Google DeepMind has developed Gemini Robotics On-Device, a new artificial intelligence model to empower robots. Its key feature is the ability to operate without requiring an internet connection, allowing robots to perform tasks without relying on the cloud or Wi-Fi. This AI model grants bi-arm robots the ability to process speech, understand language, and execute actions. The system operates with low latency, making it ideal for real-time applications. It enables robots to understand and act on human instructions. The AI model has demonstrated success in internal tests, including tasks such as folding clothes, opening and closing bags, recognizing and manipulating objects, and industrial belt assembly. The model can understand natural language instructions and execute them effectively. The system has been tested on multiple robotic platforms, including ALOHA, Franka Emika FR3, and Apptronik Apollo humanoid robots. At present, the system is available only to select developers within Google’s Trusted Tester program, allowing them to explore the Gemini Robotics SDK and MuJoCo physics simulator.
