Google's new Gemini AI can power robots and make them work without internet
Google DeepMind has unveiled the latest version of its Gemini Robotics AI model, known as Gemini Robotics On-Device, allowing robots to function independently without the need for an internet connection. This development is aimed at enhancing the speed, autonomy, and privacy of robots in real-world scenarios.
Benefits of Gemini Robotics On-Device
Unlike previous models that relied on cloud connectivity, Gemini Robotics On-Device empowers robots to process information and make decisions on the device itself. This revamp is a game-changer for robotics, especially in offline environments such as remote areas, secure facilities, and situations where low latency is critical.
Features of Gemini Robotics On-Device
Gemini Robotics On-Device introduces several innovative features. The model can perform tasks out of the box and learn new ones with minimal demonstrations, ranging from 50 to 100. Initially trained using Google's ALOHA robot, it has now been adapted for use with other robotic systems like Apptronik's Apollo humanoid and Franka FR3 with dual arms.
Google asserts that the model can execute intricate actions such as folding clothes and unzipping bags while running low-latency inference on the device itself. This stands in contrast to Tesla's humanoid robot, Optimus, which relies on internet connectivity for data processing.
Privacy and Security
A notable feature of Gemini Robotics On-Device is its local data processing, making it ideal for privacy-sensitive applications such as healthcare and industrial automation. This approach ensures data security by processing all information locally, without the need for cloud connectivity.
Despite its advantages, the on-device model does not include built-in semantic safety tools like its cloud-based counterpart. Google suggests that developers implement their safety protocols, including leveraging Gemini Live APIs and connecting to low-level safety controls.
This release follows Google's introduction of the AI Edge Gallery, an Android app allowing users to run AI models offline on their smartphones. Powered by the compact Gemma 3 1B model, the app enables users to perform tasks like generating images and interacting with AI tools directly on their devices, without internet dependency.
Similar to Gemini Robotics On-Device, AI Edge Gallery prioritizes privacy and low-latency performance, utilizing open-source models and technologies like TensorFlow Lite for seamless user experiences across various devices.
