article — General

Google Just Unveiled Its Most Advanced AI Robots Yet, Gemini Robotics 2

About this Resource

Explore Google's Gemini Robotics 2, an advanced AI system designed to help robots reason, move, manipulate objects and collaborate in the real world.<h2> </h2><h2>Google Just Unveiled Its Most Advanced AI Robots Yet, Gemini Robotics 2</h2>Artificial intelligence is beginning to move beyond computer screens and into the physical world. Google’s Gemini Robotics 2 is designed to give robots something closer to a general-purpose “brain,” allowing them to understand instructions, reason about their surroundings, control their bodies, and perform complex physical tasks. Unlike robots built to repeat one carefully programmed movement, Gemini Robotics 2 focuses on generality. The goal is for one robot to handle many different situations and tasks rather than being limited to a single job. <h2>Three Major Improvements</h2>Google highlights three important areas in Gemini Robotics 2, whole-body control, dexterous manipulation, and robot collaboration. <ul><li>Controlling the Entire Robot - Simple movements for humans can be extremely complicated for robots. Walking, reaching for something, or maintaining balance requires many different joints and actuators to work together at precisely the right time. Gemini Robotics 2 is designed to coordinate these movements across the robot's entire body. It can understand what needs to happen and translate that decision into physical movement while maintaining balance.</li></ul> <ul><li>More Precise Hands -The system also focuses heavily on robotic dexterity, the ability to manipulate objects accurately. Demonstrations include putting grapes into a Ziploc bag, closing the bag, unscrewing a light bulb, moving tools into the correct positions, and tying a knot in a trash bag. These tasks require much more than simply picking something up and putting it somewhere else. For example, unscrewing a light bulb requires the robot to understand several different movements, including twisting, pushing and pulling while keeping its fingers precisely positioned. This could eventually make robots more useful for jobs involving complicated or potentially dangerous objects, including hazardous waste.</li></ul> <ul><li>Robots Can Work Together - One of the most interesting features is the ability for multiple robots to collaborate on the same task. In one demonstration, an Apollo humanoid robot works with a Duo robot to organize tools. Apollo handles part of the job before handing control to Duo to complete the remaining steps. Each robot runs its own copy of the system, allowing them to make their own decisions while coordinating with one another. The robots can communicate with each other and determine when another robot should assist. This could allow teams of robots to tackle tasks that would be difficult for a single machine.</li></ul> <h2>Gemini Robotics 2 Can React to Changes</h2>A major part of making robots useful in the real world is dealing with situations that weren't perfectly planned in advance. During the demonstrations, the robot navigates cluttered environments, identifies the correct objects, places them where they belong, and can recognize when something goes wrong before trying again. Another demonstration challenges the robot to find a tote bag that has been unexpectedly placed on the floor and move it onto a table. The goal is to test how quickly the robot can react to changes in its environment. <h2>How the AI Controls the Robot</h2>At a simple level, Gemini Robotics 2 works by connecting understanding and reasoning with physical action. The reasoning system can interpret what the robot sees and understand natural-language instructions. It then works with a vision-language-action system that turns those decisions into movements for the robot. That means an instruction such as asking a robot to pack sports equipment isn't just a command to move one object. The robot has to understand the environment, identify the correct items, plan the task, move through the space, maintain its balance, and coordinate its joints and hands. <h2>Why This Matters</h2>The bigger idea behind Gemini Robotics 2 is general-purpose physical AI. Today's robots can be extremely impressive when they are designed for specific tasks. The challenge is creating machines that can operate in the unpredictable environments humans deal with every day. Gemini Robotics 2 is Google's attempt to address that challenge by giving robots better reasoning, whole-body control, dexterity, adaptability, and the ability to collaborate. The technology is still being developed, but the demonstrations show a direction where AI is no longer limited to generating text, images, or code. AI is increasingly being designed to understand the physical world and act within it.