Google DeepMind has introduced an upgrade to Gemini Robotics with the new AI model called Gemini Robotics On-Device. This model runs directly with the assistance of robots, without an internet connection. The new model is an extension of the earlier Gemini Robotics model, launched in March 2025.
Using this model, robots can be called upon to perform tasks through natural language commands. The primary function of the model is to expedite functional performance through robots and enable smoother operation.
Google claims that the Gemini Robotics On-Device model operates almost as well as cloud models in every respect. The company stated that it had outperformed other local AI models during the testing phase. However, Google chose not to reveal the other models that were outperformed by Gemini Robotics On-Device.
In live demos, the robot was presented with tasks, including unzipping bags and folding clothes, where it quickly adapted to new objects and tasks.
The model was originally developed for ALOHA robots, but now it is working with other types of robots like the bi-arm Franka FR3 and Apptronik's Apollo. FR3 robot used the model to perform assembly tasks it had never encountered before. This means that the model can be of practical use in the real world to a great degree.
Google has also released a Gemini Robotics SDK. The SDK assists developers in training robots using just 50 to 100 task demos. Training is conducted in the MuJoCo physics simulator.
Developers can make use of simple-language prompts to control and fine-tune the robots. This facilitates the building of robotic skills more easily and more quickly.
Other corporations are venturing into robotics as well. Nvidia is developing AI for humanoid robots. Hugging Face is building open AI models and tools for robotics. Korean startup RLWRLD is also working on robot AI. The race for the advancement of AI robotics is picking up speed on a global scale.
Gemini Robotics On-Device represents the major advancement in robotics utilizing AI. Actions built for processing in the device will be faster, more secure, and more private.
This model might just revolutionize the use of robots in homes, factories, and public spaces. It pushes the idea of intelligent, autonomous robots even closer to reality.