Gemini Robotics 2: Google DeepMind учит ИИ управлять гуманоидными роботами

In July 2026, Google DeepMind introduced Gemini Robotics 2 — an updated version of the AI model designed to control humanoid robots in the physical world. According to Wired, this is "a significant leap toward physical AGI" — artificial intelligence that acts not only in digital space, but also physically interacts with objects and people.

What can Gemini Robotics 2 do?

The key idea of the model is a unified architecture for three tasks: perception (computer vision and spatial understanding), interpretation (processing natural language instructions), and action (generating commands for motor mechanisms). Traditional robotic systems solve these three tasks with separate modules; Gemini Robotics 2 combines them in a single multimodal model.

This means an operator can give the robot a voice command — for example, "pick up the red object from the table" — and the model will independently build the chain from visual object recognition to a precise grasping movement.

• Developer — Google DeepMind
• Platform — humanoid robots
• Architecture — unified multimodal model (vision + language + motor control)
• Positioning — the next step toward "physical AGI"

The predecessor, first-generation Gemini Robotics, established the concept of "embodied AI" — systems capable of acting in a three-dimensional physical environment, rather than just processing text. Gemini Robotics 2 develops this concept, expanding the range of solvable tasks and improving the precision of control over complex humanoid platforms.

Why is physical AI more dangerous than digital?

Transferring AI from digital space to the physical world creates a fundamentally different risk profile. A language model instantly corrects a failed response — a robotic one does not: an incorrect manipulator movement means a broken object, a spilled substance, or real danger to people nearby.

"Deploying AI in the real world carries risks,"

Wired states, listing the technical and ethical challenges that arise alongside Gemini Robotics 2.

Among the key issues are the model's behavior in situations not encountered in training data; the question of legal liability in case of an error; and the safety of physical interaction with people. These are precisely the questions that make physical AGI a task not only of engineering, but also of regulation.

Google DeepMind is not the only company in this race. Tesla is developing the humanoid robot Optimus, Figure AI has attracted more than $675 million in investment, and Boston Dynamics is advancing Atlas. However, Gemini Robotics 2 arrives with a fundamental distinction: at its core is the multimodal Gemini model, rather than a disparate set of control modules loosely connected to one another.

What is happening in the robotics market

The humanoid robot market is on the verge of large-scale growth. According to Goldman Sachs forecasts, by 2035 its volume could exceed $150 billion — provided the industry solves key problems of reliability and safety. The center of competition is shifting from mechanical precision to intelligence: the ability to understand unstructured tasks, adapt to changing conditions, and work alongside people without physical barriers.

This is exactly the segment Gemini Robotics 2 is designed for — a system with "general" intelligence in a physical body, rather than a narrowly programmed industrial manipulator that can perfectly execute one movement but is helpless in a non-standard situation.

What this means

The launch of Gemini Robotics 2 confirms: leading AI laboratories view physical robotics as the next technological frontier. For business — this is a horizon of new applications in logistics, manufacturing, and service sectors. For regulators — an urgent task: to develop safety standards for AI systems whose actions have irreversible physical consequences.