Google DeepMind has unveiled Gemini Robotics 2, the latest generation of its robotics AI platform aimed at making robots more capable of performing complex physical tasks. The new release expands the company’s family of robotics foundation models with improved motion control, reasoning and task planning, further strengthening Google’s position in the growing market for embodied AI. Unlike traditional robot software that relies on predefined motion sequences, Gemini Robotics 2 is designed to reason through every movement while executing a task. According to Google, the system combines low-level motor control with higher-level planning, allowing robots to adapt their actions to changing situations instead of simply replaying programmed behaviors.
The platform consists of multiple models. Gemini Robotics 2 serves as the vision-language-action model that generates robot movements, while Gemini Robotics-ER 2 focuses on embodied reasoning, enabling robots to understand their surroundings, plan multi-step actions and solve spatial problems before acting. Google said the reasoning model is now available through Google AI Studio and in private preview on the Gemini Enterprise Agent Platform, while the motion-control models are being released to selected early-access partners.
Google says the new models improve whole-body intelligence, enabling robots to coordinate their arms, hands and body more effectively when manipulating objects or interacting with people. The company expects the technology to support a broad range of applications, from warehouse automation and manufacturing to household assistance and service robotics.
The announcement marks the next step in Google’s robotics strategy, which began with the introduction of Gemini Robotics and Gemini Robotics-ER in 2025. Since then, the company has steadily expanded its portfolio with on-device models, developer tools and reasoning capabilities aimed at making the same AI models usable across different robot platforms.
Rather than developing its own humanoid robot, Google DeepMind is positioning itself as a supplier of foundation models that can be integrated into hardware from multiple robotics manufacturers. The company has previously worked with partners including Boston Dynamics, Agility Robotics and Agile Robots to evaluate and develop its robotics models.
The launch also intensifies competition in the emerging Physical AI market. Technology companies are increasingly focusing on general-purpose AI models that can power a wide variety of robots instead of creating software for individual machines. Google joins a growing group of companies—including NVIDIA with its Isaac and GR00T platforms—that are seeking to provide the AI foundation for the next generation of autonomous robots.
