With Gemini Robotics On-Device, Google introduces an AI model that can run locally on robots without the need for an internet connection.
Google showcases Gemini Robotics On-Device. It’s an evolution of Gemini Robotics and Robotics-ER. As the name suggests, the model can run on the robot itself, without requiring an internet connection.
The AI model can control the movements of a robot. Developers can give the model instructions in plain language to make the robot perform in various scenarios. Google claims that the results of the On-Device version are in practice equivalent to those with Gemini Robotics in the cloud. It also states that the new model outperforms the competition, although it doesn’t name that competition specifically.
Complex Tasks
Gemini Robotics On-Device allows robots in a production environment to perform complex tasks, such as unzipping bags or folding clothes. The AI model was initially developed for the Aloha robotic arm, but also works on the dual-arm Franka FR3 and even the more human-like test robot Apollo.
Google claims that the built-in intelligence enables hardware to perform new tasks for which the devices were not specifically tested, such as assembly on a production line.
The combination of AI and robotics represents a next step in the deployment of the technology. It has the potential to make robotics much more ubiquitous by ensuring that robots can also perform complex variable tasks.