Big news from DeepMind! Google just launched a powerful new AI model that runs directly on robots, no internet or cloud needed. This change could make robots faster, smarter, and safer.
What’s New?
DeepMind introduced a model called Gemini Robotics On‑Device. It helps robots see, understand and act all on their own. Unlike older systems, this model doesn’t need a cloud connection to think or move.
That means robots can now work in real time, without delays or relying on outside servers.
How It Works
This AI uses vision, language, and action together. So, you can say something like “zip the bag and move it to the table,” and it listens, sees, and acts.
Even better, it can learn a new task from just 50 to 100 examples. That’s much faster than older systems.
For example, robots using this model can:
- Fold laundry
- Pour salad dressing
- Open drawers
- Pack items into boxes
All while running the model directly on the robot’s hardware. No cloud needed.
Who Can Use It?
Right now, DeepMind is testing it on different robots. These include:
- The ALOHA dual-arm robot
- Franka FR3 arms
- And the Apollo humanoid robot
Even though the robots look different, the AI still works well. That means the model is flexible, which is great for developers.
Safety Comes First
DeepMind made sure this model is safe to use. It follows strong AI safety rules and includes filters to block risky commands.
Also, because it runs locally, it can be used in places where internet is weak or not allowed. Think of hospitals, labs, or even space stations.
To make sure it stays safe, DeepMind is rolling it out slowly with trusted testers first. They also suggest teams test the model fully before real-world use.
Tools for Developers
Along with the model, DeepMind launched a full software kit (SDK). This lets builders train and test the model in both simulation and on real robots.
It works with MuJoCo, a tool for simulating how robots move. This makes it easier to test safely before putting the model on an actual machine.
Why It Matters
This isn’t DeepMind’s first step into robotics. They’ve already made models like RT-2 and Gemini Robotics-ER. But those models needed cloud help to run.
Now, robots can work without that extra layer. That makes them:
- Faster (no delays from cloud calls)
- More private (no need to send data online)
- Easier to use anywhere
As a result, this could bring robots into more everyday jobs from homes to factories and more.
Final Thoughts
With Gemini Robotics On‑Device, Google DeepMind has raised the bar for AI-powered robots. This model gives machines the power to learn and act, right on the spot.
We’re moving closer to a future where robots work beside us; fast, smart, and safe.
Learn More
Want to dive deeper? Check out these trusted sources:
