One brain. For any robot. 🤖 We’re launching Gemini Robotics 2: our next-generation physical AI bringing full body intelligence to humanoids, advanced dexterity, multi-robot teamwork and more.

Three new models power this breakthrough: 1️⃣ Gemini Robotics 2: A vision-language-action model controlling humanoids from feet to fingertips 2️⃣ Gemini Robotics ER 2: Capable of real-world video understanding and complex, multi-step planning 3️⃣ On-Device 2: Runs locally and adapts to new robot bodies in just a few hours

Gemini Robotics 2 moves physical AI beyond tabletop tasks, enabling intelligent whole-body control for humanoids. Watch @Apptronik's Apollo 2 process a single prompt to reach, bend, and pick up a watering can ↓

To be genuinely useful in our homes and workplaces, robots need finesse. Gemini Robotics 2 equips different hardware platforms with high dexterity. It can control a five-fingered hand to tie a knot or screw in a lightbulb - while also managing parallel grippers for complex packing tasks. We’re additionally introducing multi-robot collaboration so completely different types of robots can communicate and work together to solve problems a single robot couldn’t do alone. 🤝 Find out more →

why are they always grey?

can't wait for Robot Arena on X every week

Honestly it looks quite slow compared to Unitree robot

physical data can't be scraped fast, it needs preserving once caught, not thrown away after training.

Gemini Robotics soon 😂

Let’s go

Can it chase away wild boars, though?

I still stand by this and I will be right.

DeepMind is cooking 🔥🔥

Multi-robot teamwork is the sleeper feature here. One brain running several robots at once is a real milestone. Curious how it degrades. If the link to the brain gets messy, what keeps running on the robot itself? The on-device models suggest you've already thought about this.

Gemini has finally entered a field where OpenAI and Anthropic has not touched yet. Its a good field to research and innovate. But I think robots are the ones who will make humans lazy and will be reason for unemployment in the world.

Here we are going with this

The hands look great

one brain for every robot sounds normal until the robots start sharing browser history

Elias: 'Can you tidy the lighthouse?' Robot: 'We've unionised.'

Where's 3.5 Pro?!!!!!! You should be releasing Gemini 4 Pro by now!! Why are you so slow?

not gonna lie, "one brain for any robot" sounds like sci-fi but it's just really good generalization. still cool as hell though

@jimazeyu So exciting to see the model progress! And tying a trash bag with the #SharpaWave is probably the most underrated yet most difficult and useful task out there!

“One brain for any robot” is impressive. The household final boss remains undefeated: folding a fitted sheet without accidentally inventing a new geometry.

This is cool

Great, now it can physically hallucinate dropping my coffee instead of just text.

Gemini Robotics 2 shows how quickly AI is moving into the physical world. As robots gain richer perception and more complex planning, we may soon discover that today's computing architectures not the models themselves become the limiting factor. Building better AI is only part of the challenge. Building the next generation of computing for embodied AI may be just as important. That's exactly the direction I'm exploring with C.L.M. (Crystalline Lattice Matrix).

The 'any robot' line has a number behind it: the On-Device 2 section puts adapting to a new bi-arm embodiment at a few hours and typically fewer than 200 examples. Retargeting a policy to new hardware has been a data-collection problem more than a modeling one. That's the figure I'd track release to release.

looks very cool! Thanks for sharing that.

76.3% off a shelf, 45.7% off the floor. squatting is apparently harder than everything else it does.

Multi-robot teamwork is such a flex. I always thought the hardest part would be getting them to not trip over each other! What’s the real challenge - communication lag, sensors, or something else?

People underestimate how big it really is. This is a brain they're shipping: five-fingered hand, parallel gripper, robot that has to stay standing, all off the same model. Drop it on a body it's never seen and it's working in a few hours.

oh nice!

If it can handle five-fingered dexterity like screwing in a lightbulb and tying knots, we are officially 18 months away from a humanoid doing the dishes without breaking them. The real test: Can it fold fitted sheets? 😂 Jokes aside, bringing real-time video understanding on-device to physical bodies is huge! 🦾🔥

Physical AI changes the agent-security boundary. When an agent can coordinate robots and act in real-world spaces, safety cannot live only in the model. Teams need action-level policy, scoped authority, human override, audit trails, and stop conditions before physical execution.

wow. Feels like a step change in intelligence once an LLM has a body... For doubters of AGI, robotics will make it overwhelmingly convincing

One model controlling everything from dexterous fingers to multi robot teamwork is a big step past task specific robotics 🦾