A new AI model from Google DeepMind, called Gemini Robotics On-Device, is specifically designed to enhance the capabilities of robots. The primary advantage of this model is that it can operate without an internet connection. This means robots can now function effectively without needing constant access to the cloud or Wi-Fi. The AI model is built to enable bi-arm robots with speech, language processing, and action capabilities. The low latency of the system makes it suitable for immediate use in real-time applications. This advancement allows robots to understand and react to human commands in a manner akin to humans. In tests, the AI model successfully managed a variety of tasks, including folding clothes, opening and closing bags, recognizing and manipulating objects, and carrying out precision tasks such as industrial belt assembly. The model can interpret human instructions provided in natural language and perform the appropriate actions. It has been tested on various robotic systems, including ALOHA, Franka Emika FR3, and Apptronik Apollo humanoid robots. Currently, access to this system is limited to select developers through Google’s Trusted Tester program, where developers can experiment with the Gemini Robotics SDK and MuJoCo physics simulator.
