Google DeepMind has announced two new AI models designed to give robots better vision, language understanding, and physical control. The first, Gemini Robotics 2, is a vision-language-action (VLA) model that the lab describes as its most advanced yet. The second, Gemini Robotics ER 2, handles embodied reasoning and acts as a higher-level control system for robots.
The announcement, made on July 31, 2026, marks a significant step in DeepMind's robotics push. VLA models combine image recognition, language processing, and action control in a single system. That combination lets a robot see an object, understand a spoken instruction, and then physically act on it.
A New Intelligence Layer for Robots
DeepMind calls Gemini Robotics 2 an "intelligence layer" for adaptive robots. The model can control systems ranging from tabletop arms to full-body humanoid robots. According to the lab, it can manage full-body movement, fine motor tasks, and coordinate multiple robots at once.
That range is notable. A single model handling both delicate tasks and whole-body motion suggests a broad approach to robot control. DeepMind says the model can manage full-body movement, fine motor tasks, and coordinate multiple robots, which could simplify how developers build robotic systems.
Developers interested in testing the model can apply for early access through a waitlist. The rollout appears staged, with DeepMind controlling who gets in first.
ER 2 Takes Over Reasoning
Gemini Robotics ER 2 is the successor to Gemini Robotics ER 1.6, which was released in April. Embodied reasoning refers to a system's ability to understand the physical world and decide what actions to take. ER 2 acts as a higher-level control system for robots, sitting above the low-level action model.
The distinction matters. While Gemini Robotics 2 handles the direct motor commands, ER 2 figures out what should happen next. That division of labor mirrors how humans separate planning from execution.
ER 2 is now available in Google AI Studio, Google's platform for experimenting with AI models. That availability makes the reasoning model easier for developers to test than the action model, which remains behind a waitlist.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
What This Means for Robot Development
The two models together give developers a fuller stack for building robots. One model reasons about the world, and the other translates that reasoning into motion. DeepMind says the system can manage full-body movement, fine motor tasks, and coordinate multiple robots, which could reduce the need for custom code.
The announcement comes from Google DeepMind, the Alphabet-owned AI research lab. The lab has been developing robotics AI models for some time, and these releases continue that trajectory.
Availability and Next Steps
Gemini Robotics ER 2 is live in Google AI Studio right now. Gemini Robotics 2 requires a waitlist application. DeepMind has not said when the action model will see wider release.
The article was written by Matthias Bastian at The Decoder, an AI news outlet. The piece cites Gemini Robotics as its source, likely the official blog or product page. No independent analysis accompanied the announcement, and the article presents DeepMind's claims without critical evaluation.
The timing is notable. ER 1.6 arrived in April, and ER 2 followed within months. That rapid cadence suggests DeepMind is iterating quickly on its embodied reasoning models as competition in robotics AI heats up.
For developers, the practical takeaway is simple. The reasoning model is available now for experimentation. The action model will follow for those who get off the waitlist. Together, they represent DeepMind's current vision for how robots should think and move.