarios.

​

Industry Impact and Use Cases

The introduction of Gemini Robotics On-Device is expected to accelerate the adoption of smart robots in industries like manufacturing, logistics, healthcare, and home automation. In factories, robots can now assemble products, handle materials, and operate around the clock without connectivity concerns. In homes, they can assist with chores and respond to voice commands instantly and privately. The technology is also well-suited for sensitive environments, such as hospitals and secure facilities, where data privacy and reliability are critical.

​

By processing all data on-device, the model enhances security and privacy, ensuring that sensitive information never leaves the robot. This is particularly important for companies with strict data regulations or those operating in areas with unreliable networks.

​

​

Developer Access and Future Prospects

Google has released a software development kit (SDK) for Gemini Robotics On-Device, initially available to select developers and trusted testers. This SDK allows for rapid prototyping and customization, enabling companies and researchers to adapt the AI to their specific robots and use cases. The open-source MuJoCo simulator and benchmarks for bimanual dexterity are also being made available to broaden access for academics and developers.

​

Looking ahead, Google DeepMind plans to expand the capabilities of Gemini Robotics On-Device, bringing even greater dexterity, reasoning, and autonomy to robots. The move toward edge AI in robotics signals a new era of smarter, faster, and more secure machines that can operate independently in the real world.

​

​

Source: Google DeepMind
Date: June 25, 2025