New Gemini AI lets humanoid robots think and act without internet

Google DeepMind has introduced Gemini Robotics On-Device, a new AI model that enables humanoid robots to operate autonomously without internet connectivity. Unlike its cloud-dependent predecessor, this on-device version runs entirely on the robot, allowing for faster, low-latency responses and reliable performance in environments with poor or no connectivity. The model incorporates Gemini 2.0’s multimodal reasoning, natural language understanding, task generalization, and fine motor control, enabling robots to perform complex tasks such as unzipping bags and folding clothes. It is efficient enough to run locally with minimal data—requiring only 50 to 100 demonstrations to adapt to new tasks—and supports fine-tuning through teleoperation, making it highly adaptable across different robotic platforms. The Gemini Robotics On-Device model is designed with privacy and offline performance in mind, processing all data locally, which is particularly beneficial for security-sensitive applications like healthcare. Developers can access the model through Google’s trusted tester program and utilize a full software development kit