Introducing Gemini Robotics, a groundbreaking vision-language-action (VLA) model based on Gemini 2.0 with advanced spatial understanding for robotic applications. Gemini Robotics is designed for general, interactive, and dexterous tasks, surpassing state-of-the-art models in performance. The model can adapt to new situations, respond to natural language commands, and exhibit precise manipulation skills like folding origami. Gemini Robotics-ER enhances world understanding by focusing on spatial reasoning, enabling robots to perform complex tasks with improved success rates. Safety is a top priority, with measures in place to ensure responsible development and evaluation of robotic actions to align with human values. Trusted testers include companies like Boston Dynamics and Agility Robots to further explore the capabilities of these revolutionary AI models in real-world applications.
https://deepmind.google/discover/blog/gemini-robotics-brings-ai-into-the-physical-world/