Disconnected but Powerful: Google’s Gemini AI Operates Seamlessly Offline!

Google DeepMind has unveiled its new AI model, Gemini Robotics On-Device, designed specifically for robotics and capable of operating independently on physical robots without cloud connectivity. Announced in a recent blog post, this model excels in managing robotic movements and understanding human language prompts, making it ideal for environments with limited or no internet access. Building on its earlier Gemini Robotics model from March 2025, the updated version works efficiently with minimal computing resources, enabling dual-arm robots to perform intricate tasks like folding garments and unzipping containers. Comparative tests showed that Gemini Robotics On-Device outperformed other local AI models in executing multi-step tasks and adapting to unfamiliar situations. Initially trained on ALOHA robots, it was successfully tested on Franka FR3 and Apollo robots. This model is available for developers via an SDK for integration across various platforms, highlighting a significant advancement in robotics AI amid growing competition in the sector.

Source link