Google DeepMind has unveiled Gemini Robotics, a new AI model based on its Gemini 2.0 technology, designed to bring artificial intelligence into the physical world. This groundbreaking development focuses on “embodied” reasoning, the ability of robots to understand and respond to their surroundings similarly to humans.
Gemini Robotics is an advanced vision-language-action (VLA) model that integrates physical actions to control robots through improved spatial understanding. Alongside it, Gemini Robotics-ER enables roboticists to utilize embodied reasoning capabilities for various programs.
These innovations allow robots to perform a wider range of tasks than ever before, making them more adaptable and interactive. Gemini Robotics exemplifies generality, interactivity, and dexterity, crucial for effective human-robot collaboration.
The model excels in adapting to new situations and objects while understanding commands given in natural language. Additionally, Gemini Robotics can perform complex tasks requiring fine motor skills, which is a significant step toward creating general-purpose robots.
The advanced Gemini Robotics-ER elevates spatial reasoning, enabling robots to grasp and manipulate items safely and efficiently. This model improves upon Gemini 2.0’s capabilities, achieving a significant increase in success rates for controlling robots.
Safety remains a primary concern, with a layered approach to robotics safety that includes collaboration with various partners to ensure responsible development. The initiative includes the release of a new dataset aimed at enhancing semantic safety in robotics, helping to align robotic actions with human values.
As Gemini Robotics is tested and refined with the input of industry leaders like Apptronik, Agile Robots, and Boston Dynamics, Google DeepMind looks forward to advancing AI’s role in robotics, ultimately making robots more useful and helpful in everyday life.
