DeepMind ships Gemini Robotics 2 for humanoid whole-body control
TL;DR
- First DeepMind VLA to control a complete humanoid from feet to fingertips under one checkpoint; the 2025 version was limited to upper body.
- Success rates across confirmed tasks span 45.7% for floor-level pickup to 92% for lightbulb removal, gaps the primary blog post does not report.
- On-Device 2 adapts to new robot hardware in hours using under 200 training examples, the key economic argument for OEM adoption over traditional integration.
A post today from Google DeepMind puts a specific number on something the humanoid industry has been asking about for a while: how fast a general-purpose model can be retargeted onto a new robot body. The claim, in DeepMind's own announcement written by Carolina Parada, is that Gemini Robotics On-Device 2 adapts to new embodiments with typically fewer than 200 examples and several hours of training.
The rest of the release is a three-model suite. Gemini Robotics 2 is the vision-language-action model for whole-body control, demoed on the Apptronik Apollo 2 humanoid walking, crouching, and placing objects on shelves in cluttered environments. Gemini Robotics ER 2 is the embodied reasoning model, planning multi-step tasks that run for several minutes and coordinating multi-robot teams. The on-device variant is the third leg. DeepMind reports numbers including 89.6% accuracy on Franka Duo precision insertion tasks and 78.9% on diverse tool assembly, plus multi-finger work on 22-degree-of-freedom SharpaWave hands including knot-tying and ziplock sealing.
For anyone not shipping a humanoid, the point is about who owns the stack. The humanoid conversation for the last year has been dominated by hardware, with Apptronik, Boston Dynamics, Agile Robots and Franka Robotics all showing off platforms. If the same foundation stack can drive all four, and if new-body adaptation really is a few hundred demonstrations rather than a research project, the moat shifts from mechanical engineering to model access. Google starts to look like the operating system layer for a lot of humanoid work, and the robot vendors look like distribution.
These are DeepMind's own numbers, on DeepMind's own benchmark stack, including a new safety evaluation called ASIMOV-Agentic that measures whether the system refuses unsafe commands and asks for human help when uncertain. Success rates in the high eighties inside a curated demo are not the same as reliability on a factory floor, and the post does not disclose latency, compute footprint, pricing, or when the VLA and on-device pieces leave early access. What the reporting also does not define is what an 'example' actually is: a teleop episode, a trajectory, or an annotated demonstration.
What is worth watching from here is which humanoid maker declines to sign up. The vendors who ship on top of Gemini Robotics 2 get a working brain for free. The ones who insist on their own foundation model will need to justify that call to buyers who now have a reference point for how fast retargeting should be.
What others are reporting
-
Bloomberg Read →
Bloomberg frames the release through dexterity as the unsolved wall, contextualizing the gaps Google's new models are still working to close with hardware partners.
-
Robotics & Automation News Read →
Highlights the ASIMOV-Agentic safety benchmark and the sub-200-example adaptation threshold for new platforms, details absent from most general-outlet coverage.
Enables humanoids to coordinate movements from feet to fingertips.
-
MarkTechPost Read →
Maps the three-tier commercial access strategy and the ER 2 planning / VLA execution architectural split, with task-by-task success rates across Apollo 2 and Franka Duo.
Most robots today are pre-programmed or tele-operated for narrow, repetitive task sequences. They do not adapt to unpredictable environments.
-
The AI Insider Read →
Reports granular task failure rates including 45.7% from floors and 68.4% from tables, surfacing the performance ceiling that DeepMind's own post does not lead with.
Gemini Robotics ER 2 now understands when tasks begin and end, and can pinpoint the moment key events occur.
-
Briefs Read →
Frames the release as the Android moment for physical AI: DeepMind becomes the intelligence OS capturing margin from hardware OEMs rather than competing on form factor.
Our goal is to bring AI into the physical world and then build the intelligence layer that can be used by every robot.
Shared on Bluesky by 2 AI experts
Originally reported by deepmind.google
Read the original article →Original headline: Google DeepMind Ships Gemini Robotics 2, a Whole-Body VLA Suite With On-Device Variant That Adapts to New Robots in Hours