For robots to help people in everyday situations, accurate spatial reasoning is not enough. Robots must also think quickly, make their decisions and reason at the speed of real time in the world.
That's why today we're introducing Gemini Robotics ER 2, our most capable model of “integrated thinking” robotics. Think of the Gemini Robotics ER 2 as a high-end robotics brain. It allows robots to communicate with humans, understand the virtual world, and plan multi-step tasks. It then assigns motor use to any low-level vision-language-action (VLA) model. Gemini Robotics ER 2 can also natively call tools like Google Search to find information, or any other user-defined task. The Gemini Robotics ER 2 design allows the robot to “think” next while performing its actions simultaneously.
Gemini Robotics ER 2 represents a significant improvement over Gemini Robotics ER 1.6. By watching continuous video feeds, robots can now track their progress, adapt if something goes wrong, and know exactly when to move on to the next step. We also introduce multi-robot collaboration, enabling robots to work together in shared environments and complete complex workflows that a single robot cannot do alone.
Gemini Robotics ER 2 is now publicly available to developers through the Gemini API, Google AI Studio, and in private preview on the Gemini Enterprise Agent Platform. To help you get started, we share examples of how to set up a model and tell it to enable very useful portable AI functions.
Improving the agent's physical capabilities
Many tasks in the physical world are complex and require many steps to complete. Gemini Robotics ER 2 is a mobile agent, which organizes the robot's steps and enables it to adjust itself, and integrates many new situations. To build an agent setup, developers can declare low-level control interfaces – such as Vision-Language-Action (VLA) models or navigation APIs – as tools, and stream multimodal video, audio, or text directly to the model.
Gemini Robotics ER 2 improves tool orchestration workflow. We can test its performance with robots in simulations, using real-world robot controls, and even pair it with a human controlling the robot remotely.
