Google DeepMind has released the Gemini Robotics On-Device AI, a groundbreaking system that allows robots to perform complex tasks without requiring internet connectivity. This new innovation offers enhanced privacy, reliability, and adaptability for real-world applications, marking a significant step forward in robotics technology.
Unlike its cloud-based predecessor, the on-device version of Gemini Robotics runs entirely on the robot itself, eliminating the need for internet access. This makes it ideal for environments where connectivity is unreliable or unavailable, such as remote or isolated areas, or in applications requiring stringent security measures like healthcare. The system is capable of understanding natural language, performing fine motor tasks, and adapting from very little data, without the need for an internet connection.
Cariola Parada, head of robotics at Google DeepMind, highlighted the efficiency of the new system, stating that it is ‘small and efficient enough’ to operate directly on the robot. This innovation allows developers to use the model in situations where internet connectivity is limited, without compromising on performance or flexibility.
Developers can customize the model with just 50 to 100 demonstrations, and the model has already been adapted to various robotic platforms, including Apptronik’s Apollo humanoid and the Franka FR3. Google is providing access to developers through its trusted tester program and has released a full SDK to support experimental and development activities.
The on-device model’s data processing remains local, ensuring better privacy for sensitive applications. This approach also ensures uninterrupted robotic operation during internet outages or in isolated environments. Google envisions this version as highly suitable for remote, security-sensitive, or infrastructure-poor settings, offering faster response times and fewer points of failure.
Despite the absence of built-in semantic safety features, Google recommends developers integrate their own safety systems using tools like the Gemini Live API and trusted low-level controllers. The company is restricting access to select developers to better study safety risks and real-world applications. Although the hybrid model offers more overall power, this version holds its own for most common use cases and is pushing the boundaries of robotics towards everyday deployment.
The release of the Gemini Robotics On-Device represents a turning point in robotics, as robots are no longer reliant on a constant cloud connection to be smart, adaptive, and useful. With faster performance and stronger privacy, these systems are now equipped to tackle real-world tasks in environments where traditional robots might struggle.