Google DeepMind Unveils Gemini Robotics AI Models for Real-World Robot Control

Google DeepMind introduced Gemini Robotics and Gemini Robotics-ER, two AI models designed to control robots in real-world tasks. Gemini Robotics, built on Gemini 2.0, features a "physical actions" output for direct robot control, demonstrating generality, interactivity, and dexterity. It adapts to new situations, understands conversational language, and performs complex tasks like folding paper. Gemini Robotics-ER focuses on spatial reasoning, enabling robots to manipulate objects safely, such as grasping a coffee mug by its handle. These models are currently not public, with DeepMind planning to integrate them into humanoid robots for further evaluation.