Gemini Robotics
A vision-language-action model from Google DeepMind built for robots
Gemini Robotics is a vision-language-action model made by Google DeepMind and built on the Gemini 2.0 large language model. Launched on 12 March 2025 alongside an embodied reasoning variant, it handles unfamiliar situations and later gained an on-device version and a second-generation reasoning model.
- Built on the Gemini 2.0 large language model
- Gemini Robotics and Gemini Robotics-ER launched 12 March 2025
- On-Device variant released 24 June 2025 for local robot hardware
- Gemini Robotics ER 2 announced 30 July 2026 with multi-robot support
- Early testers included Boston Dynamics, Agility Robotics and Agile Robots
What it is
Gemini Robotics is a vision-language-action model from Google DeepMind. That means it takes in images and language and produces actions a robot can carry out. It is built on top of the Gemini 2.0 large language model and adapted for robotics work, with an emphasis on handling situations it has not met before. A companion model, Gemini Robotics-ER, focuses on embodied reasoning: thinking about objects, space and tasks in the physical world. Both were announced on 12 March 2025. Later additions include a version that runs on the robot itself and a second-generation reasoning model.
How it came about
Google DeepMind launched Gemini Robotics and Gemini Robotics-ER together on 12 March 2025. On 24 June 2025 the lab followed up with Gemini Robotics On-Device, tuned to run locally on robotic hardware rather than in the cloud. On 30 July 2026 came Gemini Robotics ER 2, an upgrade to the embodied reasoning line. Early access to the first models was limited to a set of trusted testers.
Inside the model
The model extends the Gemini 2.0 large language model into a vision-language-action design, so it pairs visual and language understanding with outputs that drive a robot. Gemini Robotics-ER adds embodied reasoning, covering how a system reasons about physical scenes and tasks. The On-Device variant was reworked to run on the robot's own hardware. Gemini Robotics ER 2 adds real-time video understanding, tracking of task progress, tool use and coordination between several robots, with lower-latency orchestration through the Gemini Live API.
What it can do
Gemini Robotics is described as able to understand new situations, a key requirement for robots working outside fixed factory routines. The ER 2 release brings real-time video understanding, task progress tracking, tool integration and multi-robot collaboration. No benchmark scores were reported here.
How to use it
The first Gemini Robotics models were given only to trusted testers. Gemini Robotics ER 2 was made public to developers through the Gemini API and Google AI Studio. The On-Device variant targets local execution on robotic devices.
Why it matters
The line shows Google DeepMind pushing its Gemini models into physical systems, pairing a large language model with vision and action for robots. Its trusted testers included robotics firms such as Boston Dynamics and Agility Robotics, and the later public API release opened the embodied reasoning model to a wider developer audience.
Questions people ask about Gemini Robotics
What is Gemini Robotics?
It is a vision-language-action model from Google DeepMind, built on the Gemini 2.0 large language model. It reads images and language and produces actions for robots, and is meant to cope with situations it has not seen before.
What does the ER version do?
Gemini Robotics-ER stands for embodied reasoning. It focuses on reasoning about physical scenes and tasks. The second generation, Gemini Robotics ER 2, adds real-time video understanding, task progress tracking, tool use and multi-robot collaboration.
Can I use these models?
The first models went only to trusted testers such as Boston Dynamics, Agility Robotics, Agile Robots and Enchanted Tools. Gemini Robotics ER 2 was later released to developers through the Gemini API and Google AI Studio.
Is there a version that runs on the robot?
Yes. Gemini Robotics On-Device, released on 24 June 2025, is a variant designed and optimised to run locally on robotic devices instead of relying on cloud processing.
Good reading and sources
- Official websiteOfficial site
- Gemini Robotics 2.5 — Google DeepMindGoogle DeepMind, Article
- Google’s Gemini Robotics AI Model Reaches Into the Physical WorldWired, Article
- Google DeepMind’s new AI models help robots perform physical tasks, even without trainingThe Verge, Article
The facts on this page come from the Wikipedia article "Gemini Robotics", written by its contributors and used under the CC BY-SA 4.0 licence, and from its Wikidata record (CC0). EveryLLM rewrote the text in its own words, with AI assistance, and compares every page with the article to make sure no passage is copied. It may be reused under the same CC BY-SA 4.0 licence with a link back.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.