Google DeepMind released Gemini Robotics 2, a family of models for controlling a robot — in the source's words — from the feet to the fingertips.
The previous version basically handled the upper body and tabletop tasks. This one combines walking, balancing, reaching, grasping, and reasoning in a continuous sequence.
Why this matters
- Continuous sequence: understand the order, locate the object, walk to it, pick it up, cross the room, and place it in the right spot.
- There are three models with distinct roles, not just one.
- One of them runs right on the robot, completely offline.
- You can try it out in AI Studio.
1.The three models and what each one is for
3 min- Gemini Robotics 2 — the vision-language-action model. Converts your natural language instruction, plus what the robot's cameras see, into motor commands.
- Gemini Robotics ER2 — the high-level reasoning. Understands the environment, plans tasks, corrects failures, and coordinates multiple robots.
- Gemini Robotics On-Device 2 — smaller version that runs on the robot itself, without an internet connection.
The split mirrors how a person operates: one level decides what to do and replans when things go wrong, another executes the movement. The local model solves the practical problem of connectivity — a robot that depends on the internet to move freezes when the network drops.
Spec sheet
- Access
- AI Studio + programa de testadores
- Models
- 3 (VLA, ER2, On-Device 2)
- Scope
- Corpo inteiro
Continue in the full microcourse
You've read the opening of 3 classes
The microcourse covers the complete step-by-step, the selection criteria, where the tool fails, who it's really for — and, in the Expert version, the official address to start today.
- What changed in the hands2 min
- How to get your hands dirty1 min
How we verified
We track releases straight from primary sources, transcribe what's demonstrated, check every name and number against the manufacturer's official documentation, and rewrite it in Portuguese — with what the tool no do it together, which is the part the ad leaves out.
