Google Deepmind on Tuesday released a new Gemini Robotics Artificial Intelligence (AI) model that could run perfectly on a local device. Dubbed Gemini robotics on-device, it is a voice-language-action (VLA) model that can make the robot a wide range of tasks in the real-world environment. Mountain view-based tech veteran said that since the AI model acts without the need to be associated with the data network, it is more useful for applications that are sensitive. Currently, the model is available to those who have signed up for their reliable examiner program.
Google’s new robotics model runs perfectly on the device
In a blog post, Carolina Parda, a senior director at Google Deepmind and head of robotics, announced the release of on-devices to Gemini robotics. The new VLA model can be accessed through a Gemini robotics software development kit (SDK) after signing up for its examiner program. The model can also be tested on the company’s Muzoko Physics Simulator.
Since it is a proprietary model, the details about its architecture and training methods are not known. However, Google has highlighted its abilities. The VLA model is designed for B-Arme robot and has minimal computational requirements. Despite this, the model allows for experiment, and the company claims that it can be compatible with new tasks only with 50 to 100 performances.
Gemini robotics on-device also follows natural language instructions and can perform complex functions such as unzipping bags or folding clothing. Depending on the internal test, the tech giants claim that the AI model “performs strong normalization performance during the local running.” Additionally, it is also asked to improve other on-device models on “more challenging out-of-disorder functions and complex multi-phase instructions”.
In particular, Google highlighted that while the AI model was trained for the Aloha robot, researchers were also able to adapt it to Franca FR3 and Apptronik’s Apollo Humanoid Robot. All these are B-Arm robots, the only configuration that is compatible with Gemini robotics on-device.
The AI model was able to follow the instructions on all different robots and perform normal functions. Additionally, it can also handle already unseen objects and scenarios, and even execute the industrial belt assembly functions, which requires high level of accuracy and dexterity, the company claimed.