Google DeepMind launches Gemini Robotics 2 for humanoid robots

Google DeepMind on Thursday unveiled Gemini Robotics 2, a next-generation AI system that extends robot control from the upper body to full whole-body movement, enabling humanoid machines to walk, crouch, manipulate objects, and collaborate with one another.

From Arms to Full-Body Control

The release marks a substantial expansion over the first-generation model introduced in March, which primarily governed a robot's arms and hands. Gemini Robotics 2 can now direct an entire humanoid from head to toe, according to Google's Alphabet Inc. ↗ 6.88% official blog post. The system translates natural language instructions into coordinated physical actions — for instance, asking Apptronik's Apollo 2 humanoid to place a watering can on a bottom shelf prompts the robot to walk to the table, pick up the object, and deliver it to the correct location.deepmind

The update comprises three models: Gemini Robotics 2, a vision-language-action model governing full-body motor control; Gemini Robotics ER 2, an embodied reasoning model that handles high-level task planning and multi-robot coordination; and Gemini Robotics On-Device 2, an efficient model optimized to run locally without internet connectivity. The on-device model can adapt to entirely new robot bodies with just a few hours of data and fewer than 200 examples.blog+1

Dexterity and Collaboration

The system also advances fine manipulation. Controlling a five-fingered, 22-degree-of-freedom hand on the Apollo 2 robot, the model can perform delicate tasks such as tying knots and sealing ziplock bags. Google DeepMind acknowledged, however, that precision and speed have "not yet reached human-level hand technology."mk+1

Multi-robot collaboration is another new capability. Different types of robots can communicate through shared semantic understanding to hand off and complete workflows that no single machine could manage alone. "It's another milestone in our journey toward what we refer to as physical AGI, which means enabling a robot to perform any task a human can," Carolina Parada, head of robotics at Google DeepMind, told WIRED.wired+1

Safety and Industry Context

Google is also introducing ASIMOV-Agentic, a benchmark for evaluating the safety of layered AI systems controlling robots, measuring whether commands could lead to harmful outcomes and whether robots can detect nearby humans and halt accordingly. The launch comes as the FCC this week moved to ban future sales of Chinese-made robots over security concerns, potentially limiting the hardware available to run such software in the United States.axios+2