Robots have long been good at repeating the same job over and over. But ask them to adapt when something changes, and they often struggle. Google now wants to change that. From the labs of DeepMind, the company has introduced Gemini Robotics 2, a new artificial intelligence (AI) model designed to power the next generation of humanoid robots. Think of it as a brain for humanoids that gives them the ability to reason, move and complete complex tasks in the physical world.
Unlike traditional robots that are programmed for fixed, repetitive tasks, Gemini Robotics 2 is designed to help robots understand their surroundings, reason through problems and carry out complex physical tasks on their own. Google says the model acts as an intelligence layer that allows robots to walk, think and interact more naturally in the real world.
“Think of Gemini Robotics ER 2 as a high-level brain for robots. It allows robots to chat with humans, understand the physical world and plan multi-step tasks. It then hands off motor execution to any given lower-level vision-language-action (VLA) model,” the company says in its official blog post.
Gemini Robotics 2 is teaching robots to think
Google says one of the biggest problems in robotics today is that most machines are either pre-programmed or remotely controlled. They work well in predictable environments but struggle when something unexpected happens. It is also difficult to transfer skills learned by one robot to another with a different body shape or design.
With Gemini Robotics 2, the company aims to change that. The company says the model can adapt to different types of robots and help them reason about movement instead of simply following a list of instructions. This means robots can perform a wider range of tasks in changing environments without needing to be manually programmed for every situation.
One of the biggest upgrades in Gemini Robotics 2 is whole-body control. Unlike its predecessor, which could control only a robot’s upper body, the new model can control the entire humanoid, allowing it to walk, crouch, bend, balance and use its hands at the same time. Google says it has also improved the robot’s dexterity, enabling five-fingered robotic hands to perform delicate tasks such as tying knots, sealing zip-lock bags and handling small objects, while also supporting simpler two-finger grippers for warehouse tasks like tightly packing boxes.
Gemini Robotics ER 2 acts as a project manager
Google has also introduced Gemini Robotics ER 2 to handle more complex tasks. Think of it as a project manager for robots. Instead of controlling movements directly, it understands spoken instructions, breaks tasks into smaller steps, tracks progress and adjusts the plan if something goes wrong. Google says this allows robots to complete longer, multi-step tasks involving hundreds of decisions and even collaborate with other robots in shared spaces.
Gemini Robotics On-Device 2 works without the cloud
Google has also unveiled Gemini Robotics On-Device 2, an AI model that runs directly on a robot instead of relying on the cloud. This allows robots to keep working even when they are offline. Google says developers can also customise the model for new robots in just a few hours using fewer than 200 training examples.
According to the company, together, the three models are designed to help robots move more naturally, complete complex tasks and even work with other robots.
What about safety
Google is also putting a strong focus on safety. The company says Gemini Robotics 2 combines physical safety systems with AI safety checks. It has also introduced a new benchmark called ASIMOV-Agentic to test whether robots can refuse unsafe commands, know when to ask for human help and avoid dangerous situations. Google says the robots can also detect when a person is too close, stop automatically and resume work only when it is safe.
