Google DeepMind Unveils Gemini Robotics 2 for Full‑Body Humanoid Control

Google DeepMind announced Gemini Robotics 2, an AI brain built to give a single policy control over a robot’s legs, torso, arms and fingers, enabling coordinated whole‑body movement in a lab demonstration that showcased seamless transitions between actions.

Gemini Robotics 2 integrates visual perception and motor planning, letting the humanoid balance on uneven surfaces, walk forward while adjusting posture, and manipulate objects with its hands, all without separate controllers for each limb, marking a major advance for adaptable service robots.

Full‑Body Humanoid tests revealed the system can lift a 5 kg box while stepping over obstacles and recover balance after a gentle push, demonstrating the unified policy’s capacity to handle dynamic, real‑world tasks that previously required multiple specialized models.

Unified Policy approach cuts software complexity and shortens training cycles, as engineers train a single model rather than dozens of separate networks for each joint, potentially lowering development costs and speeding up deployment of future household and industrial robots.