Google DeepMind has announced the release of Gemini Robotics 1.5 and Gemini Robotics-ER 1.5, advanced AI models designed to significantly improve robot autonomy, adaptability, and reasoning capabilities. The announcement, made on September 25, 2025, marks a significant advancement in intelligent automation, enabling machines to operate more effectively in complex physical environments.
Gemini Robotics 1.5 is a sophisticated vision-language-action (VLA) model that interprets visual input and user instructions to generate precise motor commands. This allows robots to perform intricate tasks, such as sorting laundry by color, and crucially, enables the transfer of learned motions across different robot embodiments. This cross-embodiment learning capability accelerates robot development by reducing the need for extensive retraining for new hardware configurations.
Complementing the VLA model is Gemini Robotics-ER 1.5, an embodied reasoning (ER) model specializing in understanding physical spaces and planning multi-step tasks. This model can access external tools, including Google Search, to gather information and utilize user-defined functions, facilitating more robust problem-solving in dynamic situations. Its ability to perform spatial reasoning and plan actions from natural language commands represents a major step towards more intelligent robotic systems.
The two models are designed to work synergistically, with the ER model orchestrating high-level task planning and decision-making, while the VLA model executes the physical actions. This collaborative architecture aims to foster greater generalization across a wide array of tasks and environments, moving beyond robots that merely react to commands to systems that can reason, plan, and adapt proactively. Gemini Robotics-ER 1.5 is now available to developers via the Gemini API in Google AI Studio, while Gemini Robotics 1.5 is currently accessible to select partners.