Google DeepMind's Gemini Robotics On-Device Empowers Robots Offline
Google DeepMind has unveiled Gemini Robotics On-Device, a pioneering language model that empowers robots to execute tasks offline. This innovation marks a significant step in robotics, enabling local, autonomous operation without internet connectivity.
Initially developed for ALOHA robots, the model was later adapted for the bi-arm Franka FR3 and Apollo humanoid robots. Remarkably, the Franka FR3 demonstrated its versatility by successfully performing assembly tasks beyond its training scope. Gemini Robotics On-Device has proven superior in general benchmarks, outperforming other on-device models, as per Google's claims.
To facilitate further advancements, Google DeepMind is providing a Gemini Robotics SDK. This tool allows trainers to teach robots new tasks using the MuJoCo physics simulator. Notably, the first public mention of Gemini-related on-device technology dates back to late September 2025, with the release of Gemini 2.5 Pro. The model's performance in benchmarks is on par with its cloud-based Gemini Robotics counterpart, ensuring high-quality output even offline. Gemini Robotics On-Device controls robot movements using natural language prompts, opening up new possibilities for specific applications.
Google DeepMind's Gemini Robotics On-Device is transforming robotics by enabling offline task execution. With its impressive performance and the upcoming SDK, this model promises to unlock new capabilities for robots, making them more versatile and accessible than ever before.
