Google DeepMind Unveils Gemini Robotics 2 for Smarter Robots

Google DeepMind has announced the launch of Gemini Robotics 2, an upgraded suite of AI models aimed at enhancing robot dexterity, adaptability, and collaboration. The new platform builds on the foundation of the original Gemini Robotics models, with the goal of enabling robots to perform complex, multi-step tasks with greater precision and autonomy.

Unlike traditional robots, which are often limited to repetitive, pre-programmed actions, Gemini Robotics 2 introduces whole-body intelligence. This means robots can now execute tasks requiring coordination across their entire structure—such as walking, crouching, or manipulating objects—and adapt to new hardware configurations in just a few hours. The initiative represents a significant leap in Google’s broader strategy to integrate AI into real-world environments.

Key Features of Gemini Robotics 2

At the core of this update are three advanced models:

• Gemini Robotics 2 (VLA): A vision-language-action model capable of controlling humanoid robots and bi-arm systems, enabling fine motor tasks like tying knots or packing items tightly.
• Gemini Robotics ER 2: Focused on embodied reasoning, this model allows robots to plan multi-step tasks, coordinate with humans, and collaborate with other robots.
• Gemini Robotics On-Device 2: Designed for local operation without internet connectivity, this model adapts to new robot hardware with minimal data, enhancing real-world usability.

One of the most notable capabilities demonstrated was its ability to control humanoid robots like Apptronik’s Apollo 2. In a test, the robot successfully executed a complex instruction—placing a watering can in a specific bin on a shelf—by interpreting natural language, navigating the environment, and performing the required movements.

Why This Matters

Robotics has long faced barriers in achieving adaptability and dexterity comparable to humans. Google’s Gemini Robotics 2 aims to address these limitations, making robots more versatile for both industrial and domestic use. Enhanced capabilities like multi-robot collaboration and local processing could unlock new applications, from warehouse automation to robotic assistants in homes.

Given its rapid pace of development, Gemini Robotics has emerged as a cornerstone of Google DeepMind’s push into embodied AI. The initiative aligns with broader industry trends, as competitors like Boston Dynamics and Agile Robots also integrate advanced AI capabilities into their systems. Notably, Agile Robots announced a partnership with Google DeepMind earlier this year to implement Gemini models, signaling growing adoption within the robotics sector.

Safety and Industry Implications

With greater autonomy comes heightened safety concerns. Google is addressing this with the ASIMOV-Agentic benchmark, designed to test and improve the safety of robots performing physical tasks. Gemini Robotics ER 2, for instance, is programmed to detect human proximity and halt operations to prevent accidents, a critical feature for collaborative environments.

This release underscores Google’s ambition to lead in general-purpose physical AI. By moving beyond narrow, single-task automation, Gemini Robotics 2 positions the company to capture a larger share of the emerging robotics market.

What’s Next?

Gemini Robotics 2 is already available to early-access partners, while select models like ER 2 can be previewed via Google AI Studio and the Gemini Enterprise Agent Platform. As adoption grows, expect to see these advances integrated into a wider range of robotic platforms, driving innovation in sectors like logistics, healthcare, and consumer robotics.

Image source: Shutterstock