HomeAIGoogle DeepMind Unveils Gemini Robotic
AI

Google DeepMind Unveils Gemini Robotics 2 Models

Google DeepMind has released Gemini Robotics 2, adding whole-body humanoid control, fine finger dexterity, and multi-robot collaboration capabilities.

WHAT YOU NEED TO KNOW
  • Gemini Robotics 2 converts vision and language into motor control for full humanoid bodies and bi-arm platforms.
  • The ER 2 reasoning model handles multi-step planning over several minutes and enables multi-robot collaboration.
  • Gemini Robotics On-Device 2 adapts to new robot embodiments in a few hours using under 200 examples.
  • Safety features include human proximity detection for automatic stops and the ASIMOV-Agentic evaluation benchmark.

Google DeepMind announced Gemini Robotics 2, a suite of artificial intelligence models designed to give robots whole-body control, fine physical dexterity, and multi-robot collaboration capabilities. The release includes three distinct models aimed at different aspects of robotic reasoning and motor control.

The core vision-language-action model, Gemini Robotics 2, translates visual and natural language inputs directly into physical commands. Google DeepMind said the system can control entire humanoid bodies from feet to fingertips, as well as bi-arm configurations, allowing machines to perform coordinated movements such as walking, crouching, and reaching.

In practical tests with Apptronik’s Apollo 2 humanoid robot, the model processed instructions to pick up a watering can and place it inside a specific bin on a low shelf. The same robot used a five-fingered, 22-degree-of-freedom SharpaWave hand to execute delicate actions like tying knots and sealing ziplock bags. On a Franka Duo platform, standard two-fingered parallel grippers performed tight packing tasks.

Reasoning and local execution

To manage complex sequences, Google DeepMind introduced Gemini Robotics ER 2, an embodied reasoning vision-language model. This high-level agent communicates with human operators, plans multi-step tasks lasting several minutes, and coordinates with motor control models. The ER 2 system also introduces multi-robot collaboration, allowing different types of robots to communicate and solve complex workflows together.

For hardware running without internet connectivity, Google DeepMind released Gemini Robotics On-Device 2. The company said this efficient model adapts to completely new robotic shapes, sensors, and degrees of freedom within a few hours using fewer than 200 training examples. Demonstration tasks were run across the Dexmate, SO101, and Trossen platforms.

Safety and availability

Safety evaluations for the new models rely on a benchmark called ASIMOV-Agentic, which measures an agent's ability to refuse unsafe commands and request human intervention when uncertain. Google DeepMind stated that Gemini Robotics ER 2 can detect nearby humans and bring the robot to a safe stop if someone approaches too closely.

Google DeepMind has made Gemini Robotics ER 2 available on Google AI Studio and in private preview through the Gemini Enterprise Agent Platform. The vision-language-action and on-device models are accessible to early-access partners, with development supported by robotics partners including Apptronik, Boston Dynamics, and Agile Robots.

Xentir Media
Xentir Media NewsroomSource-backed AI and technology coverage, drafted by Xentir's automated editorial system under fixed human-set rules. See our editorial policy and AI usage policy.
J
Jomon · Founder & EditorFounder and editor of Xentir Media. Sets the editorial rules the newsroom system runs under, and is accountable for its corrections. About Jomon · [email protected]
The Xentir Brief
The developments worth knowing — one useful email.
Get the Brief →