Introducing Gemini Robotics 2: Google DeepMind’s Full-Body AI Control Innovation
Google DeepMind has recently announced the launch of Gemini Robotics 2, a groundbreaking development in physical AI technology. This new system provides humanoid robots with intelligent control over their entire body, from their feet to their fingertips. This represents a significant improvement over its predecessor, which only controlled a robot’s upper body.
The Gemini Robotics 2 release includes three distinct AI models:
- The flagship Gemini Robotics 2 vision-language-action (VLA) model for whole-body humanoid control
- Gemini Robotics ER 2, an embodied reasoning model designed for multi-step planning and multi-robot collaboration
- Gemini Robotics On-Device 2, a lightweight variant that can adapt to new robot bodies with fewer than 200 training examples in just a few hours
These models allow the system to automate tasks comprising hundreds of steps and enable multiple robots to collaborate autonomously.
In testing, the system achieved an impressive 92% success rate on precision tasks such as unscrewing a light bulb, which necessitates coordinated full-body movement. Google showcased the robots performing real-world tasks, including cleaning up trash, picking up watering cans, and tying garbage bags.
Developers can access the Gemini Robotics ER 2 model via Google Cloud, the Gemini API, and Google AI Studio. However, access to the VLA and on-device models is currently limited to select partners.
Source: Google DeepMind – Official Blog