Google DeepMind's Gemini Robotics 2 Gives Humanoids Full-Body AI Control
Google DeepMind launches Gemini Robotics 2, bringing full-body humanoid control, multi-finger dexterity, and cross-robot collaboration to physical AI
- Three new models: Gemini Robotics 2 (full-body VLA), Gemini Robotics ER 2 (reasoning brain), and On-Device 2 (local, fast adaptation).
- Whole-body humanoid control: For the first time, a single model controls a humanoid from feet to fingertips, not just tabletop arms.
- Multi-robot collaboration: Different robot types (e.g., humanoid + arm) can now communicate and divide tasks via a shared semantic layer.
- ER 2 is publicly available via Google AI Studio and the Gemini API; VLA models require early-access sign-up.
- Fast hardware adaptation: On-Device 2 can adapt to a brand-new robot body in a few hours with fewer than 200 demonstration examples.
- Honest benchmarks: Multi-finger dexterity success rates range from 32-92%, with screwing in a lightbulb at only 36%, showing real frontier challenges remain.
For years, the most capable robot demos shared a common limitation: a single arm, a tabletop, and a narrow set of pre-programmed tasks. Google DeepMind's Gemini Robotics 2 is a direct attempt to break that ceiling. The new release introduces whole-body humanoid control, advanced multi-finger dexterity, and the ability for entirely different robot platforms to collaborate on a shared task, all powered by a single AI brain.
One brain, three models
The release is actually a family of three distinct models, each targeting a different layer of the robotics stack:
- Gemini Robotics 2: A vision-language-action (VLA) model, meaning it takes camera input and natural language instructions and outputs direct motor commands. This is the model that physically moves the robot, and it now controls full humanoids from feet to fingertips, not just tabletop arms.
- Gemini Robotics ER 2: The "embodied reasoning" (ER) model, acting as the high-level brain. It watches video feeds, plans multi-step tasks, calls external tools like Google Search, and hands off execution to the VLA model below it.
- Gemini Robotics On-Device 2: An efficient VLA optimized to run locally on robot hardware, with no network dependency, and capable of adapting to a completely new robot body in just a few hours.
Gemini Robotics ER 2 is now publicly available via the Gemini API, Google AI Studio, and in private preview on the Gemini Enterprise Agent Platform. The VLA and On-Device models are available to early-access partners.
The jump from tabletop to whole-body
Most robots are pre-programmed or teleoperated for narrow, repetitive task sequences, lacking the ability to truly learn for themselves or adapt to unpredictable environments. Moreover, transferring learned skills from one robot body to another remains incredibly difficult. Gemini Robotics 2 directly attacks this problem.
While previous models controlled the humanoid's upper body to achieve tabletop tasks, Gemini Robotics 2 expands physical AI into whole-body motions. The demo shows Apptronik's Apollo 2 humanoid receiving a single natural language prompt, then walking to a table, picking up a watering can, navigating to a shelf, and placing it in the correct bin. While the robots have more to advance in movement speed, this is an important step towards the skills needed to complete more complex, real-world tasks that require whole-body coordination.