Google Reveals Gemini Robotics 2.0, Promising Improved Dexterity and Safety
The new model can control humanoids from feet to fingertips, coordinate multiple robots and adapt to a new body in a few hours.
- On Thursday, Google DeepMind unveiled Gemini Robotics 2, enabling "whole-body motions" that allow humanoid robots to walk, balance, and manipulate objects from feet to fingertips.
- The release comprises three distinct models: a vision-language action model, Gemini Robotics ER 2 reasoning layer, and On-Device 2, which enable dexterous tasks like sealing a Ziploc or tying trash bags.
- Emphasizing safety as robots enter human spaces, the company introduced ASIMOV-Agentic benchmark—named after Isaac Asimov—to detect harmful commands and evaluate whether robots refuse unsafe tasks.
- Despite these advancements, Google isn't releasing consumer-facing robots anytime soon, as DeepMind acknowledged the system remains slow and true dexterity remains a "distant goal."
- As Google competes with Tesla to achieve "physical AGI," it faces political hurdles regarding Chinese-made hardware supply, complicating the goal of versatile one-model robot integration.
45 Articles
45 Articles
Robots get better at handling unpredictability
Google DeepMind unveiled a new model this week that gives humanoid robots finer control over how they interact with objects around them. Promotional videos show a robot doing complicated tasks like tying a trash bag and screwing a small lightbulb, suggesting we’re a step closer to general-purpose robotics — where humanoids can accomplish more than repetitive, pre-programmed tasks.The advancements point to the West’s lead over China in robotics s…
Gemini Robotics 2 brings us one step closer to physical AGI
This week, Google DeepMind revealed Gemini Robotics 2, an intelligence layer comprising three new models to power more adaptable physical AI. Together, Google says these models will give robots more dexterous, full-body control to work together and complete a wide range of multi-step tasks. Google announced the vision-language-action (VLA) model, Gemini Robotics 2, on Thursday, and converts vision and language inputs into motor control so robot…
Google unveils Gemini Robotics 2, a brain for humanoids that lets them walk, think and act
Imagine a robot that can understand your instructions, walk across a room, pick up objects and even pack boxes. That's exactly what Google hopes to achieve with Gemini Robotics 2, its latest AI model built for smarter humanoid robots.
Google has unveiled an artificial intelligence (AI) model for driving next-generation humanoid robots. Google announced on the 31st that it has revealed "Gemini Robotics 2," an AI model that enhances the robot's full-body control and precision manipulation capabilities. Unlike its predecessor, "Gemini Robotics 1.5," which was capable of controlling only the upper body, Gemini Robotics 2 can integrally control the entire body, including fingers a…
Coverage Details
Bias Distribution
- 46% of the sources lean Left
Factuality
To view factuality data please Upgrade to Premium
