Hello to all my tech-loving followers! Today, we’re diving into an exciting development in robotics and artificial intelligence. Google DeepMind has rolled out a new language model called Gemini Robotics On-Device, which can run directly on robots without needing the internet.

This new model builds on their previous Gemini Robotics, introduced earlier this year. It’s designed to control robot movements better and can be customized by developers using natural language prompts. Interestingly, its performance is nearly on par with the cloud-based version, but it shines because it works locally.

Google showcased demos where robots unzip bags and fold clothes using this on-device AI. Specifically, the model was trained for ALOHA robots but has been adapted to work with other machines like the Franka FR3 bi-arm robot and the Apollo humanoid by Apptronik. The Franka robot successfully handled unfamiliar tasks such as industrial belt assembly.

To support developers, Google is also releasing the Gemini Robotics SDK, allowing them to train robots using demonstrations in the MuJoCo physics simulator. Other tech giants, like Nvidia and Hugging Face, are exploring robotics as well, with Nvidia creating platforms for humanoids and Hugging Face working on models and actual robots.

In summary, Google’s latest model is a giant step toward smarter, more autonomous robots that can operate without constant internet access. This opens up new possibilities for robotics applications across industries, from industrial automation to service robots.