Google DeepMind Updates AI to Control Full Humanoid Robot Bodies

Google DeepMind has announced a major update to its robotics technology that allows artificial intelligence to control the entire body of a humanoid robot. According to The Verge, the new model, called Gemini Robotics 2, marks a shift from previous versions that could only manage a robot’s upper body. Now, the AI can coordinate movements from a robot’s feet to its fingertips, enabling machines to walk, crouch, and stretch with much better control.

This development is an important step forward for the robotics industry. As previously reported, the race to improve robot control and computing power has been a major focus for tech companies throughout 2026. While earlier models were limited in their physical reach, Gemini Robotics 2 allows robots to perform more complex, real-world tasks. Videos released by Google show the Apptronik Apollo 2 robot performing actions such as bending over to pick up a watering can and finding specific items on a shelf.

Beyond basic movement, the update brings significant improvements to dexterity—the ability to use hands and fingers skillfully. The new model can control complex, five-fingered hands, allowing robots to perform delicate tasks. These include everyday chores like sealing a plastic bag, tying a trash bag, or even unscrewing a lightbulb. Google DeepMind noted that while there is still work to be done to increase movement speed, this update provides the necessary coordination for robots to function in more complicated environments.

Another key part of the announcement is the update to Gemini Robotics ER, or embodied reasoning. This is a vision-language model—a system that allows the robot to "see" and understand its surroundings while processing human instructions. The Verge reports that the new version, Gemini Robotics ER 2, is better at completing tasks over a long period. It now understands when a task begins and ends, which helps it stay focused on multi-step jobs. In one demonstration, the system showed how different types of robots can work together, with the Apollo 2 robot instructing a dual-arm machine to help clean a garage.

Safety remains a top priority as these machines become more capable. Google DeepMind claims that Gemini Robotics ER 2 is its safest model to date. The system is designed to detect when humans are nearby. If a person gets too close, the robot can trigger safety tools to bring itself to a complete stop. This feature is essential for robots that are intended to work in spaces shared with people.

Finally, the company has improved its Gemini Robotics On-Device Model. This version is designed to run locally on the robot itself, meaning it does not need a constant internet connection to function. According to The Verge, this model can now adapt to new robot bodies much faster than before, even if those robots have very different shapes, sensors, or movement capabilities. This flexibility is crucial for companies that want to use the same AI software across many different types of hardware.

As the industry continues to evolve, these advancements suggest that humanoid robots are becoming more practical for daily use. By combining full-body control, better hand dexterity, and improved safety, Google DeepMind is moving closer to its goal of creating machines that can handle complex, real-world work alongside humans. While the technology still faces challenges regarding speed, the progress made in coordination and reasoning is a clear sign of how quickly the field of robotics is changing.