Google DeepMind unveils Gemini Robotics 2 for whole-body robot control

Gemini Robotics 2 has been launched by Google DeepMind, featuring a collection of AI models aimed at equipping robots with comprehensive body intelligence, precise dexterity, and the ability to collaborate with multiple robots. This suite comprises three distinct models: Gemini Robotics 2, a vision-language-action (VLA) model that governs full humanoid movements; Gemini Robotics ER 2, which focuses on embodied reasoning for complex task planning and interaction with humans; and Gemini Robotics On-Device 2, a streamlined VLA tailored for local use. The models can quickly adjust to new robot forms within hours using minimal data. Showcased on Apptronik's Apollo 2 and the Franka Duo, the system efficiently performs tasks such as cleaning and packing. Safety measures include the ASIMOV-Agentic benchmark and improved human proximity detection. Gemini Robotics ER 2 is accessible on Google AI Studio and in private preview on the Gemini Enterprise Agent Platform, while the VLA and On-Device models are available to select early-access partners.