Google DeepMind launched Gemini Robotics 2, a general-purpose robot model focused on "whole-body intelligence," just over a week after the WAIC exhibition. At the event, more than 200 robot-related enterprises and over 300 real machines competed, with manufacturers aiming to prove robots can work and handle diverse tasks. The article states the real bottleneck in robotics is the "brain": making robots perceive chaotic environments, decide actions, and transmit decisions to dozens of joints. The release includes three models: Gemini Robotics ER 2 for understanding the environment and planning tasks, Gemini Robotics 2, a VLA model converting vision and language into actions, and On-Device 2, which compresses the action model locally and adapts to new robot entities with less than 200 samples and several hours of data. The core advancement is using a set of models to coordinate a humanoid robot's legs, torso, arms, and fingers, allowing it to think and reason while moving, rather than performing segmented actions like standing still before reaching out.