Google Gave Robots a Gemini Brain — and Now Humanoids and Rovers Can Work as a Team

Google DeepMind's new Gemini Robotics 2 and Gemini Robotics ER 2 give robots a high-level 'embodied reasoning' brain — letting humanoids, quadrupeds and rovers plan multi-step tasks and, for the first time, coordinate with each other as a team. And developers can use it today.

By Ubedulla · 4 min read
A humanoid robot, a four-legged robot, and a wheeled rover connected by glowing blue neural lines, illustrating Google DeepMind's Gemini Robotics 2 coordinating multiple robots as a team.
Google DeepMind's Gemini Robotics 2 / ER 2 lets different robots reason, plan, and collaborate. Illustration.

Google just gave robots a shared brain. On July 30, Google DeepMind unveiled the next generation of its robotics AI — Gemini Robotics 2 and its reasoning counterpart, Gemini Robotics ER 2 — and the headline capability is one that sounds like science fiction: different robots, from four-legged machines to full-size humanoids, can now reason about the world, plan multi-step tasks, and coordinate with each other as a team.

It's one of the clearest signs yet that the "AI agent" wave isn't staying inside your chatbot. It's climbing into physical bodies. Here's what Google actually announced and why it matters.

A brain and a body

Google's system splits into two parts that work together. Gemini Robotics ER 2 ("ER" for embodied reasoning) is described as the high-level brain for a robot — the part that, in Google's words, "allows robots to chat with humans, understand the physical world, and plan multi-step tasks." Think of it as the planner and decision-maker. The companion model, Gemini Robotics 2, brings "whole-body intelligence" — the lower-level control that actually moves the machine, including humanoids that can walk and use their whole body, not just an arm bolted to a table.

The clever design is that the brain doesn't have to do the muscle work itself. Gemini Robotics ER 2 "hands off motor execution" to a lower-level vision-language-action model — so the reasoning model figures out what to do, and a specialized control model handles how to move. It can even call external tools like Google Search mid-task to look something up.

The new trick: robots that work as a team

The standout upgrade is multi-robot collaboration. Gemini Robotics ER 2 lets a mix of different robots — wheeled rovers, quadrupeds, humanoids — communicate and complete a complex workflow together, dividing up a job instead of each working in isolation. In Google's demos, that included a Boston Dynamics Spot robot fetching objects from natural-language commands, and an Apptronik Apollo 2 humanoid teaming up with a Franka robotic arm on a shared task.

There's also a new sense of self-awareness about progress: the model watches continuous video of a task to judge whether it's actually being done correctly, scoring 91.3% accuracy on "moment-finding" (spotting when a key step happens) and 57.4% on tracking overall task progress. In plain terms, the robot can now notice when it's off track — a big deal for machines that need to operate without a human babysitting every move.

Why this is a bigger deal than another chatbot update

For two years, the AI story has been about software that talks. Gemini Robotics 2 is about software that acts in the real world — and, crucially, it's model-agnostic and body-agnostic. The same reasoning brain can sit on top of a warehouse rover, a dog-shaped inspection robot, or a humanoid, because Google separated the thinking from the moving. That's the pattern that could let capable robots spread far faster than if every robot needed its own bespoke AI.

It also lands in a loaded moment. Humanoid robots are already a geopolitical flashpoint — the US just moved to block Chinese-made humanoids on national-security grounds — and now the biggest Western AI lab is handing developers a general-purpose brain to run fleets of them. The technology to put coordinated, reasoning robots into workplaces is arriving well ahead of the rules for it.

Can developers use it?

Yes — this isn't just a lab demo. Google made Gemini Robotics ER 2 available through the Gemini API and Google AI Studio, with a private preview on its Gemini Enterprise Agent Platform and code samples on GitHub. The work was led by DeepMind engineers Steven Hansen and Peng Xu. That developer availability is what turns a flashy announcement into something that shows up in real robots over the coming months.

Frequently asked questions

What is Gemini Robotics 2?

It's Google DeepMind's next-generation robotics AI, announced July 30, 2026. It pairs "whole-body intelligence" (lower-level control that moves robots, including humanoids) with Gemini Robotics ER 2, a high-level "embodied reasoning" brain that plans tasks and coordinates robots.

What does "embodied reasoning" mean?

It's the robot's high-level thinking layer. Gemini Robotics ER 2 lets a robot understand its physical surroundings, hold a conversation, break a goal into multiple steps, and decide what to do — then hand the actual movement off to a specialized control model.

What's the biggest new capability?

Multi-robot collaboration: different types of robots (rovers, quadrupeds, humanoids) can now communicate and complete a complex task together as a team, rather than each operating alone. The model can also track its own progress by watching video of the task.

Which robots does it work with?

Google demonstrated it with a Boston Dynamics Spot quadruped, an Apptronik Apollo 2 humanoid, and a Franka robotic arm. Because the reasoning brain is separated from the movement control, it's designed to work across many different robot bodies.

Can I actually build with it?

Yes. Gemini Robotics ER 2 is available via the Gemini API and Google AI Studio, with a private preview on the Gemini Enterprise Agent Platform and code samples on GitHub.

Reporting based on Google DeepMind's official announcements of Gemini Robotics 2 and Gemini Robotics ER 2, and coverage by WIRED, Bloomberg, Unite.AI and others. The Bot Post will update this story as more demos and developer details emerge.

About the author

Ubedulla

Founder & Editor

Founder and editor of The Bot Post, covering AI news and technology.

Related Articles