Google launches its first on-device embodied intelligence model: Gemini Robotics On-Device

Google DeepMind has unveiled the first on-device embodied AI model, Gemini Robotics On-Device. This model can run directly on robots without relying on cloud connectivity, offering robust offline operation capabilities. It is capable of following natural language instructions to perform delicate tasks such as opening bags and folding clothes. The model supports deployment across multiple robot platforms, features low response latency, and can quickly adapt to new tasks—learning new actions with just 50 to 100 demonstration samples.