Google has made Gemini Robotics Embodied Reasoning 2, or ER 2, available in public preview on the Gemini platform. The model helps robots plan tasks, interpret what they see and coordinate the tools that control their movements.
ER 2 acts as a high-level reasoning layer. Developers can connect it to lower-level controls, such as navigation APIs or models that turn visual input and language into actions. They can stream video, audio or text to the model and give operators a way to direct robots in natural language, according to Google.
The company says ER 2 can check whether a step succeeded before a robot continues, then adjust if it did not. It can also identify when an event occurs in video and read instruments from visual input. Google says the model supports more than 10 instrument categories, including dials, digital displays and liquid thermometers.
The model coordinates different kinds of robots
Google says the same model can coordinate different kinds of robots, from stationary arms to wheeled rovers and quadrupeds. It can help them divide work and hand tasks to one another.
For enterprise teams, Google describes a workflow that starts with testing robot behavior in simulators such as MuJoCo-Warp and NVIDIA Isaac Sim. Teams can use ER 2 to label demonstration videos, train custom robot action models and manage workflows on the Gemini platform. Google also says local tasks such as mapping and path planning can run on factory floors through Google Distributed Cloud.
ER 2 is available through Model Garden
Developers and enterprise teams can start building with ER 2 now through Model Garden on the Gemini platform. It remains in public preview.
