Google DeepMind launched Gemini Robotics 2, a general-purpose robot model focused on "whole-body intelligence," just over a week after the WAIC 2026 exhibition. At the event, more than 200 robot-related enterprises and over 300 real machines competed, including Unitree's 3-meter-tall manned transformable mech and Agibot's robot using tweezers to pick up a 2mm resistor and weld it onto a circuit board. Manufacturers are increasingly proving that robots can work and perform more types of labor, but the real bottleneck lies in the "brain"—enabling robots to understand chaotic environments, decide next steps, and transmit decisions to dozens of joints. The new release includes three models: Gemini Robotics ER 2 handles environment understanding, long-task planning, and human communication; the Gemini Robotics 2 VLA model translates vision and language into actions; and On-Device 2 compresses the action model to the local end, adapting to new robot bodies with less than 200 samples and data of several hours. The core advancement is using a set of models to coordinate the legs, torso, arms, and fingers of a humanoid robot, allowing it to think and reason while moving and operating, eliminating the need to "stand still first, then reach out."