Google DeepMind unveils Gemini Robotics 2 for robot control

Google DeepMind has unveiled Gemini Robotics 2, a major update to its artificial intelligence system designed to give robots whole-body control, advanced dexterity, and the ability to collaborate with other machines. Announced on July 30, 2026, the system represents a significant step toward creating general-purpose robots that can perform complex physical tasks in human environments.

The core advancement is the expansion of control from robotic arms to entire humanoid bodies. Earlier models were limited to tabletop tasks, but Gemini Robotics 2 can now coordinate a robot's legs, torso, arms, and hands to navigate cluttered spaces. In a demonstration, an Apptronik Apollo 2 humanoid robot successfully followed a single instruction to place a watering can on a specific shelf, requiring it to walk, pick up the object, and bend to complete the task.

The system is built on three specialized AI models working together. A Vision-Language-Action (VLA) model handles the physical execution of movements. An upgraded Embodied Reasoning model, Gemini Robotics ER 2, acts as a high-level planning brain, capable of processing live video feeds to track progress, adjust to failures, and manage...