Google DeepMind представила Gemini Robotics 2 с интеллектом всего тела робота

Google DeepMind in July 2026 introduced Gemini Robotics 2 — an updated robot control system featuring whole-body intelligence: for the first time, the model gives a robot synchronous control over its entire body — legs, torso, and arms — within a unified motion planning loop.

What is whole-body intelligence

Whole-body intelligence means the robot coordinates the movements of its entire body as a single kinematic system, rather than controlling each link sequentially. The first version of Gemini Robotics, announced by Google DeepMind in January 2025 at CES, focused on dexterous manipulation — precise object grasping, working with fragile items, and tasks requiring fine motor skills of the hands. Gemini Robotics 2 extends this principle to the entire locomotion apparatus — with the goal of creating a robot that moves naturally rather than executing pre-programmed sequences.

The system simultaneously plans movements of the ankle, hip, spine, and arms — the way a human does when carrying a heavy box across an uneven floor or bending down to pick up an object from the ground. According to the Google DeepMind blog, the model accounts for the physical constraints of the environment in advance and redistributes load across joints without waiting for a balance error to occur.

Unlike traditional control systems that optimize each degree of freedom separately, Gemini Robotics 2 treats the body as a unified system with a shared balance of forces and impulses — enabling movements to be performed without explicitly programming each joint.

How is the system trained?

Gemini Robotics 2 is built on top of the Gemini multimodal architecture: the robot receives visual and text input, reasons about the given task, and in real time translates that reasoning into motor commands. For training, DeepMind used a combination of teleoperation — where a specialist controls the robot remotely, leaving a trace of correct decisions — and synthetic motion data. This made it possible to significantly reduce the number of real demonstrations needed to master a new task.

Key characteristics of the system according to the Google DeepMind announcement:

• Architectural foundation — the Gemini multimodal model
• A unified planning loop covering legs, torso, and arms
• Support for humanoid platforms and conventional manipulators
• Training via teleoperation and synthetic motion data
• Target scenarios — unstructured and variable environments

As stated in the official

Google DeepMind blog, Gemini Robotics 2 provides "whole-body intelligence," enabling the robot to plan and execute movements the way a living organism does — using all available motor resources.

Where is the new system applied?

Whole-body intelligence opens up a class of tasks inaccessible to narrow manipulators. Warehouse logistics on uneven surfaces, cleaning in residential spaces with variable layouts, assistance for people with disabilities, and servicing of production facilities — in all of these scenarios, the key limitation was precisely the coordination of the entire body. According to DeepMind, the system was tested in conditions as close to real-world as possible: on slippery surfaces, in rooms with obstacles, and while performing two-handed tasks involving asymmetric objects.

Google DeepMind is collaborating with robotics partners to test the system on physical humanoid platforms. The pace of development has accelerated: from the first to the second version of Gemini Robotics took approximately a year and a half, whereas previous research systems were updated far less frequently.

What this means

Gemini Robotics 2 advances physical AI from the stage of laboratory demonstrations to functional systems for the real world. In the whole-body robotics market, Google DeepMind competes with Figure AI, Boston Dynamics, and Tesla Optimus — but integration with Gemini gives robots a level of language and visual reasoning that purely motor-based systems currently lack. If the system proves reliable outside the laboratory, it could set a new standard for the industry.