Google Launches On-Device Model Gemini Robotics

Google has launched Gemini Robotics On-Device, marking the first deployment of its VLA (Vision-Language-Action) model on robotic devices. This model is capable of performing efficient bimanual tasks, such as folding clothes and opening bags, and can quickly learn new skills through 50–100 demonstrations. Gemini Robotics On-Device also features strong task generalization capabilities and low-latency inference performance, making it suitable for various robot form factors, including robotic arms and humanoid robots. Additionally, Google has provided developers with the Gemini Robotics SDK to support model fine-tuning and evaluation, further advancing innovation and application in robotics technology.